File Course Module in Biostatistics

Biostatistics is coined from two words, bios, which means life, and statistics, which means the
collection, organization, and management of data. These two words allow us to define biostatistics as
the collection, organization and the management of biological data. Biostatistics is a widely taught
course as it is not just limited in the study of biology. It is also used in public health and in medicine.
Biostatistics encompasses the two branches of statistics: descriptive and inferential statistics.
Descriptive statistics applies to the summarization and presentation of data in a form that will make
them easier to understand for the reader. It involves methods like tabulation, graphical presentation,
and computation of averages to name a few. Inferential statistics on the other hand, applies to the
making of generalizations and conclusions about a target population based on the results from a
sample of a population. It involves the estimation of parameters and the testing of hypothesis to name
a few.
Biostatistics is essential because it enables each one of us to deal with the phenomenon of
variation. What is the phenomenon of variation? The phenomenon of variation refers to the tendency
of a measurable characteristic to change from one individual or one setting to another or from one
instant of time to another instant within the same individual or setting. One example of the
phenomenon of variation is when you want to take your own blood pressure reading. At one time, you
may get your blood pressure reading and on a different time, the blood pressure reading that you
obtained at that time may not be totally similar to your first blood pressure reading. The use of
biostatistics to deal with the phenomenon of variation provides a systematic way of describing and
analyzing the variability of the different phenomena that we may encounter.
Biostatistics provides us with numerous applications in the study of biology, public health, and
medicine. Biostatistics is commonly used as a tool in the decision-making process as it allows us to
properly identify the problem, assess the needs, allocate the limited resources that we have, and even
evaluate programs that necessitate a way to have a systematic process of collecting data.
Biostatistics makes use of two kinds of data, namely qualitative data and quantitative data.
Both kinds of data can either be a constant or a variable in the study of biostatistics. Let us distinguish
between and constant and a variable. A constant is a phenomenon whose values remain the same
either from person to person, or time to time, or place to place. Common examples of constants are
the number of grams in a kilogram, the speed of light in a vacuum, the number of minutes in an hour,
and the number of days in a week. These examples give us values that more or less remain the same.
On the other hand, a variable is a phenomenon whose values or categories cannot be predicted with
certainty. Common examples of variables include the color of a person’s hair, the number of children in
a family, attitudes towards certain issues, the weight of a person, and educational attainment. Even the
certainty of whether a person is a smoker or a non-smoker is something that cannot be predicted
easily.
There are two types of variables used in the study of biostatistics: quantitative variables and
qualitative variables.
Quantitative variables are those variables that can be measured and ordered according to their
quantity or amount or whose values can be expressed numerically. Examples of quantitative variables
include birth weight, head circumference, and population size.
A quantitative variable can either be discrete or continuous. Discrete quantitative variables are
those that have numerical values that are integers or whole numbers. They are characterized by gaps
or interruptions in the values that it can assume. A few examples of which are hospital bed capacity
and household size. We say that in a particular tertiary hospital, they have a hospital bed capacity of
30 beds. The 30 bed-capacity of the hospital can only serve 30 individuals. Hence, the quantity 30 is
indicated as an integer or a whole number. We cannot say that the hospital has a bed capacity of 30.5
since half a bed does not exist. The next value that this variable can take on after 30 is 31. On the other
hand, continuous variables have values that are potentially associated with real numbers. This means
that they can attain any value including fractions or decimals. They do not possess the gaps or
interruptions characteristic of a discrete random variable. Some examples include weight and height
where we can indicate that a person weighs 14.5 kilos or that a person has a height of 176.25 cm.
Qualitative variables are those variables that are used as labels to distinguish one from the
other. They have values that are intrinsically nonnumerical (categorical). Examples of qualitative
variables include sex, urban/rural classification, and the regions of the Philippines. A person can either
be a male or a female when categorized according to sex. In urban/rural classification, a person can
inhabit an urban area or a rural area, while when classifying according to the region of the country, you
have regions 1 up to the ARMM region.
In biostatistics, data collection is a major activity as the data that we collect in biostatistics is
the same data that we organize and manage. There are two categories of data according to source:
primary data and secondary data.
Primary data is obtained first hand whereas secondary data is obtained from existing data. One
important feature why researchers want to collect primary data is because it can specifically answer the
purpose of the investigator whereas in secondary data, the purpose of the original author may not
necessarily be the same as those of the investigator.
There are a number of sources where we can obtain the data that we need. These sources can
be from literature, expert’s judgment, census, interview, direct measurement, actual observations, etc.
The primary and secondary data obtained from these different sources have their advantages and
disadvantages. The table on the next page shows some examples of methods that enable us to collect
data together with their advantages and disadvantages.
Method of Data Collection Advantage Disadvantage
Documented sources Pre-collected, savings in May not answer specific
time, money and energy questions of investigator
Making observations Answer objectives Expensive
Interview Subjects or variables not Expensive, time consuming
amenable to observations
like opinions or feelings
Questionnaires Less expensive and time Lower yield of respondents
consuming
The choice of method for data acquisition depend on the level of qualitative and quantitative
data the researcher desires to collect, as well as time, money and manpower available for the data
collection.
The data that we collect in the field also has certain qualities. The qualities of good data, or
statistical data for that matter, are timeliness, completeness, accuracy, precision, relevance, and
adequacy.
1. Timeliness – the interval between the date of occurrence of the different events considered
and the time the data is ready to be used or disseminated by the researcher.
2. Completeness – the data is of coverage and can accomplish all the items necessarily
needed by the researcher.
3. Accuracy – the data collected is close to the measurement or to its true value.
Ways to check accuracy of data:
a. Compare the collected data with expected trends
b. Compare the data with totals of other record systems
c. Compare the levels of collected data with those of other places with comparable
conditions
4. Precision – the extent of the data collected refers to the consistency of information.
5. Relevance – the data collected meets the objectives of the data users. This is attained
when it answers the objectives of the users and enhanced when there is communication
between the data producers and the data users.
Adequacy – the collected data provide all the basic information needed to meet the
requirements of the user.
NAME:__________________________________________________________ DATE:______________
SECTION:__________________________________ INSTRUCTOR:____________________________
1. In the following situations/objectives, identify the possible problems that you may encounter and
indicate the data that you need to answer the objectives of your study. Indicate your answers on the
provided table. (12 points)
Type of Data Method of

Situation/Objectives Possible Problems Data Required
needed Collecting Data
a. A student is
interested in
determining the
prevalence of anemia
using pallor of
conjunctiva as the
indicator.
b. A doctor is
interested in
obtaining weekly
reports on the
number of cases of
notifiable diseases in
the neighborhood.
c. A researcher wants
to determine the
mean birth weight of
Filipino infants based
on the data indicated
from the infant’s birth
certificates.
2. As a research assistant, you were assigned to collect and report the quarterly data on the
ectoparasites in freely roaming animals in the community. Explain how your reports can have
problems with respect to: timeliness, accuracy, precision, completeness, relevance and adequacy.
(7 points)
3. What is the best method for collecting data on each of the following and why? (6 points)
Reason for choosing the best

Situation Best method for collecting data
method for collecting data
Prevalence of drug and

substance abuse among the
U.P. students.
Effect of iron
supplementation on the
hemoglobin levels of
pregnant women in Barangay
X.
Average heavy metal

concentrations on the plant
tissues in a garden Y.
Biologists are in constant pursuit of new knowledge through observations, explorations and
analyses conducted in the laboratory and in the field. These acts of observations, explorations and
analyses constitute the process called research. The biological research process is described as
systematic, objective and reproducible.
It is systematic because it is akin to a problem-solving technique that follows the scientific

method of inquiry. The objectivity of biological research is based on its empiricism, whereby
conclusions are founded on observed facts and logical reasoning. Statistics is not only limited to the
summary, presentation and analysis of data but also in the design and execution of research to avoid
haphazard data gathering and faulty data analysis.
Biological research carried out by a biology student, whether in the form of a special research
problem or undergraduate thesis, must be reproducible such that it can be validated by peers. Good
biological research must be disseminated in scientific gatherings so that it can serve as a stimulus for
further study and contribute to the greater body of knowledge.
1. Identification of the research problem

2. Formulation of research objective/s
3. Review of related literature (RRL)
4. Formulation of testable hypotheses
5. Construction of the research design
6. Drawing inferences or conclusions
7. Dissemination and/or utilization of results
Biologists investigate research problems related to their specific expertise or field of

knowledge. As biology students, a research topic may be pursued on the basis of factors such as field
of interest, feasibility of investigating the research problem (e.g. limitations in time, availability of
research funds), relevance (i.e. timeliness and impact), and ethical considerations (i.e. moral
considerations in the use of human or animal subjects). The student researcher must avoid trivial
research problems and focus on ethically sound scientific research. In student research proposals, the
research problem is written succinctly, either in the form of a question or a statement.
The research objectives must reflect the questions that need to be addressed in the biological
research process. It can be formulated as a single statement or in the form of general and specific
objectives. When writing the research objectives, the marketing mnemonic known as “S.M.A.R.T.” may
be used as a guide:
 Specific: the statements are clear, unambiguous, and straight-to-the-point
 Measurable: success or achievement of the objective can be gauged or established
 Attainable: the objectives are realistic, practical and achievable based on limitations
 Relevant: the objectives are related to the research problem
 Time-bound: there is a definite end in relation to its measurability
When considering a research problem to formulate appropriate objectives, students can seek
the opinions of scientists and academicians who have research experience in specific areas of biology
for guidance. Coupled with personal observations and a review of similar studies that have been done
on the topic of interest, students will have a clear direction toward a valid research problem that is
relevant and worth pursuing.
Ideally, in student research proposals a section on “scope and limitations” is included so that
assumptions, restrictions and limitation are explicitly stated with respect to the coverage of the study.
Matters such as time allotted for the conduct of the study, cooperation needed from collaborating
institutions or scientists, ethical considerations (e.g. method of handling live vertebrate specimens),
and available funds and facilities must be specified.
A comprehensive review of related literature (RRL) involves the collation and integration of
available information which are related to the research problem of interest. The RRL incorporates basic
principles and existing studies on the topic and must not be written as a litany of previous work, but as
a meaningful critique of current evidence. Attention must be given to the scientists who have
contributed most to the body of knowledge, the research designs and appropriate statistical analyses
executed in published research and the conclusions of these studies. The goal of the RRL is to establish
a rationale or a conceptual framework for pursuing the research problem.
There is much confusion in the declaration of the research hypothesis. One school of thought
considers the research hypothesis as the researcher’s perceived answer to the research problem. In
most biological research involving the use of statistical analysis, the research hypothesis is an assertion
about the relationship between two or more variables which are to be observed in the research
problem. The hypotheses are declared as a pair of statements known as the null and alternative
hypotheses.
1. The Null Hypothesis

 Represented symbolically as H0 or Ho
 The hypothesis statement of equality or no difference
 Example: On general weighted averages (GWA) of 2nd year biology students
o Ho: The GWA of 2nd year males is equal to the GWA of 2nd year females
o Ho: ̅ ̅
o Ho: (Male GWA) – (Female GWA) = 0
Statistics and the Biological Research Process
2. The Alternative Hypothesis

 Represented symbolically as H1 or Ha
 The hypothesis statement of non-equality or difference
 Possible alternative hypothesis counterparts for the example above:
o Ha: The GWA of 2nd year males is not equal to the GWA of 2nd year
females
o Ha: ̅ ̅
o Ha: (Male GWA) – (Female GWA) 0
o Ha: GWA of males > GWA of females or
o Ha: GWA of males < GWA of females
The actual research hypothesis of the investigator may either be the null or alternative
depending on the nature of the study. The statistical test to be used will be the basis for the rejection
of one of the two hypotheses and the “non-rejection” of the other. By convention, a hypothesis may
only be rejected or not rejected and never “accepted” since this poses the danger of making hasty
generalizations and declaring blanket statements.
The research design is a careful scheme for data collection and analysis that may be considered
as the “plan of attack” in order to answer the research objectives. Bulk of this plan is stipulated in the
“methodology” section of student research proposals. In designing a study, the researcher must focus
on the research objectives, the variables of interest, facilities and equipment required and the time
frame involved. The first step is to clearly define the variables relevant to the research problem.
During high school, science investigatory projects often require the identification of the research
variables: the independent (control) variable, dependent (outcome) variable, and possible extraneous
(confounding) variables. In statistics, the independent and dependent variables must be well-defined
and the extraneous variables controlled or minimized.
 Constant versus Variable

o Constant – a phenomenon in which its value remains the same regardless of time,
place or individual (e.g. speed of light, π, gravitational constant); a physical
phenomenon
o Variable – a phenomenon that can take different values depending on current
circumstances (e.g. weight, height, exam scores); a biological phenomenon
 General Types of Variables

o Qualitative – a categorical variable
 Categories of the variable are considered as labels to distinguish one group
from another
 Categories cannot be used as basis for saying that one group has a higher
value than another group
 e.g. name, sex, nationality, religion
 When using statistical software, a categorical variable can be represented
numerically for simplification or coding purposes (e.g. 0-male; 1-female)
o Quantitative – a variable that can be measured numerically using a specific unit of
measurement
 Values of the variable specify quantity or amount and may be arranged
according to magnitude
 Quantitative variables can be further classified as discrete or continuous
1. Discrete: only counts or whole numbers are meaningful (e.g. number of
offspring)
2. Continuous: fractions and decimals are meaningful values (e.g. tree
height)
When dealing with variable data, it is imperative to classify the collected data according to the
level of measurement. This allows the researcher to determine the appropriate statistical methods to
summarize and analyze the data. The levels of measurement (or scales of measurement) include
nominal, ordinal, interval and ratio.
 Nominal Level Data

o Level of measurement where numbers or names represent a set of mutually exclusive
classes to which observations of a variable may be assigned
o Variables that yield nominal-level data are all qualitative
o Examples of nominal data for qualitative variables:
 Sex: male & female
 Disease status: sick, not sick
 Student number: 1998-12345, 2009-54321
 Ordinal Level Data

o Similar to nominal in terms of being categorical in nature
o Unique feature is that the mutually exclusive classes can be ordered or ranked (e.g. 1st,
2nd, 3rd, 4th, so on)
o The exact distance between categories cannot be quantified
o Both quantitative and qualitative variables may yield ordinal-level data
o Examples of ordinal level data for qualitative variables:
 Perception: strongly agree, agree, neutral, disagree, strongly disagree
 Severity of Disease: mild, moderate, severe
 Height: short, medium, tall
 Weight class: lightweight, welterweight, heavyweight
If the actual height or weight measurement based on a specific unit is used to
classify the data, then the level of measurement is no longer ordinal.
 Interval Level Data
o Numerical observations of a quantitative variable, in which 0 (zero) is not absolute

(artificial)
o Conceptually, these scales are infinite (-∞ to + ∞)
o Examples of variables with interval data:
 Temperature (°C) – 0 °C doesn’t mean absence of temperature (or heat)
 Voltage – 0 volts doesn’t mean absence of charged particles
 Ratio Level Data

o Numerical observations of a quantitative variable, in which 0 (zero) is absolute (fixed
zero)
o Ratio of two numbers is meaningful
o Examples of variables with ratio data:
 Weight in Kg, Income in $, Total population
 In the ratio scale, we can say that 2kg is twice as heavy as 1kg or $60 is 3x more
than $20…
Once the research variables have been identified and defined, the research design can be
planned. Although a course on research methodology is a more appropriate venue to delve into the
nitty-gritty concepts of research design and exhaust all possible kinds of studies, it is worthy to discuss
the general and common research schemes suitable for students learning the basics of biostatistics.
Research designs may be generally divided into observational and experimental.
1. An observational study design involves the acquisition of variable data from existing phenomena,
whereby the researcher serves as an observer.
 A design in which the levels of all the explanatory variables are determined as part of the
observational process
 The investigator has not control over the explanatory variables
 Example: Study on the effect of pregnant women’s alcohol consumption on birth-weights
of their babies
Various observational study designs are common in medicine and public health in the form of
surveys known as cross-sectional studies, cohort studies and case-control studies. Data such as patient
history and demographics are collected and inferences are made with respect to a disease or condition
of interest. In biological research, a common form of observational study design would be a field
survey or prevalence study involving collection of specimens for identification and reporting.
Although there is no manipulation of variables involved, observational study designs require a
systematic method of data collection involving some form of randomization so that selection bias may
be avoided. Bias can be avoided by making sure that all possible subgroups of a target population,
such as humans, are represented in the data collection scheme. The chapter on sampling methods will
introduce the biostatistics student to the various means of acquiring data based on probabilistic
assumptions.
2. An experimental study design involves the manipulation of certain conditions (i.e. independent
variable) and measures the dependent variable.
 A stringent study design performed in a controlled environment wherein extraneous
variables are minimized if not eradicated.
 Basic components of an experiment:

o Treatment: a combination of the levels of one or more independent variables; also
known as factors
o Experimental/Observational unit: smallest unit of the study material sharing a
common treatment; e.g. an animal, a plot of land, a specimen sample, a human
subject, etc.
The experimental unit may be subjected to various conditions (independent variable or

treatment) in order to measure a reaction (dependent variable). The objective of an experiment is to
separate the treatment effects from the uncontrolled variation among units. Sources of variation in
observed responses of the experimental unit may include:
 Variation due to the effects of the independent variable/s

 Variation due to the effects of identified extraneous variable/s (e.g. genetics or health
condition of subject)
 Variation due to unidentified sources (i.e. error variation)
An experimental research design must be formulated in a manner wherein variation in

observed data is mainly due to the effects of the independent variable/s. In order to minimize
variation in observed data due to extraneous variables or unidentified sources, data acquisition
techniques may be incorporated in the experimental design:
 Randomization
o Any experimental study requires random allocation of treatments to experimental
units to avoid bias.
o All experimental units have the same chance (equal probability) of being given any
of the treatments
 Replication
o Commonly known as “replicates”
o Number of experimental units in a single treatment at one time = number of
replicates for that treatment (e.g. 5 mice per treatment, 3 plants per treatment,
etc.)
o The mean of the observations from each replicate can be subjected to statistical
tests/ inferences
 Blocking
o Groups of experimental units sharing a common level of an extraneous variable
o Based on agricultural field experiments where blocks such as plots of land share
the same soil conditions
o e.g. In comparing the effects of soil characteristics to plant yield: Experimental units
were 3 plots of land and the treatments/ independent variable is variety of a
particular plant. Plant yield is the outcome of interest.
After the complete and successful implementation of the research design, the data acquired
must be summarized and analyzed appropriately. Both descriptive and inferential statistics play key
roles in making sense out of the data collected and in answering the research objectives. Separate
chapters on data presentation, data summary and various methods of statistical tests will be covered in
this module. It is important to note that descriptive and inferential statistical methods are applicable
based on the type of variables (qualitative or quantitative), level of measurement of the data acquired
(nominal, ordinal, interval, ratio) and the number of variables being analyzed.
The ultimate goal of all research is to contribute to the greater body of knowledge.
Dissemination of research output may be achieved through publication in reputable, peer-reviewed
science journals, technical reports, oral or poster presentations in scientific meetings/conferences and
even social media for real-time critique. For immediate utilization of results, technical reports may be
submitted to policy-making bodies such as local government units for policy development or
improvement.
NAME:__________________________________________________________ DATE:______________
1. Classify whether each listed item is a constant or variable by checking on the appropriate box. If it
is a variable, specify whether qualitative, quantitative-discrete or quantitative-continuous.
(10 points)
a. Blood glucose level in mg/100ml constant variable _________________________
b. Blood type constant variable _________________________
c. Avogadro’s number constant variable _________________________
d. Dialect constant variable _________________________
e. Cases of pneumonia constant variable _________________________
f. Nutritional status constant variable _________________________
g. Grams in a Kilogram
constant variable _________________________
h. Civil status
i. Moons of Saturn
j. Per capita income
2. Specify the level of measurement of the given variables. (10 points)
a. Patient Name ____________________________________
b. Age in years and months ____________________________________
c. Gender ____________________________________
d. Digital Systolic Blood Pressure ____________________________________
e. History of Surgery ____________________________________
f. Pain Scale ____________________________________

g. Degree Program ____________________________________
h. Students Enrolled ____________________________________
i. Distance Traveled ____________________________________
j. Calories per Serving ____________________________________
3. Answer the questions related to the given research topics.
a. Methionine is a sulfur-amino acid that is added to the diets of turkeys to enhance growth.
Three types of methionine, labeled as M1, M2 and M3, were compared. Weight gain of
young turkeys over a three-week period was measured. The experimental units were 12
cages of young turkeys. The cages were stacked on top of each other, four layers high, in a
room, with three cages on the floor, three cages on the second level, three cages in the
third level up and three cages on the top level, near the ceiling. Because of concern over
temperature as a possible confounder, the cages were grouped: Group1 - 3 floor cages,
Group 2 - 3 second level cages, Group3 - 3 third level cages, Group4 - 3 ceiling cages. In
each group, the cages were assigned to M1, M2 or M3 at random.
i. What general type of research design is described above?
_______________________________________________________________________
ii. Independent Variable __________________________________________________
iii. Dependent Variable ___________________________________________________
iv. Variable Type of the Dependent Variable and Level of Measurement:
_______________________________________________________________________
v. Data Acquisition Techniques Employed in the Design:
_______________________________________________________________________
vi. Draw a schematic diagram of the research design described above.
b. In a study that would like to compare the effects of two different drugs, Drug A and Drug B
on the white blood cell count of human volunteers, the medical researcher formulated the
following null hypothesis: There is no difference in the white blood cell counts of patients
administered with Drug A and Drug B.
i. Formulate an appropriate alternative hypothesis ___________________________
___________________________________________________________________
___________________________________________________________________
ii. What general type of research design is described above?
_______________________________________________________________________
iii. Dependent Variable ___________________________________________________
iv. Variable Type of the Dependent Variable and Level of Measurement:
_______________________________________________________________________
After the collection of data, the researcher must make sense of raw data in a manner that can
contribute to the achievement of the research objectives. Data must be presented so that the research
results can be understood by peers, mentors or the general public. Data presentation methods are part
of descriptive statistics and may serve as preludes to inferential statistical methods.
Basic methods of data presentation:
1. The Narrative
2. Tabular Presentation
3. Graphical Methods
Also known as the textual method of presentation, the narrative method is a written account of
data that is more appropriate for small data sets or a few numerical figures. The narrative can be
confusing for larger data sets and should only be used to introduce or supplement tables or graphs.
For example:
“Out of the 7 subjects included in the investigation to determine the dependence of nutritional
needs on body size, 4 subjects had fat-free mass sizes that were below 68Kg. All subjects had recorded
24-hour energy expenditures that were beyond 1,500Kcal.”
Ideally, the narrative shown in the example above should be followed by other data
presentation methods including tables and/or graphs so that other information gathered that may be
relevant to the research problem will not be omitted or taken for granted.
Tabular presentation involves the systematic arrangement of statistical data in columns and
rows for a specific purpose. Tables are suitable for presenting large sets of numerical data in a
compact and orderly fashion. A large set of nominal data may also be tabulated for presentation. A
well-constructed table can clearly emphasize trends or relationships among different variables, which
may not be obvious in a narrative format. Tables must be simple, direct, and clear to the point that is
self-explanatory.
There are various table layout formats that researchers can follow such as those prescribed by
the American Psychological Association (APA), Modern Language Association (MLA) and Chicago/
Turabian. Use the layout format that your mentor, school, institution or organization recommends and
be consistent. Do not mix different table layout formats in a single research paper.
Regardless of the layout format, the table must possess the following parts:
 Table number and title
o Tables must be numbered chronologically within a research paper. A table must
appear after it is mentioned in the body of the paper.
o The title must mention the variables shown in the paper. The general location
where the data was collected and the year of data collection may be included.
 Labels for the columns and rows including the unit (e.g. %, g/dL, m/s, years, $, etc.)
 Table cells containing data
o String data (non-numerical) must be aligned consistently (all left or all center)
o Numerical data must be aligned at the right so that the place values are observed;
values may be centered as place values are aligned
o Numerical data must be consistent in the number of decimal places and must use
commas to delineate thousands, millions, billions and so on
 Footnotes or sources of data, if necessary.
Follow the font (face and size), emphasis (bold or capitalization) and capitalization rules
prescribed by the specific table layout format recommended for your paper. The tables below show
the essential elements of standard tables for numerical and textual data.
A common table that is used for presenting count data, or a tally, is the frequency distribution
table. The frequency distribution table usually shows the number of observations (frequency) for each
category and the percentage of occurrence (relative frequency in %), such as the sample table below:
Frequency distribution tables may also include cumulative frequencies and relative cumulative
frequencies. By definition:
 Cumulative frequency – sum of the frequency of the class under consideration and
frequencies of preceding classes
 Cumulative relative frequency – same as cumulative frequency applied to the relative
frequency
Modifying Table 3-3 above wherein columns for cumulative frequency and relative cumulative
frequency are included, the following frequency distribution table can be constructed:
In the previous examples for the frequency distribution table, the row items are
straightforward categorical. There are instances wherein a large numerical data set, whether interval or
ratio level data, must be reduced or condensed for better comprehension of trends. The grouped
frequency distribution table will require the construction of class intervals or groups, such as the sample
grouped frequency distribution below.
In constructing a grouped frequency distribution table, examine the raw data from the
following hypothetical study: “A study on serum creatine phosphokinase (CK) enzyme in U/l and its
effect on skeletal muscle activity among 36 male sprinters resulted to the serum CK values enumerated
below.”
58 82 151 121 100 68

163 145 201 95 64 163
94 57 60 84 139 94
203 104 113 119 110 203
110 83 93 62 67 110
42 123 48 25 70 42
First, arrange the data in ascending order from the lowest observed value to the highest.
25 60 82 95 113 151
42 62 83 100 119 163
42 64 84 104 121 163
48 67 93 110 123 201
57 68 94 110 139 203
58 70 94 110 145 203
Calculate the range of the data set, that is, subtract the lowest observed value from the highest
= 203-25 =178. Determine the number of class intervals or groups for the data set which may be from
5 groups to 15 groups. The researcher’s discretion as to the number of classes to be used must be
based on how meaningful the grouped frequency distribution will be. The table must not hide (too
few classes) nor be irrelevant (too many classes). For the example above, 10 groups will be used for
the serum CK in U/l.
The next step is to calculate the class width, or magnitude of each group, by dividing the range
by the number of classes desired = 178/10 = 17.8 or 18. The class width must have the same number
of decimal places as the observed data. After determining the class width, tally the frequency of
observations under each group then construct the grouped frequency distribution table.
It is apparent from the table above that the most frequently occurring serum CK range is 82-
100 U/l and that around half the total serum CK levels observed (55.5%) lie from 100 U/l and lower.
Grouped frequency distribution tables give meaning to numerical data sets and allow the researcher to
recognize trends.
Other tables relevant to research are dummy tables and master. A dummy table contains a
proposed table number, title, column and row headings but cells are empty. Dummy tables are
essential in student research proposals, because these guide the student researcher as to how the data
will be collected and analyzed. A master table is a single table that contains raw data on each variable
used/ examined with respect to all elementary units. The master table is where the researcher encodes
collected data in preparation for better tabulation, graphical presentation or analysis.
For the listed research topics, construct appropriate and complete tables for the given data using word
processing software (Open Office Document or Microsoft Word) at the Biostatistics computer
laboratory. Submit electronic and/or printed copies (as specified by laboratory teacher) at the end of
the laboratory period.
1. In the study by Allen and Gorski (1992) on the sexual orientation and the size of the anterior
commissure in the human brain published in the Proceedings of the National Academy of
Science, 89:7199-7202, the midsagittal area of the anterior commissure of the brain of
homosexual men, heterosexual men and heterosexual women were measured in square
millimeters. The average midsagittal areas of the anterior commissures for homosexual men,
heterosexual men and heterosexual women were 14.20, 10.61 and 12.03, respectively (5 pts).
2. Tripepi and Mitchell (1984) studied the metabolic response of river birch and European birch
roots to hypoxia which was published in Plant Physiology, 76:31-35. The researchers flooded 4
birch tree seedlings and kept 4 other seedlings as controls. All seedlings were harvested and
the amounts of ATP (nmoles per milligram tissue) in the roots were analyzed. The flooded
seedlings had ATP amounts of 1.45, 1.19, 1.05 and 1.07. Recorded ATP for controls were 1.70,
2.04, 1.49 and 1.91 (5 pts).
3. The hypocholesterolemic effects of oat-bran or bean intake for hypercholesterolemic men was
investigated by Anderson et al (1984) and published in The American Journal of Clinical
Nutrition, 40:1146-1155. The study measured the decrease in serum cholesterol level
(milligrams per deciliter) of subjects given either an oat diet or bean diet. The mean fall in
cholesterol level for oat diet subjects was at 53.6 with a standard deviation of 31.1, while for
bean diet subjects the mean decrease was at 55.5 with a standard deviation of 29.4 (5 pts).
4. Paleontologists at the Museum of Natural History measured the width (in millimeters) of the
terminal molar at the right side of the upper jaw of 36 specimens of the extinct mammal
“auroch” (Bos primigenius). The measurements are listed below (10 points):
5.9 6.1 6.1 5.7 5.9 6.0

6.5 6.1 6.7 5.8 5.9 6.2
6.3 6.3 5.7 6.2 5.7 6.0
6.2 6.2 5.7 5.8 6.1 6.2
6.1 6.2 6.0 6.1 5.9 5.4
6.5 5.9 5.7 6.1 6.0 6.1
a) What is your calculated range?

b) If the number of classes prescribed is 5, what would be the class width?
c) Construct a grouped frequency distribution table with relative, cumulative and relative
cumulative frequencies.
Data Presentation
When dealing with more voluminous raw data, various graphs may be used to show trends or
patterns which could be missed in tabulated data. Certain situations may require the use of both a
table and graph to show different perspectives of the same data set. Usually, tables and graphs should
not be redundant so it is best to use either tables or graphs. Just like tabular presentation of data,
graphs must also be simple and self-explanatory.
A graph must be labeled as “Figure,” numbered chronologically within a research paper, contain a
title and should be position after it is mentioned in the body of the text. Various statistical software
(e.g. SAS, Stata, IBM SPSS, R Statistics), technical computing software (e.g. MatLab) and spreadsheet
software (e.g. Microsoft Excel) allow the construction of graphs after encoding of raw data through a
spreadsheet interface or plain text(e.g. .csv). The common graphs used by student researchers are
discussed in this section of the module.
 Histogram
o Graphical form of a grouped frequency distribution table
o The horizontal axis (abscissa) contains the classes or groups and the vertical axis (ordinate)
corresponds to the frequency or relative frequency
o Horizontal axis variables: quantitative discrete or continuous, often used in graphing the
distribution across age groups; the frequencies are represented by the areas of the bars
o Vertical axis must always start at zero (0).
o The example below is the histogram complement of Table 3-6.
 Frequency Polygon
o Also known as an area chart, this is an alternate form of the histogram wherein the class
midpoints (middlemost value of each class) is plotted against the corresponding frequency
or relative frequency in the vertical axis. The points may be connected to each other
through a line and immediately show a trend.
o Frequency Polygons may be used to juxtapose and compare frequency distributions of two
groups.
o Horizontal axis variables: quantitative discrete or continuous
 Bar Graph
o The bar graph is commonly used to compare the frequencies or relative frequencies
among different qualitative variables (nominal or ordinal). It may be oriented vertically or
horizontally.
o Vertical bar graph: the categories are situated along the horizontal axis; compared to the
histogram, the bars are separated from each other. Subcategories of a group can be
represented by adjacent bars.
o Horizontal bar graph: the categories are situated along the vertical axis
Data Presentation
 Pie Chart
o Similar to the bar graph, pie charts are useful in graphing relative frequencies of qualitative
variables (nominal or ordinal) with few categories. There is no clear-cut rule regarding the
number of categories appropriate for a pie chart or a bar graph. What would be more
relevant is that the graph is clear, self-explanatory and concise.
o The categories are represented by the slices and the magnitude (in percent) is represented
by the size of the slice.
 Component Bar Graph
o Instead of visualizing several pie charts of different groups, a component bar graph (a.k.a
stacked bar graph) can condense the information into a single graph.
o The component bar graph is also appropriate for qualitative variables (nominal or ordinal)
with few categories.
o The graph may be oriented vertically or horizontally, where each bar corresponds to a
group and the components of the bar represent the percent distribution of the categories
of interest.
Data Presentation
 Line Graph
o The line graph is suitable for graphing time series of frequencies of qualitative variables or
quantitative variables
o The horizontal axis is for time and the vertical axis is for the frequency of the qualitative
variable or value of the quantitative variable. The aim is to show the trend of the
frequencies or values across time.
o Trends for two or more groups can be juxtaposed in a single line graph for comparison
 Scatterplot
o A scatterplot is a graphical representation of the relationship between two quantitative
variables (interval or ratio data) and usually supplement correlation or linear regression
analysis.
o Typically, the independent quantitative variable is plotted in the x-axis (abscissa) and the
dependent quantitative variable is plotted along the y-axis (ordinate).
 Stem and Leaf Diagram
o A simple graphical presentation of the distribution of a small set of observed quantitative
variables (discrete or continuous variables that have been rounded off). This is can be
considered as an oversimplified histogram.
o All values of the observed data are preserved (more or less).
o Values of data are arranged in ascending order. Each value us split into a “stem” (e.g. tens)
and a “leaf” (e.g. ones). In the diagram, the stem is adjacent to all its leaves.
o Example: radish growth in millimeters after 5 days of total darkness
15 11 20 29 8 30 33
20 35 10 22 15 37 25
 Boxplot
o To graphically compare the distribution and measures of central tendency, dispersion and
location of quantitative variables (interval or ratio data; continuous or discrete) across
different groups, boxplots or box-and-whisker plots are useful.
o A separate chapter in this module will discuss how to determine and calculate various
measures of central tendency, dispersion and location. Typically, statistical software can
generate boxplots from raw data. A single data set boxplot may be oriented horizontally.
LABORATORY EXERCISE 4
Graphical Presentation (35 points)
NAME:__________________________________________________________ DATE:______________
Generate the required graph for the given raw data using spreadsheet or statistical software (Microsoft
Excel or trial version of SPSS, STATA or open source software such as R or PSPP) at the Biostatistics
computer laboratory. Combine all complete and numbered graphs in a document (e.g. M.S. Word) file,
and submit electronic and/or printed copies (as specified by laboratory teacher) at the end of the
laboratory period.
1. Stem-and-Leaf Diagram and Single Boxplot. A study by Connolly (1968) on fruit fly preening
behavior published in Animal Behaviour, 16:385-391, the total preening time spent (in seconds)
by each fruit fly specimen (n=20) during a 6-minute observation period were recorded as
follows: (10 points)
48 22 48 29 19
18 26 57 32 25
76 33 31 46 24
34 24 10 16 52
2. Bar Graph. Based on the CIA World Fact Book, the estimated life expectancies at birth (2011) of
ASEAN countries are as follows: (5 points)
Life Expectancy Life Expectancy

Country Country
(years) (years)
Brunei 75.74 Malaysia 73.29
Burma 63.39 Philippines 71.09
Cambodia 62.1 Singapore 82.14
Indonesia 70.76 Thailand 73.10
Laos 56.68 Vietnam 71.58
3. Pie Chart. The Manila Zoological and Botanical Garden has the following groups of animal
specimens: (5 points)
Group Number of Species

Mammals 30
Reptiles 63
Birds 13
TOTAL 106
4. Component Bar Graph. A marketing student conducted a survey on the preferential rice
varieties of consumers in the 4th district of Metro Manila (CAMANAVA) and discovered that the
top three preferred rice varieties were Sinandomeng, Dinorado and Thai Jasmine. The survey
results are as follows: (5 points)
City Dinorado (%) Sinandomeng (%) Thai Jasmine (%)

Caloocan 32 40 28
Malabon 28 56 16
Navotas 56 13 31
Valenzuela 41 32 27
5. Line Graph. The average monthly precipitation data from DOST-PAGASA are summarized
below: (5 points)
Month Ave. Rainfall (cm)

Jan 1.5
Feb 0.8
Mar 1.5
Apr 2.5
May 12.1
Jun 29.2
Jul 35.3
Aug 47.5
Sep 40.5
Oct 18.5
Nov 12.3
Dec 6.2
6. Scatterplot. A resident pediatrician measured the weights and heights of 13 newborn babies
delivered for the month of August 2012 at the Philippine General Hospital for routine census.
Data are tabulated below. (5 points)
Baby 1 2 3 4 5 6 7 8 9 10 11 12 13
Weight (oz) 81 125 83 106 88 118 86 101 86 102 98 95 88
Height (in) 17.0 19.7 17.3 19.0 17.2 20.0 17.5 18.7 17.0 18.8 19.4 18.0 18.1
In this chapter, we will learn several techniques for organizing and summarizing data so that
we may easily determine what information they contain. The ultimate in data summarization is the
calculation of a single value that in some ways conveys important information about the data from
which it was calculated. Such single values that are used to describe data are called descriptive
measures. After studying this chapter, you will be able to calculate several descriptive measures for
both populations and samples of data.
First, we will learn how to calculate measures of central tendency, location, dispersion when
data are ungrouped.
 Convey information regarding the average value of a set of values.

 They indicate where the majority of values in the distribution are located.
 These measures can be considered as the center of the probability distribution from which the
data were sampled.
 Sum of the individual values in a data set divided by the number of values in the data set.
General formula for a finite population mean General formula for the sample mean
µ
where: where:
 µ = population mean  ̅ = sample mean
 N = population size  n = no. of observations
 xi = individual observation  xi = individual observation
 i = position of observation  i = position of observation
 Σ = summation sign  Σ = summation sign
Example:
In an outbreak of dengue fever, 10 people became ill with clinical symptoms 3 to 14 days after
exposure to the virus. In this example, we will illustrate how to calculate the sample mean period for
the dengue outbreak. The incubation periods of the affected people (xi) were 7, 4, 3, 12, 8, 9, 6, 5, 5,
and 14 days.
1. To calculate the numerator, sum the individual observations
2. For the denominator, count the number of observations: n = 10

3. To calculate the mean, divide the numerator (sum of observations) by the denominator
(number of observations):
̅ *
*A reasonable rule is to express the mean with one more significant digit than the observations
Therefore, the mean incubation period for this outbreak was 7.3 days.
 The mean or average of a set of data measured on a logarithmic scale.

 Consider the value of 100 and a base of 10 and recall that the logarithm is the power to which
a base is raised. For example, the logarithm of 100 at base 10 is 2 since 102 is equal to 100.
 An antilog raises the base to the power (logarithm). For example, the antilog of 4 at base 2 is
16.
 Is calculated as the nth root of the product of n observations.

 The geometric mean is a good summary measure in situations when data follow an
exponential or logarithmic pattern typical in dilution assays such as serum antibodies, as well
as environmental sampling data.
Note: to calculate the geometric mean, you will need a scientific calculator with log and yx keys.
Descriptive Measures
̅ √
In practice, the geometric mean is calculated as:
̅ ( ∑ )
where:
 ̅ = geometric mean
 x1 = lowest value in the set of observations
 xn = highest value in the set of observations
 n = number of observations
 Σ = summation sign
Example:
Using the titers given on the next page, calculate the geometric mean titer of antibodies
against human parainfluenza virus among the seven patients.
ID# Dilution Titer

1 1:256 256
2 1:512 512
3 1:4 4
4 1:2 2
5 1:16 16
6 1:32 32
7 1:64 64
Using the second formula, we get

̅ [ ⁄ ]
̅ [ ⁄ ]
̅ ( ⁄ )
̅
̅
Therefore, the geometric mean titer is 32, and the geometric mean dilution is 1:32
 The mean is the “center of mass”

 It uses all the observed values in the calculation
 It may or may not be an actual observed value in the data set
 It is algebraically tractable, which means that mean values can be computed directly
 Its value is affected by outliers
 The mean of a finite data set always exists and is unique
 Data values should be measured using at least an interval scale
Calculating the arithmetic mean using Microsoft Excel:
1. Input your data into the spreadsheet in an organized manner.

2. Select the cell directly underneath the individual observations. In this case, cell B19.
3. Under the Formulas tab, click Auto Sum and wait for the drop down menu to appear.
4. Click on Average.
5. Press Enter.
Calculating the geometric mean using Microsoft Excel:


2. Select the cell directly underneath the individual observations. In this case, cell C11.
3. Under the Formulas tab, click More Functions > Statistical and wait for the drop down menu to
appear.
4. Click on GEOMEAN.
5. A menu called Function Arguments will appear.
6. Select the cells comprising the individual observations. In this case, select C4 through C10.
7. Click on OK.
 From the previous example on the average incubation period during a dengue fever outbreak,
only four observations were higher than the mean of 7.3 days.
xi : 7, 4, 3, 12, 8, 9, 6, 5, 5, 14
 Thus, the mean is not very representative of the data set as a whole. The median might be
more appropriate
 The median is defined as the value which divides a finite set of values into two equal parts
such that the number of values equal to or greater than the median is equal to the number of
values equal or less than the median.
 In other words, the median is the middlemost observation in a set of observations arranged in
numerical order or in an array.
1. Arrange the observations in increasing or decreasing order. Microsoft Excel has a sort function
for arranging data into an array.
2. Find the middle rank with the following formula:
a. If the number of observations (n) is odd, the middle rank falls on an observation.
b. If n is even, the middle rank falls between two observations
NOTE: The formula only computes for the position of the median and the median itself!
3. Identify the value of the median:
a. If the middle rank falls on a specific observation (that is, if n is odd), the median is equal to
the value of that observation.
b. If the middle rank falls between two observations (that is, if n is even), the median is equal
to the average (i.e., the arithmetic mean) of the values of those observations.
Example:
From the previous example, the incubation period of 10 patients affected by the dengue fever
outbreak, arranged from lowest to highest, are:
3 4 5 5 6 7 8 9 12 14
1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th
[The median is between the 5th and 6th observations]
Median = (6+7)/2 = 6.5 days
Interpretation: Half of the patients who had dengue fever had an incubation period less than
6.5 days and half had an incubation period greater than 6.5 days.
Suppose there were 11 patients, and the 11th had an incubation period of 14 days.
3 4 5 5 6 7 8 9 12 14 14
1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th 11th
[The median falls exactly on the 6th observation]
Thus, the median is 7 days.
Interpretation: Half of the patients who had dengue fever had an incubation period less than
or equal to 7 days and half had an incubation period greater than 7 days.
 The median is the “center” of the array
 Unlike the mean, it uses only the middle value(s) in the array for its computation
 The median is not affected by outliers
 The median itself is not algebraically tractable. Only the position of the median is.
 The median is still interpretable when the scale of measurement used is as low as ordinal
 The median will always exist and is unique
Calculating the median using Microsoft Excel:

appear.
4. Click on MEDIAN.

6. The cells of your individual observations will be automatically selected.
7. Click on OK.
 The value which occurs most frequently in a set of observations.
 The mode is usually found by creating a frequency distribution in which we tally how often
each value occurs.
 It is possible to have no mode, one mode (unimodal distribution), two modes (bimodal
distribution), or more than two modes (multimodal distribution)
Example:
Using the previous example where the incubation period of 10 patients during a dengue fever
outbreak was recorded, the mode would be 5 days since the value occurred twice while all other
values occurred only once.
7, 4, 3, 12, 8, 9, 6, 5, 5, 14
Interpretation: The usual incubation period of a dengue fever patient is 5 days.

 The mode is the “center” in the sense that it is the most typical value in a set of observations
 Outliers do not affect the mode
 The mode is not algebraically tractable for ungrouped data
 The mode will not always exist; and if it does, it may not be unique
 The value of the mode is always one of the observed values in the data set
 The mode can be obtained for both quantitative and qualitative types of data; that is, the
mode is interpretable even if the scale of measurement is as low as nominal
 Generally, the mode is not as useful as a measure of central tendency as the mean and the
median when the data consists of only a few observations. For example, for the values 5, 14,
20, 26, 37, 37, the mode is 37 since it occurred twice and all other numbers only once.
However, 37 cannot be considered a good measure of central tendency for this set of data
since it is in fact, at the extreme high end of the values and its frequency exceeds the
frequency of the other values by only 1.
Calculating the mode using Microsoft Excel:

2. Select a few cells directly underneath the individual observations. In this case, cell B14. This is
important in case the distribution is multimodal. Selecting only a single cell will give only one
of the modes if the distribution is multimodal.
appear.
4. Click on MODE.MULT. [This will ensure that the values of all modes are returned in case the
distribution is multimodal in contrast to selecting MODE.SNGL which will only return one
mode]

6. Select the cells comprising the individual observations. In this case, select B4 through B14.
7. Press CTRL + SHIFT + ENTER. [Clicking on OK will return only a single mode even if the
distribution is multimodal which is similar to using the MODE.SNGL function].
8. In this case, all the cells selected in the vertical array will contain the value “5”. If the data were
changed to have two modes, the output will look like this:
*The value in B8 was changed from 8 to 9, making the distribution multimodal.
Note that both modes (9 and 5) were given in the vertical array. The remainder of the cells in
the vertical array show #N/A which means that there are no other modes to be displayed
within those cells.
When we observe the graph of a frequency distribution, we generally notice two main features:
1) The graph has a peak, commonly near the center; and 2) it spreads out on either side of the peak.
Just as a measure of central tendency is used to describe where the peak is located, a measure of
dispersion is used to describe how much spread there is in the distribution. Several measures of
dispersion are available.
 The difference between the smallest and largest value in a set of observations.
 In the statistical world, the range is reported as a single number, the difference between the
maximum and minimum. In the epidemiologic community, the range is often reported as
“from (the minimum) to (the maximum)”, i.e., two numbers.
where:
 R = range
 xL = largest value
 xS = smallest value
Example:
Still using the previous example where the incubation period of 10 patients during a dengue
fever outbreak was recorded (xi : 7, 4, 3, 12, 8, 9, 6, 5, 5, 14), the range is equal to:
Interpretation: This means that the patients differed in the time when they began presenting
symptoms of dengue fever by as much as 11 days.
 Measure of a variable’s spread or distribution around its mean.
 Takes into account the squared deviations of individual observations from the mean
General formula for the finite population variance General formula for the sample variance
Where Where
 2
σ = population variance  s2 = sample variance
 µ = population mean  ̅ = sample mean
 N = population size  n = sample size
 xi = individual observation  xi = individual observation
 i = position of observation  i = position of observation
 Σ = summation sign  Σ = summation sign
Example:
Still from our previous example: 7, 4, 3, 12, 8, 9, 6, 5, 5, 14
s2 = (7 – 7.3)2 + (4 – 7.3)2 +… (3 – 7.3)2
s2 = 12.5 days2
Alternative formula for the sample variance:
means square the individual observations then take the sum of the squared observations
means take the sum of the individual observations, then square the sum
Incubation Period
Patient No. xi2
in days (xi)
1 7 49
2 4 16
3 3 9
4 12 144
5 8 64
6 9 81
7 6 36
8 5 25
9 5 25
10 14 196
Total 73 645
Calculating the variance using Microsoft Excel:

appear.
4. Click on VAR.P if you wish to calculate the population variance or VAR.S if you wish to calculate
the sample variance.

6. The cells of your individual observations will be automatically selected.
7. Click on OK.
 Simply the square root of the variance

 Note that the units of measurement for the variance are in units square (e.g. grams2, meters2).
This is a major drawback for the variance’s use as a measure of dispersion – it is difficult for
most people to think in terms of squared units.
 The formula for the sample standard deviation is:
From our previous example, the standard deviation is √

Interpretation: This means that on average, the incubation period of patients affected by the
dengue fever outbreak is ± 3.5 units away from the mean (7.3 ± 3.5).
 It uses every observation in its computation.

 It may be distorted by outliers. This is because squaring large deviations from the mean will
give more weight to these outliers.
 It is algebraically tractable.
 It is always nonnegative. A value of 0 implies the absence of variation.
 The scale of measurement must at least be interval for the standard deviation to be
interpretable.
Calculating the standard deviation using Microsoft Excel:
Follow the same steps as if you were calculating variance except that you must change the
function to STDEV.P if you wish to calculate the population standard deviation or STDEV.S if you wish to
calculate the sample standard deviation.
 A measure of relative dispersion which expresses the standard deviation as a percentage of the
mean
 May be used to compare SDs of two variables measured in different units or used when two
means, although measured in the same unit, differ appreciably.
 The formula for the coefficient of variation is:
̅
where s is the sample standard deviation, and ̅ is the sample mean
Example:
Suppose two samples of female chimpanzees yield the following results:
Sample A Sample B
Age 5 years 15 years
Mean Height 45 cm 70 cm
Standard Deviation 8 cm 8 cm
We wish to know which is more variable, the heights of the 5-year old female chimpanzees or
the heights of the 15-year old female chimpanzees. Comparing the standard deviation of the two
samples may mislead us that both are equally variable. However, by using the coefficient of variation,
we get a different impression.
 CV for Sample A
 CV for Sample B
Interpretation: The heights of the 5-year old chimpanzees are more dispersed than the heights
of the 15-year old chimpanzees.
The coefficient of variation (CV) is a sample statistic. Its corresponding population parameter is
called the coefficient of dispersion (CD) and is given by the following formula:
Measure
Type of Distribution
Central Tendency Dispersion
normal arithmetic mean standard deviation
skewed median interquartile range
exponential or logarithmic geometric mean consult statistician

SOURCE: Principles of Epidemiology, Centers for Disease Control and Prevention (1992)
 Percentile ( ) – one of the 99 values of a variable which divides the distribution into 100 equal
parts.
 There are 99 percentiles, denoted by . The percentile, denoted by
, is a value such that at least of the observations are less than or equal to it and
at least are greater than it, where
 Decile ( ) – one of the 9 values of a variable which divides the distribution into 10 equal parts.
 Quartile ( ) – one of the 3 values of a variable which divides the distribution into 4 equal
parts.
P P P
25 50 75
P P P P P P P P
10 20 30 40 60 70 80 90
D D D D D D D D
1 2 3 4 6 7 8 9
Q D Q
1 5 3
Q
2
1. Order the data in increasing order of magnitude.

2. Compute where n is the sample size while k is the percentile of interest.
3. If is not an integer, the percentile is the largest measurement where j is the
largest integer less than .
[1’ denotes adding the necessary value to to raise it to the nearest integer]
4. If is an integer, the percentile is the average of the and
[ ] largest observations.
Additional Notes on the Interpretation of
 will be an interpolated value if is not an integer. If the values used in the

interpolation are not tied values then will not be one of the observations. In such a case, the
interpretation of will simplify as follows:
“ of observations are less than . Likewise, are greater than ”.
 That is, observations are less than so that the remaining =
are greater than .
A note on the calculation of percentiles
There is no standard definition of a percentile. The method used here relies on calculating
to determine the position of the percentile. Another method to estimate the position of the
percentile is while Excel uses . In addition, the latter two methods uses linear
interpolation to obtain a more accurate estimate of the percentile when it does not fall on a single
observation. When calculating percentiles using Excel, we suggest the use of the PERCENTILE.EXC
function. For more information about the proper use of this function, please visit the following URL:
http://office.microsoft.com/en-us/excel-help/percentile-exc-function-HA010345439.aspx
Note that the PERCENTILE.EXC function is included only in the 2010 version of Microsoft Excel.
The PERCENTILE function present in earlier versions of Excel is similar to the PERCENTILE.INC function
currently present in Excel 2010. These two functions, including the method we discussed here for
calculating percentiles by hand, yield different results. However, when the number of observations is
very large, all these methods will yield similar results.
Example:
Using once again the data on the incubation period of the 10 patients affected by the dengue
fever outbreak, find the 90th percentile (xi : 3, 4, 5, 5, 6, 7, 8, 9, 12, 14).
Since 9 is an integer, the percentile is the average of the 9th and 10th largest observations.
Interpretation: 90% of the patients had an incubation period less than 13 days while 50% of
the patients had an incubation period greater than 13 days.
Using the same data, this time, find the 23rd percentile.
Since 2.3 is not an integer, the 23rd percentile is the largest measurement where j is the
largest integer less than 2.3. Thus, j = 2 and j + 1 = 3. The 23 percentile falls on the 3rd observation.
rd
Interpretation: 23% of the patients had an incubation period less than or equal to 5 days while
77% of the patients had an incubation period greater than 5 days.
Note the difference in the interpretation when is an integer and when it is not.
 To compute for deciles and quartiles, first determine the equivalent percentile, then use the
equation for the percentile.
 Using the previous data, compute for D8.
1. Determine the corresponding percentile
D8 = P80
2. Calculate as you would a percentile
 Represents the central portion of the distribution, and is calculated as the difference between
the third or upper quartile ( ) and the first or lower quartile ( ).
 This range includes about one-half of the middlemost observations in the set, leaving one-
quarter of the observations on each side.
1. Arrange the observations in increasing order.
2. Find the position of the 1st and 3rd quartiles.
3. Identify the value of the 1st and 3rd quartiles.
4. Calculate the interquartile range as minus .
Example:
Still using the data on the incubation period of the 10 patients affected by the dengue fever
outbreak, find the interquartile range (xi : 3, 4, 5, 5, 6, 7, 8, 9, 12, 14).
The position of , which is equivalent to is given by:
Since 2.5 is not an integer, the position of is the 3rd observation and is equal to 5 days.
The position of , which is equivalent to is given by:
Since 7.5 is not an integer, the position of is the 8th observation and is equal to 9 days.
Therefore, the interquartile range is
Now that we have learned how to calculate the measures of central tendency, dispersion, and
location for ungrouped data, let us now discuss how to calculate these measures when the data are
arranged in a frequency distribution.
 In calculating the mean from grouped data, we assume that all values falling into a particular
class interval are located at the midpoint of the interval.
where:
 xi is the midpoint of the ith interval
 fi is the frequency of observations in the ith interval
 k is the number of categories
 n is the total number of observations
 The midpoint of a class interval is also called the class mark. It may be calculated by dividing by
taking the sum of either the stated or true lower and upper limits of the same interval and
dividing it by 2.
We will use the following data in all examples for grouped data calculations:
Clasen et al. studied sparteine and mephenytoin oxidation in a group of participants who
inhabit Greenland. Two populations were represented in their study: East Greenlanders and West
Greenlanders. The investigators were interested in genetic polymorphisms between the two groups.
The ages of the participants in this study are summarized in the following frequency distribution table:
Class Interval Frequency

10 – 19 4
20 – 29 66
30 – 39 47
40 – 49 36
50 – 59 12
60 – 69 4
Total 169
SOURCE: Knud Clasen, Laila Madsen, Kim Brøsen, Kurt Albøge, Susan Misfeldt, and Lars F. Gram, “Sparteine and Mephenytoin
Oxidation: Genetic Polymorphisms in East and West Greenland,” Clinical Pharmacology & Therapeutics, 49, (1991), 624-631
Example:
Class Interval Frequency (fi) Midpoint (xi) fixi
10 – 19 4 14.5 58.0
20 – 29 66 24.5 1617.0
30 – 39 47 34.5 1621.5
40 – 49 36 44.5 1602.0
50 – 59 12 54.5 654.0
60 – 69 4 64.5 258.0
Total 169 5810.5
̅
Interpretation: The average age of the participants in the sparteine and mephenytoin oxidation
study is 34.4 years.
The steps for calculating the median from grouped data is as follows:
1. Construct the cumulative frequency distribution (CFD)
2. Calculate n/2, where n is the number of observations
3. Starting from the top, locate the value in the CFD column that is ≥ n/2 for the first time. The
class interval corresponding to that value is the median class.
4. Approximate the median using the formula:
where:
 lm is the true lower limit of the median class
 wm is the class width of the median class
 n is the total no. of observations
 cfm-1 is the cumulative frequency of the class before the median class
 fm is the frequency of the median class
 Unlike computing the mean from grouped data where it is assumed that the values within a
class interval are located at the midpoint, in computing the median, we assume that they are
evenly distributed through the interval.
Example:
Class Interval Class Boundaries Frequency (fi) Cumulative Frequency (cf) cf ≥ n/2 = 84.5?
10 – 19 9.5 – 19.5 4 4 No
20 – 29 19.5 – 29.5 66 70 No
30 – 39 29.5 – 39.5 47 117 Yes
40 – 49 39.5 – 49.5 36 153
50 – 59 49.5 – 59.5 12 165
60 – 69 59.5 – 69.5 4 169
Total 169
*The class shaded pink is the median class
Interpretation: Half of the participants in the sparteine and mephenytoin oxidation study have
an age less than 32.6 years while the other half have an age greater than 32.6 years.*
*Note that the interpretation uses the less than statement rather than the less than or
equal to statement. This is the appropriate interpretation for the median of grouped data
and is true for the percentiles of grouped data as well.
The steps for calculating the mode from grouped data is as follows:
1. Locate the modal class. For a frequency distribution with equal class sizes, the modal class is
the class with the highest frequency.
2. Approximate for the mode using the formula:
where:
 lmo is the true lower limit of the modal class
 wmo is the class width
 fmo is the frequency of the modal class
 f1 is the frequency of the class preceding the modal class
 f2 is the frequency of the class following the modal class
Example:
Class Interval Class Boundaries Frequency (fi)
10 – 19 9.5 – 19.5 4
20 – 29 19.5 – 29.5 66
30 – 39 29.5 – 39.5 47
40 – 49 39.5 – 49.5 36
50 – 59 49.5 – 59.5 12
60 – 69 59.5 – 69.5 4
Total 169
*The class shaded pink is the modal class
( )
Interpretation: The usual age of the participants involved in the study of sparteine and
mephenytoin oxidation is 27.15 years.
The range is simply the lower class limit of the first class subtracted from the upper class limit
of the last class.
where:
 is the upper class limit of the last class
 is the lower class limit of the first class
Example:
Class Interval Frequency
10 – 19 4
20 – 29 66
30 – 39 47
40 – 49 36
50 – 59 12
60 – 69 4
Total 169
Interpretation: The age difference between the oldest and the youngest sparteine and
mephenytoin oxidation study participant is 59 years.
 It is a simple measure (easy to compute and understand)

Weaknesses:
 It fails to communicate any information about the clustering or the lack of clustering of values
in the middle of the distribution since it uses only the extreme values (minimum and
maximum)
 An outlier can greatly affect its value
 It tends to be smaller for smaller collections than for larger collections
 It cannot be approximated from frequency distributions with an open-ended class
 It is not algebraically tractable
 In calculating the variance (and standard deviation) from grouped data, we assume that all the
values falling into a particular class interval are located in the midpoint of the interval.
where:
 s2 = sample variance
 k = no. of categories
 fi = frequency of the ith category
 xi = midpoint of the ith category
 n = total no. of observations
Example:
Class Interval Frequency (fi) Midpoint (xi) fixi fixi2
10 – 19 4 14.5 58.0 841
20 – 29 66 24.5 1617.0 39616.5
30 – 39 47 34.5 1621.5 55941.75
40 – 49 36 44.5 1602.0 71289
50 – 59 12 54.5 654.0 35643
60 – 69 4 64.5 258.0 16641
Total 169 5810.5 219972.3
As mentioned earlier, the variance is difficult to interpret due to the units of measurement
being squared. Thus, let us compute for the standard deviation which is simply the square root of the
variance.
√ √
Interpretation: The ages of the participants in the sparteine and mephenytoin oxidation study
vary by, on average, ±10.96 years from the mean of 34.4 years (34.4±10.96 years).
The steps for calculating the percentile from grouped data is as follows:
1. Construct the CFD
2. Compute for
3. Locate the percentile class ( class). The class is the class interval where the less than
cumulative frequency is ≥ for the first time, starting from the top
4. Use the following formula to approximate :
where:
 Pk = kth percentile to be computed
 lk = true lower limit of the percentile class
 wk = width of the percentile class
 n = total no. of observations
 cfk-1 = cumulative frequency of the class before the percentile class
 fk = frequency of the percentile class
Example:
Using the previous data, compute for
Class Interval Class Boundaries Frequency (fi) Cumulative Frequency (cf) cf ≥ nk/100 = 126.75?
10 – 19 9.5 – 19.5 4 4 No
20 – 29 19.5 – 29.5 66 70 No
30 – 39 29.5 – 39.5 47 117 No
40 – 49 39.5 – 49.5 36 153 Yes
50 – 59 49.5 – 59.5 12 165
60 – 69 59.5 – 69.5 4 169
Total 169
*The class shaded pink is the percentile class
[ ]
Interpretation: 75% of the participants in the sparteine and mephenytoin oxidation study had
an age less than 42.21 years while 25% had an age greater than 42.21 years.
NAME:__________________________________________________________ DATE:______________
1. When computing for the sample variance or standard deviation, what is the rationale behind
dividing by instead of ? (5 points)
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
2. The following table shows the number of hours 45 hospital patients slept following the
administration of a certain anesthetic.
7 10 12 4 8 7 3 8 5
12 11 3 8 1 1 13 10 4
4 5 5 8 7 7 3 2 3
8 13 1 7 17 3 4 5 5
3 1 17 10 4 7 7 11 8
a. Determine the mean, median, and mode. Interpret the results (6 points)
Mean (2 pts) Interpretation (1 pt)
Median (2 pts) Interpretation (1 pt)

Mode (2 pts) Interpretation (1 pt)
b. Determine the variance and standard deviation. Interpret the standard deviation (5 points)
Variance (3 pts)
Standard Deviation (1 pts) Interpretation (1 pt)
c. Determine , , and the interquartile range. Interpret the results (11 points)
(2 pts) Interpretation (1 pt)

Interquartile Range (4 points) Interpretation (1 pt)
3. The following table shows the frequency distribution of serum cholesterol levels (mg/dl) in a 4,462
individuals who reported for hypercholesterolemia. Use the supplied columns to write down the
necessary information that you need to solve the following problems.
Class Class Midpoint Freq. Cum. cf ≥ cf ≥ nk/100?

Interval Boundaries fixi fixi2
(xi) (fi) freq. (cf) n/2? P88 D3 Q1
60-99 9
100-139 111
140-179 811
180-219 1677
220-259 1285
260-299 464
300-339 88
340-379 11
380-419 6
Total 4462
a. Determine the mean, median, and mode. Interpret the results (9 points)
Mean (2 pts) Interpretation (1 pt)
Median (2 pts) Interpretation (1 pt)
Mode (2 pts) Interpretation (1 pt)

b. Determine the variance and standard deviation. Interpret the results (5 points)
Variance (3 pts)
Standard Deviation (1 pts) Interpretation (1 pt)
c. Determine the 88th percentile, 3rd decile, and 1st quartile. Interpret the results (9 points)


Probability is a mathematical concept that determines the likelihood of occurrence of events that
are subject to chance. When we say an event is subject to chance, we mean the outcome is in doubt
and there are at least two possible outcomes.
The origins of probability can be traced to gambling. Games of chance provide good
demonstrations of what the possible events are. Typical statements that you might see concerning
probability include: the chance of getting a flush in a 5-card poker hand is about 2 in 1000 or 0.2%; the
chance of throwing a sum of 10 with two dice is 1 in 12 or 8.33%, or the probability that a ball will land
on red in a roulette wheel is 1 in 2 or 50%. Meanwhile, in the health sciences setting, one might be
interested in the probability that a patient who receives a novel medical treatment will live for two or
more years. We may hear a physician say that a patient has a 50-50 chance of surviving a particular
operation. Knowledge of the probability of these outcomes can help in making better informed
decisions, for example, whether or not the patient should undergo the operation if the chance of
surviving is small. Another demonstration of probability lies on the fact that many events in life are
uncertain. We do not know whether it will rain tomorrow or when the next disaster would strike.
Probability is a formal way to measure the chance of these uncertain events.
As these examples suggest, probabilities are usually expressed in terms of percentages or

fractions (percentages are the result of fractions multiplied by 100). Therefore, the probability of
occurrence of some event is measured by a number between zero and one. The more likely the event,
the closer it is to one and the more unlikely the event, the closer it is to zero. An event that cannot
occur has a probability of zero, and an event that is certain to occur has a probability of one.
1. Objective probability – the view that the likelihood of the outcome of any event is an objective
phenomenon derived from objective processes. This view of probability can be further classified
into classical (or a priori) probability and the relative frequency (or a posteriori) probability.
This treatment of probability originated from the work of two mathematicians, Pascal and
Fermat, and dates back to the 17th century. Much of this theory culminated out of attempts to solve
problems regarding games of chance, such as those involving cards or the rolling of dice. The
principles involved in classical probability are very well illustrated by examples from games of chance.
For example, if a fair six-sided die is rolled, the probability that a 1 will be observed is equal to 1/6 and
is the same for the other five faces. If a card is drawn at random from a well-shuffled deck of standard
playing cards, the probability of drawing a spade is 13/52 or ¼. In this view of probability, probabilities
are calculated by the process of abstract reasoning. Rolling a die or drawing cards from a deck is not
required to compute for these probabilities. In the rolling of the die, there is an equal likelihood of
observing any of the six sides if there is no reason to favor any one side. Likewise, if there is no reason
to favor the drawing of a specific card from a deck of cards, it can be said that each of the 52 cards has
an equal likelihood of being drawn. Probability in the classical sense can be defined as follows:
If an event can occur in N mutually exclusive and equally likely ways, and
if m of these possess a characteristic, E, the probability of occurrence of E
is equal to m/N.
If P(E) is read as “the probability of E,” this definition can be expressed as
Also known as the frequentist approach to probability, this relies on the repeatability of some
process and the ability to count the number of repetitions, as well as the number of instances that
some event of interest takes place. In this context, the probability of observing some characteristic, E,
of an event can be defined as follows:
If an event can occur in N mutually exclusive and equally likely ways, and
if m of these possess a characteristic, E, the probability of occurrence of E
is approximately m/N.
In compact form, this definition can be expressed as
For example, we are interested in knowing the probability of some event in some process. This
event could be the number of times a head is obtained in the process of tossing a coin. Suppose that
we toss the coin many times, where we are careful to toss the coin in the same manner each time the
process is repeated. Suppose this experiment was repeated 50 times and the following results were
recorded:
Toss No. Result Toss No. Result Toss No. Result Toss No. Result Toss No. Result
1 H 11 T 21 T 31 T 41 T
2 T 12 H 22 H 32 T 42 H
3 H 13 T 23 H 33 H 43 T
4 T 14 H 24 T 34 H 44 H
5 H 15 T 25 H 35 T 45 T
6 H 16 H 26 H 36 H 46 H
7 T 17 H 27 T 37 H 47 T
8 H 18 T 28 T 38 T 48 H
9 H 19 T 29 H 39 T 49 H
10 T 20 T 30 H 40 H 50 H
Basic Concepts in Probability
To approximate the probability of obtaining a head or P(H) during a coin toss, we can count
the number of heads (H) obtained in the trials (27) and divide by the total number of trials (50). That is,
the probability of observing a head is roughly the relative frequency of heads observed in the
experiment.
Comments about the frequentist definition of probability:
a. The observed relative frequency is only an approximation to the true probability of an

event. However, as the number of trials is increased, one might expect the relative
frequency to become a better approximation of the true probability. If the coin was
tossed 100 times, 200 times, 300 times, and so on, we would observe that the
proportion of heads observed would become closer and closer to the true probability
of 0.50. A controversial claim of this approach is that in the long run, as the number of
trials approaches infinity, the relative frequency will converge exactly to the true
probability:
b. This view of probability depends on the important assumption that the process or
experiment can be repeated many times under similar circumstances. In the case
where this assumption does not hold true, the subjective interpretation of probability
is useful.
2. Subjective probability – also called the personalistic concept of probability, this view holds that
probability measures the confidence that a particular individual has in the truth of a particular
proposition or occurrence of event. Unlike objective probability, this view does not rely on the
repeatability of any process. In fact, one may evaluate the probability of an event that can only
happen once by applying this concept of probability. For example, you are likely interested
that you will get a grade of 1.00 in this course. Most likely, you will take this course only one
time; even if you retake this course next semester, you would not be taking it in the same
conditions as this semester. You will have a different instructor, a different set of quizzes and
exams, and possibly different work conditions. In the case where the process occurs only once,
how do we view probabilities? In the subjective view of probability, a person assigns a number
to this event which reflects his or her personal belief in the likelihood of this event occurring. If
you are doing well in this course and you think that a grade of 1.00 is a certainty, then you
would assign a probability of 1 to this event. Meanwhile, if you are experiencing difficulties in
this course, you might think that getting a grade of 1.00 is close to impossible and so you
would assign a probability close to 0. What if you are not sure about the grade that you will
get? In this case, you would assign a number to this event between 0 and 1.
Comments about this view of probability:
a. Subjective probability reflects a person’s opinion about the likelihood of an event. If

the event of interest is “you will get a 1.00 in this class”, then your opinion about the
likelihood of this event is probably different from your instructor’s or your classmate’s
view about this event. Subjective probabilities are personal and they will differ between
people.
b. Can I assign any number to events? The numbers you assign must be proper
probabilities. That is, they must satisfy some basic rules that all probabilities obey. In
addition, they should reflect your opinion about the likelihood of the events.
c. Assigning probabilities to events is not an easy task, especially when you are uncertain
whether the event will occur or not. However, comparing the likelihoods of different
events can be used as a guide in assigning subjective probabilities in a process called a
calibration experiment.
 An event is the basic element to which probability can be applied. It is the result of an
observation or experiment, or the description of some potential outcome.
 Elementary events are the building blocks of a probability model. They are events that cannot
be broken down or decomposed into smaller sets of events.
For example, we roll two dice at the same time. Assume that the two dice are fair (they do not
favor any face or number) and are independent of one another (that is, the outcome in one die would
not affect the outcome in the other). We sum the two faces and are interested in the event that the
faces add up to 7. For each die there are 6 faces numbered 1 to 6 with dots. Each face is assumed to
have an equal 1/6 chance of landing up. In this case, there are 36 equally likely elementary events or
outcomes for a pair of dice. These elementary events are denoted by pairs, such as {2, 3}, which
denotes a roll of 2 on one die and 3 on the other. The 36 elementary events are: {1, 1}, {1, 2}, {1, 3},
{1, 4}, {1, 5}, {1, 6}, {2, 1}, {2, 2}, {2, 3}, {2, 4}, {2, 5}, {2, 6}, {3, 1}, {3, 2}, {3, 3}, {3, 4}, {3, 5}, {3, 6}, { 4, 1}, {4, 2},
{4, 3}, {4, 4}, {4, 5}, {4, 6}, {5,1}, {5, 2}, {5, 3},{5, 4},{5, 5}, {5, 6}, {6, 1}, {6, 2}, {6, 3}, {6, 4}, {6, 5}, and {6, 6}
These 36 elementary events constitute the sample space (commonly denoted as S, Ω, or U)

which is the set of all possible outcomes for the experiment. Meanwhile, the sample space for a coin
flip is S = {H, T}. The probability of an event E is determined by first defining the set of all possible
elementary events, associating a probability with each elementary event, and then summing the
probabilities of all elementary events that imply the occurrence of E. The elementary events are distinct
and mutually exclusive.
The term mutually exclusive means that for elementary events E1 and E2, if E1 happens then E2
cannot happen and vice versa. For example, you can pass this course but not fail at the same time and
vice versa. This property is necessary to sum probabilities, as we will discuss later.
In this example, seven occurs if we have {1, 6}, {2, 5}, {3, 4}, {4, 3}, {5, 2}, or {6, 1}. That is, the
probability of observing a sum of seven is 6/36 = 1/6 ≈ 0.167.
1. Given some process (or experiment) with n mutually exclusive outcomes (called events),
E1, E2,…, En, the probability of any event Ei is assigned a nonnegative number. That is,
P(Ei) ≥ 0
In other words, all events must have a probability greater than or equal to zero, a reasonable condition
in view of the difficulty of thinking of negative probability. A key concept in the statement of this
property is the concept of mutually exclusive outcomes. Two events are said to be mutually exclusive if
they cannot occur simultaneously.
2. The sum of the probabilities of all mutually exclusive events is equal to 1.
P(E1) + P(E2) + … + P(En) = 1
This is the property of exhaustiveness and denotes the fact that the observer of a probabilistic process
must consider all possible events, and when all are taken together, their total probability is 1. The
requirement that the events be mutually exclusive is specifying that the events E1, E2,…,E3 do not
overlap.
3. Consider any two mutually exclusive events, Ei and Ej. The probability of the occurrence of
either Ei or Ej is equal to the sum of their individual probabilities.
P(Ei or Ej) = P(Ei) + P(Ej)
Suppose the two events were not mutually exclusive; that is, suppose they could occur at the same
time. In attempting to compute for the probability of the occurrence of either Ei or Ej the problem of
overlapping would be discovered, and the process could become rather complicated.
Operations with Events
 Intersection – the intersection between two events A and B, denoted as A ∩ B, is defined as the
event “both A and B”.
A A∩B B
∩
 Union – The union of two events A and B, denoted as A ∪ B, is defined as the event “either A or
B”.
A∪B
∩
 Complement – The complement of an event A, denoted as Ac or A , is defined as the event

“not A”
A
A
Sample Problem: Calculating the probability of an event
In an article in The American Journal of Drug and Alcohol Abuse, Erickson and Murray state that
women have been identified as a group at particular risk for cocaine addiction and that it has been
suggested that their problems with cocaine are greater than those of men. Based on their review of
scientific literature and their analysis of the results of an original research study, the authors argue that
there is no evidence that women’s cocaine use exceeds that of men, that women’s rate of cocaine use
are growing faster than men’s, or that female cocaine users experience more problems than male
cocaine users.
The subjects in the study by Erickson and Murray consisted of a sample of 75 men and 36
women. The authors state that the subjects are a fairly representative sample of ‘typical’ adult users
who were neither in treatment nor in jail. Table 4.1 shows the lifetime frequency of cocaine use and the
gender of these subjects. Suppose we pick a person at random from this sample. What is the
probability that this person will be a male?
SOURCE: Patricia G. Erickson and Glenn F. Murray, “Sex Differences in Cocaine Use and Experiences: A Double Standard?”
American Journal of Drug and Alcohol Abuse, 15 (1989), 135-132 as printed in Biostatistics: A Foundation for Analysis in The Health
Sciences 6e by Wayne W. Daniel (1995)
Lifetime Frequency of Cocaine Use Male (M) Female (F) Total

1-19 times (A) 32 7 39
20-99 times (B) 18 20 38
100 + times (C) 25 9 34
Total 75 36 111
Solution:
We assume that male and female are mutually exclusive categories and the likelihood of
selecting any one person is equal to the likelihood of selecting any other person. We define the desired
probability as the number of subjects with the characteristic of interest (male) divided by the total
number of subjects. We may write the result in probability notation as follows:
P(M) = Number of males/Total number of subjects

= 75/111 = 0.6757
In some instances, the set of “all possible outcomes” may comprise a subset of the total group.
In other words, the size of the group of interest may be diminished by conditions not applicable to the
total group. When probabilities are computed with a subset of the total group as the denominator, the
result is a conditional probability.
We may think of the probability calculated in the previous example as an unconditional

probability since the size of the total group served as the denominator. No conditions were enforced to
restrict the size of the denominator. This probability can also be thought of as a marginal probability
since one of the marginal totals was used as the numerator. We may illustrate the concept of
conditional probability by referring again to Table 6-2.
Sample Problem:
Suppose we pick a subject randomly from the 111 subjects and find that he is a male (M). What
is the probability that he will be one who has used cocaine 100 times or more during his lifetime (C)?
Solution:
In this particular problem, the total number of subjects is not of concern anymore, since, with
the selection of a male, the females are eliminated. The desired probability can then be defined as
follows: Given that the selected subject is a (M), what is the probability that the subject used cocaine
100 times or more (C) during his lifetime? This is a conditional probability written as P(C|M) in which
the vertical line is read “given”. In this conditional probability, the 75 males become the denominator,
and 25, the number of males who have used cocaine 100 times or more during their lifetime, becomes
the numerator. Therefore, our desired probability is
P(C|M) = 25/75 = 0.3333
 The probability of two events in conjunction
 The probability that a subject picked at random from a group of subjects possesses two
characteristics at the same time
Sample Problem:
What is the probability that a person picked at random from the 111 subjects will be a male (M)
and a person who has used cocaine 100 times or more during his lifetime (C)? [Refer to Table 6-2]
Solution:
The probability we are looking for may be expressed in symbolic notation as P(M ∩ C) in which
the symbol ∩ is read either as “intersection” or “and”. The expression M ∩ C denotes the joint
occurrence of conditions M and C. The number of subjects who satisfy both of the desired conditions is
found in Table 6-2 at the intersection of the column labeled M and the row labeled C and is equivalent
to 25. Since the selection will be made from the total set of subjects, the denominator is 111. Therefore,
this joint probability may be written as
P(M∩C) = 25/111 = 0.2252
 When two independent events occur simultaneously, the combined probability of the two
outcomes is equal to the product of their individual probabilities of occurrence.
 The probability that two events A and B will both occur is equal to the probability of A
multiplied by the probability of B given that A has already occurred.
In addition, the multiplication rule of probability is a relationship where a joint probability may be
computed as the product of an appropriate marginal probability and an appropriate conditional
probability. This relationship can be illustrated by the following example:
Sample Problem:
We wish to compute the joint probability of male (M) and a lifetime frequency of cocaine use of 100
times or more (C). [Refer once again to Table 6-2]
Solution:
The probability we are looking for is P(M∩C). We have already calculated a marginal
probability, P(M) = 75/111 = 0.6757, and a conditional probability, P(C|M) = 25/75 = 0.3333.
Coincidentally, these are the appropriate marginal and conditional probabilities for calculating the
desired joint probability. We can now calculate P(M∩C) = P(M) P(C|M) = (0.6757)(0.3333) = 0.2252. Note
that as expected, this is the same result we calculated earlier for P(M∩C).
The multiplication rule can be stated in general terms as follows:
 For any two events A and B,
P(A∩B) = P(B)P(A|B), if P(B) ≠ 0
 For the same two events A and B, the multiplication rule may also be written as
P(A∩B) = P(A)P(B|A), if P(A) ≠ 0
Through algebraic manipulation, these equations can be used to find any one of the three
probabilities if the other two are known. For example, we may find the conditional probability P(A|B) by
dividing P(A∩B) by P(B). This relationship allows us to formally define conditional probability as follows:
The conditional probability of A given B is equal to the probability of A∩B

divided by the probability of B, provided the probability of B is not zero.
That is,
|
Addition Rule of Probability
The third property of probability previously stated that the probability of the occurrence of
either one or the other of two mutually exclusive events is equal to the sum of their individual
probabilities. For example, suppose that we pick a person at random from the 111 subjects
represented in Table 6-2. What is the probability that this person will be a male (M) or a female (F)? This
probability can be stated as P(M ∪ F) where the symbol ∪ is read either as “union” or “or”. Since the
two genders are mutually exclusive, P(M ∪ F) = P(M) + P(F) = (75/111) + (36/111) = 0.6757 + 0.3243 = 1.
What if two events are not mutually exclusive? This case is covered by what is known as the
addition rule, which may be stated as follows:
Given two events A and B, the probability that event A, or event B, or

both occur is equal to the probability that event A occurs, plus the
probability that event B occurs, minus the probability that the events
occur simultaneously.
The addition rule may be written as
P(A ∪ B) = P(A) + P(B) – P(A ∩ B)
Sample Problem:
If we select a person at random from the 111 subjects represented in Table 6-2, what is the
probability that this person will be a male (M) or will have used cocaine 100 times or more during his
lifetime (C) or both? [Refer to Table 6-2]
Solution:
The probability we are looking for is P(M ∪ C). Based on the addition rule, this probability may
be expressed as P(M ∪ C) = P(M) + P(C) – P(M ∩ C). We previously calculated that P(M) = 75/111 =
0.6757 and P(M ∩ C) = 25/111 = 0.2252. From the information in Table 6-2, we can calculate P(C) =
34/111 = 0.3063. Substituting these results into the equation for P(M ∪ C), we have P(M ∪ C) = 0.6757 +
0.3063 – 0.2252 = 0.7568.
Note that the 25 subjects who are both male and have used cocaine 100 times or more are
included in the 75 who are male as well as the 34 who have used cocaine 100 times or more. Since, in
calculating the probability, these 25 have been added in the numerator twice, they have to be
subtracted out once to cancel the effect of duplication or overlapping.
Suppose that we are told that event B has occurred, but this fact has no effect on the
probability of A. That is, the probability of event A is the same regardless of whether or not B occurs. In
this situation, P(A|B) = P(A). In cases such as these, we say that A and B are independent events.
• Two events are said to be independent, if the outcome of one event has no effect on the
occurrence of the other. If A and B are independent events,
P(A|B) = P(A) and P(B|A) = P(B)
• In this special case of independence, the multiplicative rule of probability may be written as
P(A∩B) = P(A)P(B)
Therefore, we can observe that if two events are independent, the probability of their joint
occurrence is equal to the probabilities of their individual occurrences.
Note that when two events with nonzero probabilities are independent, each of the following
statements is true:
P(A|B) = P(A), P(B|A) = P(B), P(A∩B) = P(A)P(B)
Two events are not independent unless all these statements are true.
Sample Problem:
In a certain high school class, consisting of 60 girls and 40 boys, it is observed that 24 girls and
16 boys wear eyeglasses. If a student is picked at random from this class, the probability that the
student wears eyeglasses P(E) is 40/100 or 0.4.
a. What is the probability that a student picked at random wears eyeglasses, given that the
student is a boy?
b. What is the probability of the joint occurrence of the events of wearing eyeglasses and being a
boy?
Solution:
a.
|
We may also demonstrate that the event of wearing eyeglasses, E, and not being a boy, B ,
are also independent as follows:
̅
|̅
̅
b. P(E∩B) = P(B)P(E|B) = P(B)P(E)

= (40/100)(40/100)
= 0.16
In the previous sections, we discussed a common method for finding probabilities: we

calculated the probability of an event by counting the number of possible ways an event can occur
and dividing the resulting number by the total number of equally likely elementary outcomes [P(E) =
m/N]. Because we used simple examples, such as the rolling of two dice which has 36 different
possibilities at most, we did not encounter any problem applying this formula.
However, there are situations when the number of ways that an event can occur is so large that
complete and exhaustive enumeration becomes tedious and impractical. The combinatorial methods
discussed in this section will facilitate the calculation of the numerator and denominator for the
probability of interest. Combinatorics, also called combinatorial mathematics, is the field of mathematics
involved with the problems of selection, arrangement, and operation within a finite or discrete system.
Its objective is to apply methods that allow you to count without counting. Thus, one of the
fundamental problems of combinatorics is to determine the number of possible configurations of
objects or events of a given type.
Again, let us consider the experiment where we roll dice. On any roll of a die, there are six
elementary outcomes. Suppose we roll the die three times so that each roll is independent of the other
rolls. We want to know how many ways we can roll a 4 or less on all three rolls of the die without
repeating a number.
Direct enumeration is difficult and tedious since there are a total of 6 x 6 x 6 = 216 possible
outcomes. Moreover, the number of successful outcomes may not be obvious. There is a shortcut
solution that becomes even more vital as the space of possible outcomes, and possible successful
outcomes, become even larger than in this example.
Thus far, our problem is not well defined. We must also specify whether or not the order of
distinct numbers maters. When order matters we are dealing with permutations. When order does not
matter we are dealing with combinations.
First, let us consider the case in which order is important; therefore, we will be determining the
number of permutations. If order matters, then the triple {4, 3, 2} is a successful outcome but differs
from the triple {4, 2, 3} because order matters. In fact, the triples {4, 3, 2}, {4, 2, 3}, {3, 4, 2}, {3, 2, 4}, {2,
4, 3}, and {2, 3, 4} are six distinct outcomes when order matters but count only as one outcome when
order does not matter since they all correspond to an outcome in which the three numbers 2, 3, and 4
each occur once.
A successful outcome is when a 4 or lower number is rolled. In this case, there are four objects
– the numbers 1, 2, 3, and 4 – to choose from because a choice of 5 or 6 on any trial leads to a failed
outcome. Because there are only three rolls of the die, and a successful roll requires a different number
on each trial, we want to determine the number of ways of selecting three objects out of four when
order matters. This type of selection is called the number of possible permutations for selecting three
objects out of four.
In general, let r be the number of objects to choose from and n the number of objects
available. The number of permutations of r objects chosen out of n can be determined using the
following formula:
The symbol “!” represents the function called the factorial. The notation n! (read as, n factorial)
means by definition the product:
For example, 3! = (3)(2)(1) = 6; while 4! = (4)(3)(2)(1) = 24; and 5! (5)(4)(3)(2)(1) = 120. Note that
0! exists and is equal to 1.
The problem of determining the number of ways we can roll a 4 or less on three rolls of a die
without repeating any number is solved as:
Now let us consider combinations. In combinations, only distinct subsets but not their order,
are considered. In the example of distinct outcomes of three rolls of the die where success means three
distinct numbers less than 5 without regard to order, the triplets {2, 3, 4}, {2, 4, 3}, {3, 2, 4}, {3, 4, 2},
{4, 3, 2}, and {4, 2, 3} differ only in order and not in the objects included.
Observe that for each different set of three distinct numbers, the common number of
permutations is always 6. For example, the 1, 2, and 3 contains the six triplets {1, 2, 3}, {1, 3, 2}, {2, 1, 3},
{2, 3, 1}, {3, 1, 2}, and {3, 2, 1}. Notice that the number six occurs because it is equal to P(3,3) = 3!/0! = 6.
The formula for the number of combination of r objects taken out of n is:
In our example of three rolls of the die resulting to three distinct numbers less than 5, the
number of combinations for choosing 3 objects out of 4 is:
These four distinct combinations are the sets consisting of (1) 1, 2, and 3; (2) 1, 2, and 4;
(3) 1, 3, and 4; and (4) 2, 3, and 4.
NAME:__________________________________________________________ DATE:______________
1. Why is it that science is not always certain? How do scientists cope with uncertainty? (3 points)
______________________________________________________________________________
______________________________________________________________________________
______________________________________________________________________________
______________________________________________________________________________
2. One hundred married women were asked to specify which type of birth control method they
preferred. The following table shows the 100 responses cross-classified by educational level of the
respondent. (7 points)
Birth Control Educational Level

Method High School (A) College (B) Graduate School (C) Total
S 15 8 7 30
T 3 7 20 30
V 5 5 15 25
W 10 3 2 15
Total 33 23 44 100
Specify the number of members of each of the following sets:

a. _______
b. ∪ _______
c. _______
d. ̅ _______
e. ̅ _______
f. _______
g. ̅̅̅̅̅̅̅̅̅ _______
Solve the following problems on the space provided. Write the pertinent probability expression(s) for
each one.
3. Laveist and Nuru-Jeter conducted a study to determine if doctor-patient race concordance was
associated with greater satisfaction with care. Toward the end, they collected a national sample of
African-American, Caucasian, Hispanic, and Asian-American respondents. The following table
classifies the race of the subjects as well as the race of their physician:
Patient’s Race
Physician’s Race Caucasian African-American Hispanic Asian-American Total
White 779 436 406 175 1796
African-American 14 162 15 5 196
Hispanic 19 17 128 2 166
Asian/Pacific-Islander 68 75 71 203 417
Other 30 55 56 4 145
Total 910 745 676 389 2720
Source: Thomas A. Laveist and Amani Nuru-Jeter, “Is Doctor-patient Race Corcondance Associated with Greater Satisfaction
with Care?” Journal of Health and Science Behavior, 43 (2002), 296-306 as printed in Biostatistics: A Foundation for Analysis in the
Health Sciences by Wayne W. Daniel (1995)
a. What is the probability that a randomly selected subject will have an Asian/Pacific-Islander
physician? (2 points)
b. What is the probability that an African-American subject will have an African-American

c. What is the probability that a randomly selected subject in the study will be Asian-
American and have an Asian/Pacific-Islander physician? (2 points)
d. What is the probability that a subject chosen at random will be Hispanic or have a Hispanic
e. Use the concept of complementary events to find the probability that a subject chosen at
random in the study does not have a white physician. (2 points)
4. The probability is 0.6 that a patient selected at random from the current residents of a certain
hospital will be a male. The probability that the patient will be a male who is in for surgery is 0.2. A
patient randomly selected from current residents is found to be a male; what is the probability that
the patient is in the hospital for surgery? (3 points)
5. The probability that a person selected at random from a population will exhibit the classic
symptom of a certain disease is 0.2, and the probability that a person selected at random has the
disease is 0.23. The probability that a person who has the symptom also has the disease is 0.18. A
person selected at random from the population does not have the symptom; what is the
probability that the person has the disease? (5 points)
6. Systematically enumerate all 24 permutations of rolling a die three times and observing a number
less than 5 on each separate occasion without any number repeating. Explain the method that you
used to systematically enumerate the permutations. (5 points)
(1) _______ (7) _______ (13) _______ (19) _______

(2) _______ (8) _______ (14) _______ (20) _______
(3) _______ (9) _______ (15) _______ (21) _______
(4) _______ (10) _______ (16) _______ (22) _______
(5) _______ (11) _______ (17) _______ (23) _______
(6) _______ (12) _______ (18) _______ (24) _______
Explain the method you used:

_____________________________________________________________________________________
_____________________________________________________________________________________
_____________________________________________________________________________________
_____________________________________________________________________________________
7. How many ways can a researcher select a distinct set of 10 participants from a pool of 50
volunteers? Show your solution. (3 points)
The term population refers to a collection of people or objects that share common observable
characteristics. For example, a population could be all the inhabitants of your city, all of the students
enrolled in a certain university, or all of the people who are afflicted by a particular disease (e.g., all
men diagnosed with prostate cancer during the last five years). Meanwhile, a sample is a subset of the
population. One approach of statistics, called inferential statistics relies upon the use of samples in
making inferences about populations. Sampling is the act of studying a subset of the population to
make inferences on the whole.
If a researcher wishes to gather information regarding a population through questioning or

testing, he/she has two basic options:
1. Every member of the population can be questioned or tested (i.e., a census); or

2. A sample can be drawn from the population of interest where only selected members of the
population are questioned or tested.
Understandably, choosing the first option means that the processes of contacting, questioning,
and gathering information from a large population, such as all of the households within Metro Manila,
is extremely expensive, challenging, and time-consuming. However, a properly designed sampling
method provides a reliable way of inferring information about a population without examining each of
its members or elements.
Another advantage of sampling is that it is more accurate than a census of the entire
population. The smaller sampling operation allows the application of more rigorous controls which
guarantees better accuracy. These rigorous controls give researchers the ability to reduce nonsampling
errors like interviewer bias and mistakes, nonresponse problems and attrition, questionnaire design
flaws, as well as data processing and analysis errors.
These nonsampling errors are partly reduced through pretesting which gives the researcher
information on whether or not his/her tools of data gathering are accurately and appropriately
designed by administering the test to a small subset of respondents. When conducting a census,
pretesting cannot be done without risking possible contamination of some of the respondents. In
addition, the scope and detail of information that can be asked in a sample is greater that in a census
due to cost and time limitations under which most researchers are operating. Administering a relatively
long and difficult survey to a sample is easier than administering a concise questionnaire to the entire
population. However, be mindful that not all samples are accurate or the appropriate vehicle for
obtaining information or testing a hypothesis about a population. In addition, there are cases where
sampling is the only possible method for destructive procedures. The following sections in this chapter
will discuss the advantages and disadvantages of various sampling procedures.
 Population – the totality of individuals or objects of interest → parameter is measured
 Target population – a subset of the population where representative information is desired
and to which inferences will be made
 Sampling population – a subset of the population where the sample is actually taken →
statistic is measured
o Sampling unit – the units which are chosen in selecting the sample, and may be
comprised of a non-overlapping collection of elements.
o Elementary unit/element – the sample where observations or data will be acquired
 Sampling frame – listing of all sampling units from which a sample will be drawn or the
collection of all the sampling units
 Sampling error – the difference between the population value (parameter) and the estimate of
this value based on the different samples (statistics)
Example:
 Population: Filipino college students

 Target population: state university students
 Sampling population: UP Manila students
 Elementary unit: BS Biology majors
 Sampling frame: OCS list of enrolled students
 Representative of the population

 Adequate sample size
 Practical and feasible
 Economical and efficient
 Non-Probability Sampling
o Any sampling scheme in which the probability of a population element being chosen is
unknown
o Data normally analyzed using non-parametric statistical tests (normal distribution not
assumed)
o Appropriate when the researcher has no intention of generalizing beyond the sample
o Advantages:
 More easily administered
 Samples tend to be less complicated and less time consuming
 May occasionally serve as the only possible means of getting a sample
(especially for “hidden populations”, e.g. MSM, drug users, etc.)
Sampling Methods
o Disadvantages:
 Does not allow the study’s findings to be generalized from the sample to the
population
 When discussing the results of a nonprobability sample, the researcher must
limit his/her findings to the persons or elements sampled
 More likely to produce biased results
 No defined rules to compute for estimates
 Cannot compute the reliability of estimates
 Probability Sampling
o Any sampling scheme wherein each population element has a known non-zero chance
of being included in the sample
o Analyzed using parametric statistical tests (normal distribution assumed)
o Uses random selection procedures to ensure that each unit of the sample is chosen on
the basis of chance where all units of the study population should have an equal or at
least a known chance of being included in the sample.
o Requires that a listing of all study units (sampling frame) exists or can be compiled.
o Advantages:
 Probability samples are the only type of samples where the results can be
generalized from the sample to the population
 Allows the researcher to calculate the precision of the estimate as well as the
sampling error
o Disadvantages
 More difficult and costly to conduct
 Judgmental/Purposive – a representative sample of the population is selected based on a

researcher’s “expert” judgment. Prior knowledge and research skill are used in choosing the
respondents or elements to be sampled.
o selection of patients in a clinical trial by a medical specialist
o choice of participants based on a pre-test questionnaire or focus-group discussion
(FGD)
 Haphazard/Accidental – also known as convenience sampling, is a sampling design where the

samples are selected by an arbitrary method that is easy to carry out. Convenience samples
have been used when it is very difficult or impossible to draw a random sample. Studies based
on convenience samples have results that are descriptive and may be used to recommend
future research but should not be used to draw inferences about the population under study.
o friends as a sample of college students
o ambush interviews of random people in the area
o households of relatives or friends as sampling sites
 Quota – dividing the population into predetermined classes then obtaining haphazard samples
of a fixed size (quota) within each class. As each class fills or reaches its quota, additional
respondents that would have fallen into these classes are rejected or excluded from the results.
o Interviewing the first 10 males and females in a public restroom regarding shampoo
preference
o Obtaining the RH bill opinion of 20 people per region in a municipality
o A researcher desires to obtain a certain number or respondents from different income
categories. Generally, researchers have no knowledge of the incomes of the people
they are sampling until they ask about income. Thus, the researcher is unable to
subdivide the population from which the sample is drawn into mutually exclusive
income categories prior to drawing the sample. In this type of sample, bias can be
introduced when the respondents who are rejected (because the class to which they
belong has already reached its quota) differ from those who are used.
 Simple Random Sampling – the most basic type of sampling design wherein every element in
the population has an equal chance of being included in the sample.
Simple random sampling is carried out by following these steps:

 Prepare an exhaustive list (sampling frame) of all members of the population of
interest.
 Decide on the size of the sample
 Select the necessary number of sampling units, using a “lottery” method, a table of
random numbers, or the RAN function of a calculator.
o Advantages:
 Simple design, easy to analyze
 Random sampling guarantees unbiased estimates of population parameters.
Unbiased means that the average of the sample estimates over all possible
samples is equal to the population parameter.
o Disadvantages:
 Not cost efficient because elementary units may be too widespread
 Requires a sampling frame or listing of all elementary units of the population
which might be costly and tedious to prepare
 Does not guarantee balance in any particular sample drawn at random. Even
though the probability is very small, it is possible that nonrepresentative
samples can be drawn from the population. For example, suppose a catheter
ablation treatment is known to have a 95% chance of success. Thus, we can
expect only about one failure in a sample size of 20. However, even though it is
highly unlikely, it is possible that we could select a random sample of 20
individuals with the result that all 20 individuals have failed ablation
procedures.
Sampling Methods
 Systematic Sampling – in this method, the researcher selects samples at regular intervals (every
1st, 2nd, 3rd,… or kth). The sampling interval is computed as where is the sampling
interval, is the population size, and is the desired sample size.
o Example:
 Sampling interval = 20/10 = 2 or every 2nd unit
 Randomly draw a number from 1 to 20 = 3 for instance
 Samples are selected every 2nd unit (example: households 3, 5, 7… and so on)
until is reached.
o Advantages:
 Tend to be easier to draw and execute compared to simple random sampling
since the researcher does not have to jump forward and backward through the
sampling frame to draw the members to be sampled.
 Compared to simple random sampling, a systematic sample may spread the
members selected for measurement more evenly across the entire population.
Thus, in some cases, systematic sampling may have better representativeness
and be more precise.
 It can allow the researcher to draw a probability sample without complete prior
knowledge of the sampling frame.
o Disadvantages:
 Can lead to difficulties when the variable of interest is periodic (with period n).
For example, when conducting a sample of financial records, or other objects
that follow a calendar schedule, setting the sampling interval to 7 would mean
that all observations would fall on the same day of the week. This introduces
bias into the sample since an inappropriate interval was chosen.
 Periodic or list effect – a foregoing source of sampling error. For example, if we
used a very long list such as a telephone directory for our sampling frame and
needed to sample only a few names using a short sampling interval, there is a
possibility that by accident, we select a sample from a portion of the list
wherein a certain ethnic group is concentrated. Thus, the representativeness of
the sample would not be very good. If the characteristics we are interested in
varied considerably by ethnic group (such as average lifespan, allelic
frequencies, genetic polymorphisms, etc.), our estimate of the population
parameter could be very biased.
 Stratified Random Sampling – a modification of simple random sampling that is used when we
want to guarantee that each stratum (subgroup) constitutes an appropriate portion or
representation in the sample. It involves categorizing the members of the population into
mutually exclusive and collectively exhaustive groups. Simple random sampling is then
independently carried out for each group.
Stratified random sampling is carried out by following these steps:

 Define subgroups of strata
 For the subgroup, we select a simple random sample of size
 Follow this procedure for each subgroup.
 The total sample size is then ∑ . The notation Σ stands for the summation of the
individual ’s. For example, if three groups, then ∑ . In general,
we have a total sample size “ ” in mind.
o Advantages:
 Can be used to improve the accuracy of the sample estimates when there is
prior knowledge that the variability in the data is not constant across the
subgroups.
 It can enable the researcher to determine the desired level of sampling
precision for each stratum.
 It can be demonstrated by statistical theory that in many situations, stratified
random sampling produces an unbiased estimate of the population mean with
better precision compared to simple random sampling with the same total
sample size n. Choosing large values of for the subgroups with the largest
variability and small values for the subgroups with the least variability
improves the precision of the estimate.
o Disadvantage:
 May require a very large if reliable estimates for each stratum are desired
 Cluster Sampling – a method of sampling in which the element selected is a group (rather than
an individual), called a cluster. For instance, the clusters could be city blocks. Cluster sampling is
similar to stratified sampling because the population to be sampled is subdivided into mutually
exclusive groups. However, in cluster sampling the groups are defined so as to maintain the
heterogeneity of the population. The goal of the researcher is to establish clusters that are
representative of the population as a whole, although in practice this may be difficult to
achieve. Once the clusters are established, a simple random sample of the clusters is drawn
and the members of the chosen clusters are sampled.
o One-stage cluster sampling – all of the elements (members) of the clusters selected are
sampled
o Two-stage cluster sampling – a random sample of the elements of each selected
cluster is drawn.
o Advantages:
 Can be employed in the absence of a sampling frame. For example, a
researcher is interested to measure the age distribution of people residing in
Metro Manila. It is easier to compile a list of residential addresses in Metro
Manila than to compile a list of every person residing in Metro Manila. In this
case, each address would represent a cluster of elements (people) to be
sampled.
 Since the clusters are randomly selected, the samples can be representative of
the population and unbiased estimates of the population total or mean value
for a particular parameter can be obtained.
 Reduced cost of data collection
o Disadvantage:
 Sometimes, there is loss of precision for the estimate relative to the simple
random sampling if the heterogeneity within the clusters is not similar to the
heterogeneity within the population.
Sampling Methods
 Multistage Sampling – a procedure carried out in phases involving a combination of

probability sampling designs. The population is divided into sets of primary or first stage
sampling units and then a random sample of secondary stage units is obtained from each of
the selected units in the first stage. This type of sampling method is appropriate for very large
and diverse populations where the selection of elementary units may be done in two or more
stages.
Example: Nationwide survey of all 15 regions (stratified)
o 1 province/region – primary sampling unit (simple random sampling)
o 1 urban & 1 rural barangay – secondary sampling unit (stratified random sampling)
o 1 cluster of 35 households – tertiary sampling unit (cluster sampling)
o Choose the households per cluster – elementary unit (systematic sampling)
Sampling design: 4-stage, stratified, systematic, cluster, simple random sampling.
o Advantages:
 Cost-efficient
 Samples are easier to select if a sampling frame is available
o Disadvantages:
 May require complicated analyses
 Sample size must be large enough to obtain representative estimates of the
parameters
NAME:__________________________________________________________ DATE:______________
1. How can bias affect a sample design? Explain by using the terms selection bias, response bias, and
periodic effects (5 points)
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
2. How is sampling with replacement different from sampling without replacement? (2 points)
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
3. Why would a convenience sample of college students on vacation at Boracay not be representative
of students at a particular college or university? (3 points)
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
_________________________________________________________________________________
4. Identify the possible sampling frame that could be used for each of the following surveys
(2 points each)
a. A survey among the alumni of the BS Biology program of UP Manila on the relevance of the BS
Biology curriculum in their present work
______________________________________________________________________________
b. A survey to determine the percentage of CAS students who intend to go to medical school.
______________________________________________________________________________
c. A survey to find out the distribution of mammals at the Manila Zoo.
______________________________________________________________________________
5. On the next page is a map of a certain community with 20 houses. Using the following random
numbers,
1402 1185 0920 3610 0879 1304
Draw or obtain:
a. A simple random sample of 6 households (write the household numbers that you were able to
draw) (5 points).
______________________________________________________________________________
b. A systematic sample with (write the household numbers that you were able to draw)
(5 points).
______________________________________________________________________________
c. Based on the simple random sample you acquired in (a), compute for the mean age of the
household heads (2 points).
Place your computations here:
1. How does your computed mean age compare with that of your classmate’s result?
(Indicate your classmate’s name and her/his computed mean) Is it larger, lower or of the
same value? (2 points)
___________________________________________________________________________
___________________________________________________________________________
2. How does your computed mean age compare with the mean age of the household heads
for the entire population? Is it larger, lower or of the same value? (2 points)
___________________________________________________________________________
___________________________________________________________________________
3. To what do you attribute the difference between your computed sample mean, your
classmate’ result and that of the total population? (3 points)
___________________________________________________________________________
___________________________________________________________________________
___________________________________________________________________________
Probability Theory is an essential mathematical concept that serves as a foundation of
statistics. Statistical tests which are used to make inferences related to a set of hypotheses are based
on probabilistic models or distributions. A probability distribution is a function, that is, a graphical
relationship between all possible outcomes of an event (x, domain) and the probabilities of occurrence
of each outcome (y, codomain).
 The “event” can be considered in the light of a quantitative variable (e.g. count, weight)
 The “possible outcomes of an event” as any possible value of the quantitative variable; also
called a “random variable”
 The “probabilities of occurrence” corresponds to the frequency of occurrence of the value
 If all “possible outcomes of an event” are tabulated in a frequency distribution table (where
frequency/ relative frequency is the probability of a specific outcome), a corresponding
histogram or frequency polygon serves as the graphical representation of the probability
distribution
There are two general kinds of probability distributions: probability mass functions and
probability density functions. Without going into mathematical rules and set notations, for the
purpose of this general biostatistics course:
1. Probability mass function (PMF)

 probability distribution of values of quantitative discrete random variables; a histogram
 especially useful in genetics, e.g. determining the probability of a combination of a
certain number of boys and girls in a 3-child family
o this specific PMF is a binomial distribution (i.e. only two possible outcomes: girl
and boy)
o given that the probability of having a boy is 50% and a girl is also 50%, then to
calculate the different combinations for a 3-child family, binomial expansion of
the equation can provide a guide in calculating
the probabilities of the combinations:
– probability of 3 boys = =0.125
– probability of 2 boys =
– probability of 1 boy =
– probability of 0 boys =
Figure 7-1. Probability Mass Function for Number
of Boys in a 3-Child Family
2. Probability density function (PDF)

 probability distribution of values of quantitative continuous random variables; a
curve or frequency polygon with small class widths so as to appear smooth
 the probability of values of a quantitative continuous random variable correspond
to the areas under a curve and where the total area under the curve is equal to 1 or
the integral: ∫
 A specific type of PDF or curve is the normal curve or the normal distribution
The normal distribution is the most important probability distribution in statistics. The idea of
the normal curve was first introduced by the French mathematician Abraham De Moivre in the early
1700s whose probability theory focused on the binomial distribution and its applications on his side
job as a gambling consultant (i.e. coin flips in most probability theory problems).
Pierre-Simon de Laplace rediscovered the normal curve in the 1780s through his central limit
theorem. An important concept of Laplace’s central limit theorem states that for a random variable of
interest, when a large enough sample size (n) is acquired from a population (N), the mean of all
samples ( ̅ from the population approximates the population mean ( ̅ . The population
mean, , is at the center of the symmetrical distribution.
In the 1830s a German mathematics and astronomy professor, Carl Friedrich Gauss, determined
that the distribution of accidental errors in astronomical measurements also followed a normal curve.
His Gaussian Law of Errors in the Theory of Additive Functions was studied and applied to other
situations that the normal distribution was occasionally called the Gaussian distribution. However, in
France, the normal distribution is mostly called the Laplacian distribution in reference to de Laplace.
1. Bell-shaped curve that is symmetrical at the mean and extends from (asymptotic
horizontally)
2. The measures of central tendency (mean, median and mode) are equal
Normal Distribution
3. ∫ , that is, the total area under the curve is equal to 1

4. Determined by the mean, , and the standard deviation, , so that every specific pair of mean
and standard deviation corresponds to a specific normal curve
A more specific kind of normal distribution is the standard normal distribution, wherein the
µ=0 and σ=1. A standard normal curve is obtained by transforming any value of the random variable
(x) into a z-score and is calculated as:
A table that provides the results of all the integrations of ∫ that we might be interested
in is provided at the end of this module. In the body of the table are found the areas under the curve
between and the values of z shown in the left-most column of the table. The shaded area of Figure
7-6 represents the area listed in the table being between and z0 where z0 is the specified value of z.
Normal Distribution
Example:
Given the standard normal distribution, find the area under the curve, above the z-axis between
z= and z = 2
Solution:
It will be helpful to draw a graph of the standard normal distribution and shade the desired
area as in Figure 7-7. If we locate z = 2 in Table C and read the corresponding entry in the body of the
table, we find the desired area to be .9772. We can interpret this value in the following ways:
• 97.72% is the probability that a z picked at random from a population of z’s will have a value
between and 2.
• 97.72% is the relative frequency of occurrence (or proportion) of values of z between - and
2, or we may say that 97.72% of the z’s have a value between and 2.
Alternatively, instead of looking up the desired areas on the table, you may use Microsoft Excel’s
NORM.S.DIST function. The syntax for the function is =NORM.S.DIST(z,cumulative) where z is the value
for which you want the distribution and cumulative is a logical value that determines the form of the
function. If cumulative is TRUE, NORM.S.DIST returns the cumulative distribution function or the
probability that the value will be less than or equal to z. If cumulative is FALSE, NORM.S.DIST returns
the probability mass function or the probability that the value z will occur. We will mostly use TRUE as
the argument for our purposes since we desire to find the area to the left of z.
Therefore, from the example above, the correct way to calculate the area between z = and
z = 2 is to input =NORM.S.DIST(2,TRUE) into a cell. The output 0.97725 will be generated.
Note that for versions of Excel earlier than 2010, the analogous function is NORMSDIST whose
syntax is =NORMSDIST(z). Notice that the argument for cumulative is not required since this function
always returns the cumulative distribution function. When using Excel 2007 or earlier, the correct way
to input the syntax into a cell is =NORMSDIST(2). This will give the same output of 0.97725
Recall that the NORM.S.DIST function returns the area from to the specified value of z,
similar to the tabulated values. How should you write your syntax if you want to find areas from z to
instead? How about when you want to find the probability that a z picked at random from the
population of z’s will have a value between two z’s such as P(-2.55 < z < 2.55)?
Another useful function is NORM.S.INV. The NORM.S.INV function works in reverse as the NORM.S.DIST
function (i.e., it returns the inverse of the standard normal cumulative distribution). Its syntax is
=NORM.S.INV(probability) where the argument probability refers to a probability corresponding to the
normal distribution (i.e., an area under the curve). NORM.S.INV considers the area from to the
specified probability to calculate the z value. The analogous function for Excel versions 2007 and earlier
is NORMSINV with syntax =NORMSINV(probability).
Example:
As part of a study of Alzheimer’s disease, Dusheiko reported data that are compatible with the
hypothesis that brain weights of victims of the disease are normally distributed. From the reported
data, we may compute a mean of 1076.80 grams and an SD of 105.76 grams. If we assume that these
results are applicable to all victims of Alzheimer’s disease, find the probability that a randomly selected
victim of the disease will have a brain that weighs less than 800 grams.
SOURCE: S.D. Dusheiko, “Some Questions Concerning the Pathological Anatomy of Alzheimer’s Disease,” Soviet Neurological
Psychiatry, 7 (1974), 56-64 as printed in Wayne Daniel, Biostatistics: A Foundation for Analysis in the Health Sciences” 6e (1995).
Solution:
1. Draw a graph of the distribution and shade the area corresponding to the probability of interest.
2. If this were a standard normal distribution, with a mean of 0 and a standard deviation of 1, we
could look up the probability in the tabulated values of z and find it with ease. Fortunately, it is
possible for any normal distribution to be transformed to the standard normal with little effort.
What we do is transform all values of X to corresponding values of z. This means that the mean of X
must become 0, the mean of z. In this particular problem, we must determine what value of z
Normal Distribution
corresponds to an x of 800. This can be accomplished by using this formula which was presented
earlier:
Therefore, P(x < 800) = P(z < -2.62)
Alternatively, you can use Excel’s STANDARDIZE function to transform values of x into z. The
syntax for this function is =STANDARDIZE(x,mean,standard_dev) where x is the value you want to
normalize, mean is the arithmetic mean of the distribution, and standard_dev is the standard deviation
of the distribution. In this example, the proper way to input the syntax into a cell is
=STANDARDIZE(800,1076.8,105.76). This will return a result of -2.61725.
3. Find the area to the left of z = -2.62 by looking at the tabulated values of the standard normal
distribution of by using the NORM.S.DIST function.
However, there exists a simpler, one-step solution to solve for the desired probability when the
given values are not normalized such as the case in this problem. This is through the use of Excel
2010’s NORM.DIST function which returns the normal distribution for the specified mean and standard
deviation. Its syntax is is =NORM.DIST(x,mean,standard_dev,cumulative) where all four arguments are
as defined earlier. For this particular problem, the function should be used as
=NORM.DIST(800,1076.8,105.76,TRUE). The result of using this function is 0.004432. The analogous
function to NORM.DIST for Excel 2007 and earlier is NORMDIST which uses the same combination and
sequence of arguments.
Simply put, using =NORM.DIST(x,mean,standard_dev,cumulative) is similar to using

=NORM.S.DIST(STANDARDIZE(800,1076.8,105.76),TRUE).
Interpretation: The probability that a randomly selected victim of Alzheimer’s disease will have
a brain that weighs less than 800 grams is 0.44%
 In inferential statistics, statistical tests used (e.g. t-test, analysis of variance) assume a normal
distribution. Statistical tests that assume a normal distribution are also termed as parametric
tests.
 Basis for estimation of parameters and calculation of the probability of occurrence of random
variables based on areas under the normal curve in a sampling distribution
1. To determine the proportion (%) of values or proportion of the population that belong
to certain categories (given the μ & σ)
2. To estimate the probability that a member of the population will belong to a category
(given the μ & σ)
3. To determine the bounding variables/ values (i.e. x1 & x2) given the proportion or
probability
An important goal of data analysis is to distinguish between features of the data that reflect
real facts and features that may reflect only chance effects. Because it is impossible to actually acquire
values of a random variable from all members of a population so as to measure the population mean,
μ, and standard deviation, σ. It is essential to acquire values of a random variable from a random
sample of a population in order to get the sample mean, ̅ , and sample standard deviation, s.
 Parameter versus Statistic

o Parameter = numerical constant acquired by observing the population (e.g. μ, σ)
o Statistic = numerical variable acquired by observing a sample from the population
(e.g. ̅ )
 Sampling Distribution: probability distribution of a statistic that would be obtained under
repeated sampling
o Sampling Distribution of the Mean: probability distribution of sample means ( ̅
obtained from all possible samples of size n.
o Sampling Distribution of Proportions: probability distribution of sample proportions (p)
obtained from all possible samples of size n. (note: Population Proportion = P and
Sample Proportion = p)
1. Sampling Distribution of the Mean

 Properties:
o It is approximately normally distributed
o μ x = μ, mean of all sample means is equal to the population mean
o s (a.k.a. standard error of the mean)
√
 Applications:
o Determining the probability of occurrence of x given a pre-specified magnitude (or
bounding values under the normal curve) from the population
o Estimation of the μ
o Hypothesis testing about μ
 Sample Problem: If the distribution of systolic blood pressure (SBP) of non-hypertensive
chimpanzees (Pan sp.) has a mean of 110 mmHg & standard deviation of 12 mmHg:
a) What is the probability that a sample of 49 chimpanzees will yield a mean that is:
(1) Greater than or equal to 112 mmHg?
(2) Between 112 and 115 mmHg?
b) Within what range will the middle 95% of the sample means fall?
Given: mmHg n = 49 chimpanzees

= 12 mmHg ̅ = 112 mmHg
Required: a1) Probability of a mean that is ≥ 112mmHg

a2) Probability of a mean that is between 112mmHg and 115mmHg
b) The bounding values ̅ ̅ for the middle 95% of the sample means
Normal Distribution
Solution:
 Step 1 – sketch the distribution as a guide; shade areas under the curve that are required
or given
 Step 2 (for a1 and a2) – compute for the z-deviate
̅ ̅
a1) a2)
⁄√ ⁄ ⁄√ ⁄
√ √
̅
⁄√ ⁄
√
Alternatively, you may take advantage of Excel’s STANDARDIZE function and substitute the
value of (the standard error) for the standard_dev argument. Take note however that standard
error is not the same as standard deviation. We will discuss this in a further lesson. We are only
exploiting the algorithm employed by Excel to easily compute for the z-deviate of the sampling
distribution.
Example for a1: =STANDARDIZE(112,110,12/SQRT(49))
The result is 1.166667
 Step 2 (for b) – Since the area under the curve is = 1 or 100%, then the total remaining area
under the tail regions is 0.05 or 5% where each tail has an area of 0.025 or 2.5%
 Step 3 (for a1 and a2) – refer to a z-table (or table for the areas under the standard normal
curve) that may either provide the probabilities for the tail region or middle region of the
curve. Assuming that the z-table used for this problem provides areas under the tail based
on the z deviate calculated:
a1) 1.17  0.1210 or 12.10%
Alternatively, you may also exploit the NORM.DIST function for a one-step solution. The
appropriate syntax would be =1‒NORM.DIST(112,110,12/SQRT(49),TRUE). Once again, we have
substituted the value of the standard error into the standard_dev argument. We began the
expression with “1‒” since we want to get the area to the right of the specified value of x.
Interpretation: The probability of a mean that is ≥ 112 mmHg is 12.10%
a2) z1 = 1.17  0.1210 or 12.10% and z2 = 2.92  0.0018 or 0.18% (since these probabilities
are tail end probabilities and the middle is need then subtract these from 100%)
̅
Interpretation: The probability of a mean that is from 112 mmHg to 115 mmHg is 11.92%
 Step 3 (for b) – Also based on a z-table that provides areas under the tail region of a
standard normal distribution; look for the area that is 0.025. This time, the corresponding
z-deviate for the area 0.025 must be acquired, which in this case is 1.96. Based on the
symmetrical nature of the standard normal distribution:
Normal Distribution
In order to get the appropriate ̅ and ̅ , simply use the given z-deviates:
̅ ̅
  ̅ ( ⁄ )
⁄
√
⁄
√ √
= -3.36 + 110
= 106.64 mmHg
̅ ̅
  ̅ ( ⁄ )
⁄
√
⁄
√ √
= 3.36 + 110
= 113.36 mmHg
Interpretation: The middle 95% of the sample means fall between 106.64 mmHg to 113.36 mmHg
2. Sampling Distribution of Proportions

 Properties (in a large sample size, n, both n(P) and n(1-P) must be ≥ 5):
o It is approximately normally distributed
o μp = P, sample mean proportion is equal to the population proportion
o √
 Applications:
o Determining the probability of occurrence of p given a pre-specified magnitude from
the population
o Estimation of the P
o Hypothesis testing about P
 In calculating the z-deviate, the following formula is used:
√
 Sample Problem: If the cure rate for a new intestinal parasite drug for dogs is 80%, what is the
probability that up to 70% of 50 dogs in an animal shelter given the drug will be cured?
o Note: nP and n(1-P) must both be greater than or equal to 5
o nP = (0.80)(50) = 40  n(1-P) = (0.20)(50) = 10 
Given: P = 80% p = 70% n = 50 dogs
Required: Probability that up to 70% of 50 dogs will be cured by the parasite drug
Solution: Sketch the graph, transform to the standard normal distribution (z deviate) and get the
probability of the computed z from the z table.
√ √
For a z table with tail areas, z= -1.77

 0.0384  3.84%
Interpretation: The probability that up to 70% of 50 dogs in a shelter will be cured by the drug.
NAME:__________________________________________________________ DATE:______________
1. Given the values of z, find the area under the normal curve that lies between. (2.5 points each)
a. z = -1.85 and z = 1.85 b. z = -0.76 and z = 1.13
________________________________ ________________________________
________________________________ ________________________________
________________________________ ________________________________
a. Answer: _____________________ b. Answer: _______________________
2. Determine the area under the normal curve towards the tail end from the z deviates below. (2.5
points each)
a. z = -2.41 b. z = 1.73
________________________________ ________________________________
________________________________ ________________________________
________________________________ ________________________________
Answer: _____________________ b. Answer: _______________________
3. If the age of at onset of a hypothetical disease Y has a normal distribution with a mean of 55
years old and a standard deviation of 10 years. What is the probability that that onset of
disease Y for a person is before 40 years old? (5 points)
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
4. If the nursing board exam passing rate of a new nursing school is 75%, what is the likelihood
that half of 350 nursing students who will take the boards this year will pass? (5 points)
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
5. If the mean serum cholesterol level of Manila residents is 217mg/dL and the variance is 750,
calculate the probability that a random person will have a cholesterol value that is < 150
mg/dL. (5 points)
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
________________________________________
6. Use the NORM.S.INV function to solve for z1 given the following probabilities (write all pertinent
________________________________________
syntax as well as probability notations):
Example:
P(z ≤ z1) = 0.1234

Syntax: =NORM.S.INV(0.1234)
z1 = -1.16
Probability Notation(s):
a. P(z ≤ z1) = 0.0077 (2 points)

Syntax: _____________________________________________________________________
z1: ____________
_____________________________________________________________________
b. P(z > z1) = 0.5250 (2 points)

Syntax: _____________________________________________________________________
z1: ____________
_____________________________________________________________________
c. P(-2.75 ≤ z ≤ z1 ) = 0.9885 (4 points)
Syntax: _____________________________________________________________________
z1: ____________
___________________________________________________________________________
d. P(-z1 ≤ z ≤ z1 ) = 0.6263 (5 points)

Syntax: _____________________________________________________________________
z1: ____________
___________________________________________________________________________
e. P(z1 ≤ z ≤ 3.60) = 0.1071 (4 points)

Syntax: _____________________________________________________________________
z1: ____________
___________________________________________________________________________
7. Use all Excel functions at your disposal to solve the following problem:
Suppose the average length of stay in a chronic disease hospital of a certain type of patient is
45 days with a standard deviation of 10. If it is reasonable to assume an approximately normal
distribution of lengths of stay, find the probability that a randomly selected patient from this
group will have a length of stay:
a. Greater than 30 days (2 points) Answer: __________

b. Between 20 and 45 days (3 points) Answer: __________
c. Less than 10 days (1 points) Answer: __________
d. Greater than 70 days (2 points) Answer: __________
On occasion, we calculate descriptive measures to describe a particular set of data. At other
times, when the data represent a sample from a larger population, we might be interested in drawing
inferences from the data. A large collection of statistical techniques is available to allow us to perform
this and constitutes the branch of statistics called inferential statistics. Statistical inference is the process
by which we reach a conclusion about a population based on the information present in a sample
drawn from that population.
When inferences are drawn from normally distributed data, conclusions are based on the
relationships of the standard deviation and the mean to the normal curve. When the graph of a
frequency distribution seems normal, we can assume that the population of data where our sample
originated is normally distributed as well given that the sample size is sufficiently large. In most
practical situations, a sample size of 30 is satisfactory. We then assume that if we possessed all possible
observations from that population of data, we would discover that 68.3%, 95.5%, and 99.7% of the
population would lie between the mean and ±1, 2, and 3 standard deviations. In addition, we assume
that 95% of the population would like between the mean and ±1.96 standard deviations.
SOURCE: Principles of Epidemiology, Centers for Disease Control and Prevention (1992)
In this chapter, we will discuss estimation, the first of the two general areas of statistical
inference, the other being hypothesis testing. Estimation is the process by which a statistic computed
for a random sample is used to approximate the corresponding parameter.
Here are some examples of situations when estimation is useful:
 A biochemist is interested in estimating the average concentration of a certain protein in a

patient population.
 A geneticist is interested in estimating the allele frequencies of certain genes in a target
population.
The interests above are to estimate a certain numerical quantity associated with a particular
population.
Random Sample
I am 95% confident
Population that µ is between
(mean, µ, is Mean 40 and 60
̅ = 50
𝒙
unknown)
Sample
The rationale behind estimation in the health sciences field rests on the assumption that
professionals in this field are interested in parameters such as means and populations of various
populations. If this is the case, there are two good reasons why one must rely on estimation to gather
information about these parameters.
1. Many populations of interest, although finite, are so large that a 100% examination would be
prohibitive from the standpoint of cost.
2. Populations that are infinite are incapable of complete examination
 Point estimate – a single numerical value used to estimate the corresponding population
parameter.
 Interval estimate – consists of two numerical values defining a range of values that, with a
specified degree of confidence, includes the parameter being estimated.
The estimator is the sample statistic used to make inferences about an unknown parameter.
Example:
∑
̅
where the sample mean ̅ is an estimator of the population mean µ.
Summarizing Figure Population Parameter Sample Statistic (Estimator)

Mean ̅
Variance
Proportion
Difference between
̅ ̅
two means
Difference between
to proportions
1. Unbiased – its expected value is equal to the parameter being estimated (regardless of sample
size n). It should neither neither consistently overestimate nor underestimate the parameter.
The sample mean and sample variance have this property and are therefore unbiased.
However, the sample standard deviation is not.
2. Precise – it is repeatable because its standard error is small; should not vary too much from
sample to sample
3. Consistent – its deviation from the parameter being estimated decreases as sample size
increases.
 Simply the mean computed from a sample


Example:
Prostate specific antigen (PSA) is a protein produced by the cells from the prostate. The blood
concentration of PSA is often used as a biomarker of prostate cancer. Results under 4 ng/mL are usually
considered normal. The higher the PSA level, the more likely a patient has prostate cancer. Because of
this relationship, postproctectomy PSA has also been used to measure the success of the operation.
0.2 0.1 0.0 0.0 0.1 0.1 0.1 0.1 0.1 0.0
0.4 0.0 0.0 0.2 0.2 0.1 2.7 0.1 0.0 0.2
1.3 0.0 0.2 0.0 0.1 0.3 0.1 0.0 0.0 0.1
The point estimate for PSA levels in patients measured 6 months after proctectomy is
0.2267 ng/mL.
 Simply the difference between two sample means, ̅ ̅

 ̅ ̅
Example:
A study aims to determine the relationship between salt intake and blood pressure of persons aged 15
years and older. The mean SBP of 20 subjects with low salt diet was 120 mmHg while the mean SBP of
those with high salt diet was 138 mmHg.
̅ ̅
The point estimate for the difference in mean systolic blood pressure between patients with
high and low salt diet is 18 mmHg
 simply its corresponding statistic,

 the proportion of the sample possessing the characteristic of interest
Example:
In a prospective cohort study, troponin T levels were obtained for a sample of 801 patients
who had been hospitalized with acute myocardial ischemia. Whether or not the patients died within 30
days was then obtained and is summarized in the following table. What is an overall estimate of dying
within 30 days for the patients with high troponin T levels?
Troponin T Level
Status >0.1 ng/mL ≤0.1 ng/mL Total
Alive 255 492 747
Dead 34 20 54
Total 289 512 801
The proportion of dying within 30 days for patients with high troponin T levels is 0.12.
 Simply the difference between two sample proportions,


Example:
From the previous table, what is an estimate of the difference in proportion of deaths between
the patients with high versus low troponin T levels?
̂ ̂ ⁄ ⁄
The point estimate for the difference in proportion of deaths between patients with high
versus low troponin T levels is 0.08.
An interval estimate gives a range of values, taking into consideration the variation in sample
statistics form sample to sample.
 It conveys information about the probable magnitude of the population parameter.

 It is stated in levels of confidence. It can never be 100% confident.
 Express the probability that a prescribed interval will contain the true parameter.
 Estimator – point estimate of the population parameter in the center
 Reliability or confidence coefficient – the degree of certainty that the population parameter
being estimated is within the computed confidence interval
 Standard error of the mean – or simply standard error, it is the standard deviation of the
sampling distribution of the statistic
The standard deviation and standard error of the mean should not be confused. The
standard deviation is a measure of the variability or dispersion of a set of observations about
the mean. Meanwhile, the standard error of the mean is a measure of the variability or
dispersion of sample means about the true population mean.
The general form of an interval estimate is expressed as:
Notice that the center of the interval estimate is the point estimate of parameter of interest.
The quantity obtained by multiplying the reliability factor by the standard error of the sample
statistic is called the precision of the estimate or the margin of error.
Reliability or Confidence Reliability Factor or

Coefficient z-deviate
90% 1.645
95% 1.960
99% 2.576
Take note that researchers may use any confidence coefficient they wish but the most
frequently used values are presented above.
Formula for estimating the standard error of the mean:
√
Note that the standard error of the mean is affected by two components, the standard
deviation and the sample size. The greater the observations vary about the mean, the greater the
uncertainty of the mean, and the greater the standard error of the mean. The larger the sample size,
the more confidence we have in the mean, and the standard error of the mean becomes less.
Example:
Occupational health researchers measured the weights of a random sample of 120 male workers at an
industrial factory, factory F. The mean weight was 92.546 kg, with a standard deviation of 5.265 kg.
Calculate the standard error of the mean for the height of workers at factory F.
√ √
Suppose the reliability coefficient is 95% with a corresponding reliability factor of 1.96. We can
say that in repeated sampling, approximately 95% of the intervals constructed will include the
population parameter being estimated. This interpretation is based on the probability of occurrence of
different values of the sampling statistic. This interpretation can be generalized if we designate the
total area under the curve of the point estimate that is outside the interval [estimator ± (reliability
factor X standard error)] as α and the area within the interval as 1 – α and give the following
probabilistic interpretation:
Probabilistic interpretation: In repeated sampling, from a normally distributed population with a

known standard deviation, 100(1 – α) of all intervals constructed will in the long run include the
population parameter of interest.
̅ , ̅ , ̅ , and ̅ all fall within the

95% interval about µ, and these
intervals about these sample
means include the value of µ
Suppose that in the figure above, 15 additional confidence intervals are constructed so that
there are a total of 20 confidence intervals for the population mean. When the confidence level is 95%,
we can expect that only one out of the 20 confidence intervals that we constructed would not include
the population parameter being estimated. In the case above, ̅ did not include the true value of the
population mean, µ.
The quantity 1 – α, in this case is called the confidence coefficient (or confidence level), and the
interval [estimator ± (reliability factor X standard error)] is called a confidence interval for the population
parameter of interest. When (1 – α) = 0.95, the interval is called the 95% confidence interval for the
population parameter. Another way of interpreting confidence intervals is through the practical
interpretation which may be expressed as follows:
Practical interpretation: When sampling is from a normally distributed population with known
standard deviation, we are 100(1 – α) percent confident that the single computed interval,
[estimator ± (reliability factor X standard error)], contains the population parameter of interest.
̅
√
where = standard error (SE)
√
Sample Problem:
A physical therapist wished to estimate, with 99% confidence, the mean maximal strength of a
particular muscle in a certain group of individuals. He is willing to assume that strength scores are
normally distributed with a variance of 144. A sample of 15 subjects who participated in the study
yielded a mean of 84.3
Solution:
The z value corresponding to a confidence (or reliability) coefficient of 0.99 is 2.576.
The 99% confidence interval for µ is
( ⁄ )
√
Interpretation (Practical): We are 99% confident that the mean maximal strength of the
particular muscle in the population is between 76.3 and 92.3.
It will not always be possible or prudent to assume that the population of interest is normally
distributed. However, based on the central limit theorem, this will not deter us if the sample size is
sufficiently large. For sufficiently large samples , the sampling distribution of ̅ is
approximately normally distributed regardless of how the parent population is distributed.
Sample Problem:
Punctuality of patients in keeping appointments is of interest to a research team. In a study of

patient flow through the offices of general practitioners, it was found that a sample of 35 patients were
17.2 minutes late for appointments, on the average. Previous research had shown the standard
deviation to be about 8 minutes. The population distribution was felt to be nonnormal. What is the
90% confidence interval for µ, the true mean amount of time late for appointments?
Solution:
Since the sample size is satisfactorily large (greater than 30), and since the population standard
deviation, σ, is known, we draw on the central limit theorem and assume the sampling distribution of ̅
to be approximately normally distributed.
( ⁄ )
√
Interpretation (Practical): We are 90% confident that within the population, the average
amount of time patients are late for appointments with their general practitioners is between
15.0 to 19.4 minutes.
In the previous section, we outlined the procedure for constructing a confidence interval for
the population mean which requires the knowledge of the variance of the population from which the
sample is drawn. It may seem somewhat strange that one can have knowledge of the population
variance but not know the value of the population mean. Indeed, the usual case is that in most
situations, both the population variance as well as the population mean are unknown. For example,
although the statistic
̅
√
is normally distributed when the population is normally distributed, and is at least approximately
normally distributed when n is large, regardless of whether or not the population is normally or
nonnormally distributed, we cannot use the z statistic because σ is unknown. However, all is not lost,
and the most logical solution to the problem is to use the sample standard deviation, s, in place of σ.
When the sample size is sufficiently large , our faith in s as an approximation of σ is usually
substantial, and we may feel justified in using normal distribution theory to construct a confidence
interval for the population mean just like we have done previously.
̅
√
It is when we do not have a sufficiently large sample size that it becomes necessary for us to
find an alternative procedure for constructing confidence intervals.
This alternative is the Student’s t distribution, usually shortened to t distribution which is the
result of the work of Gosset writing under the pseudonym of “Student”. The following quantity follows
the t distribution:
̅
√
1. It has a mean of 0.
2. It is symmetrical about the mean.
3. Generally, it has a variance greater than 1, but the variance approaches 1 as the sample size
becomes large. For df > 2, the variance of the t distribution is df/(df – 2), where df is the
degrees of freedom. Alternatively, since here df = n – 1 for n > 3, the variance of the t
distribution can be written as (n – 1)/(n – 3).
4. The variable ranges from to .
5. The t distribution is actually a family of distributions, since there is a different distribution for
each sample value of n – 1, the divisor used in computing s2. Recall that n – 1 is referred to as
degrees of freedom.
SOURCE: Daniel, Wayne. Biostatistics: A Foundation For Analysis in the Health Sciences 6e (1995).
6. Compared to the normal distribution, the t distribution is less peaked in the center and has
higher tails.
Despite the need to use the t distribution rather than the standard normal distribution, the
general procedure for constructing confidence intervals is still the same. The expression
still applies. The only difference is the source of the reliability coefficient. It is now obtained from the
table of the t distribution instead of the table of the standard normal distribution. To be more specific,
when sampling is from a normal distribution whose standard deviation, σ, is unknown, the 100(1 – α)
percent confidence interval for the population mean, µ, is given by
̅
√
Note that the sample must still be drawn from a normal distribution to justify the valid use of
the t distribution. However, it has been empirically demonstrated that moderate departures from this
requirement can be tolerated. Therefore, the t distribution is used even when there is knowledge that
the parent population somewhat deviates from normality. Most researchers consider the assumption
of at least a mound-shaped population to be acceptable.
Sample Problem:
Maureen McCauley conducted a study to evaluate the effect of on-the-job body mechanics
instruction on the work performance of newly employed young workers. She used two randomly
selected groups of subjects, an experimental group and a control group. The experimental group
received one hour of back school training provided by an occupational therapist. The control group did
not receive this training. A criterion-referenced Body Mechanics Evaluation Checklist was used to
evaluate each worker’s lifting, lowering, pulling, and transferring of objects in the work environment. A
correctly performed task received a score of 1. The 15 control subjects made a mean score of 11.53 on
the evaluation with a standard deviation of 3.681. We assume that these 15 controls behave as a
random sample from a population of similar subjects. We wish to use these sample data to estimate
the mean score for the population.
SOURCE: Maureen McCauley, “The Effect of Body Mechanics Instruction on Work Performance Among Young Workers,” The
American Journal of Occupational Therapy, 44 (1990), 402-407 as printed in Wayne Daniel, Biostatistics: A Foundation for Analysis
in the Health Sciences” 6e (1995).
Solution:
1. Since the level of confidence was not stated, assume that a 95% confidence interval is desired.
2. Find the reliability coefficient, the value of t associated with a confidence coefficient of .95 and
n – 1 = 14 degrees of freedom. Since a 95% confidence interval leaves .05 of the area under the
curve of t to be equally divided between the two tails, we need the value of t to the right of
which lies .025 of the area. Alternatively, Microsoft Excel’s T.INV.2T function can be used. The
syntax for this function is as follows: =T.INV.2T(probability, deg_freedom) where probability
stands for α and deg_freedom stands for degrees of freedom or n – 1. The T.INV.2T function
returns the two-tailed inverse of the t distribution. Note that a similar function, T.INV exists but
returns the left-tailed inverse of the t distribution. For Excel versions earlier than 2010, the
analogous function to T.INV.2T is the TINV function with the syntax =TINV(probability,
deg_freedom) where probability and deg_freedom are as previously defined. The value of t,
which is our reliability coefficient, is found to be 2.1448.
3. Construct the 95% confidence interval as follows:
̅
√
( ⁄ )
√
Interpretation (Practical): We are 95% confident that the mean score for the population lies
between 9.49 and 13.57
 sample size
 functional form of the sampled population (whether it is normally or nonnormally distributed)
 knowledge of the population variance
SOURCE: Daniel, Wayne. Biostatistics: A Foundation for Analysis in the Health Sciences 6e (1995).
̅ ̅ ⁄
√
Sample Problem:
A research team is interested in the difference between serum uric acid levels in patients with
and without Down’s syndrome. In a large hospital for the treatment of the mentally retarded, a sample
of 12 individuals with Down’s syndrome yielded a mean of ̅ = 4.5 mg/100 ml. In a general hospital a
sample of 15 normal individuals of the same age and sex were found to have a mean value of ̅ = 3.4
mg/ 100 ml. If it is reasonable to assume that the two populations of values are normally distributed
with variances equal to 1 and 1.5, find the 95% confidence interval for .
Solution:
1. Since the level of confidence was not stated, assume that a 95% confidence interval is desired.
2. Find the reliability coefficient, the value of z associated with a confidence coefficient of .95
from the table of the standard normal distribution or using Microsoft Excel’s NORM.S.INV
function.
3. Construct the 95% confidence interval as follows:
̅ ̅ ⁄
√ (√ )
Interpretation (practical and probabilistic): We are 95% confident that the true difference,
, is somewhere between 0.26 and 1.94, because, in repeated sampling 95% of the
intervals constructed in this manner would include the difference between the true means.
Constructing confidence intervals for the difference between two population means when
sampling is from nonnormal populations is similarly to what we have previously done in constructing
confidence intervals for a single population mean when the samples were taken from a nonnormal
population. Remember that both n1 and n2 must be sufficiently large for the central limit theorem to be
valid.
In cases when the population variances are unknown, and the confidence interval for the
difference between two population means is desired, the t distribution can be used as a source of the
reliability factor if certain assumptions are met. We must know, or be willing to assume, that both
sampled populations have a normal distribution.
If the assumption of equal population variances is justified, the two sample variances that we
compute from our two independent samples may be considered as estimates of the same quantity, the
common variance. To capitalize on this, a pooled estimate of the common variance is calculated. The
pooled estimate is obtained by calculating the weighted average of the two sample variances where
each sample variance is weighted by its degrees of freedom. If n is equal for both populations, the
pooled estimate is the arithmetic mean of the two sample variances. However, if n is unequal for both
populations, the weighted average takes advantage of the additional information provided by the
larger sample. The pooled estimate is given by the formula:
Thus, the 100(1 – α) percent confidence interval for is given by
̅ ̅ ( ⁄ )
√
Sample Problem:
The purpose of a study by Stone et al. was to determine the effects of long-term exercise
intervention on corporate executives enrolled in a supervised fitness program. Data were collected on
13 subjects (the exercise group) who voluntarily entered a supervised exercise program and remained
active for an average of 13 years and 17 subjects (the sedentary group) who elected not to join the
fitness program. Among the data collected on the subjects was maximum number of sit-ups
completed in 30 seconds. The exercise group had a mean and standard deviation for this variable of
21.0 and 4.9, respectively. The mean and standard deviation for the sedentary group were 12.1 and 5.6,
respectively. We assume that the two populations of overall muscle condition measures are
approximately normally distributed and that the two population variances are equal. We wish to
construct a 95% confidence interval for the difference between the means of the populations
represented by these two samples.
SOURCE: William J. Stone, Debra E. Rothstein, and Cynthia L. Shoenhair, “Coronary Health Disease Risk Factors and Health
Related Fitness in Long-Term Exercising Versus Sedentary Corporate Executives,” American Journal of Health Promotion, 5,
(1991), 169-173 as printed in Wayne Daniel, Biostatistics: A Foundation for Analysis in the Health Sciences” 6e (1995).
Solution:
1. Calculate the pooled estimate of the common population variance.
2. Substitute the given values into the formula for constructing the confidence interval for the
difference between two means when the population variances are unknown and equal
population variances are assumed.
Interpretation (practical and probabilistic): We are 95% confident that the difference between
population means is somewhere between 4.9 and 12.9. We can say this because we know that
if we were to repeat the study many, many times, and construct confidence intervals in the
same manner, around 95% of the intervals would include the difference between the
population means.
When it cannot be determined whether the variances of two populations of interest are equal,
even though both populations may be assumed to have normal distributions, it is not correct to use
the t distribution as we have discussed for constructing some of the confidence intervals. The problem
lies in the fact that the quantity
̅ ̅
does not follow a t distribution with n1 + n2 – 2 degrees of freedom when the population variances are
not equal. Cochran proposed a solution which consists of computing the reliability factor, ( ⁄ ) by
the following formula:
( ⁄ )
where


 ( ⁄ ) for n1 – 1 degrees of freedom
 ( ⁄ ) for n2 – 1 degrees of freedom
An approximate 100(1 – α) percent confidence interval for is given by
̅ ̅ ( ⁄ )
√
Sample Problem:
In the study by Stone et al. described in the previous example, the investigators also reported
the following information on a measure of overall muscle condition scores by the subjects:
Sample n Mean Standard Deviation

Exercise group 13 4.5 .3
Sedentary group 17 3.7 1.0
We assume that the two populations of overall muscle condition scores are approximately
normally distributed. We are unwilling to assume, however, that the two population variances are
equal. We wish to construct a 95% confidence interval for the difference between the mean overall
muscle condition scores of the two populations represented by the samples.
Solution:
1. Find the values of t1 and t2. Using the tabulated values of the t distribution or Microsoft
Excel’s T.INV.2T function, it can be seen that with 12 degrees of freedom and α = 0.05,
t1 = 2.1788. Similarly, with 16 degrees of freedom and α = 0.05, t2 = 2.1199.
2. Compute for the value of t’
( ⁄ ) ( )
( ) ( )
3. Substitute the values into the formula for constructing the confidence interval for the
difference between two means when the population variances are unknown and equal
population variances are not assumed.
When constructing a confidence interval for the difference between two population means,
use the following figure do decide quickly whether the reliability factor should be z, t, or t’.
Estimation
Health workers often ask questions related to population proportions. What proportion of
patients who receive a particular type of treatment recover? What proportion of some population if
affected by a certain disease? What proportion of a population is immune to a particular disease?
The estimation of a population proportion is similar to what we have done with estimating
population means or the difference between two population means. The expression
still applies. A sample is taken from the population of interest, and the sample proportion, , is
calculated. The sample proportion serves as the point estimate of the population proportion.
When both and are greater than 5, the sampling distribution of may be
considered to be quite close to the normal distribution. When this condition is satisfied, a z-value from
the standard normal distribution is used as the reliability factor. The confidence interval for a
population can be constructed by using the following formula:
( ⁄ )√
Sample Problem:
Mathers et al. found that in a sample of 591 admitted to a psychiatric hospital, 204 admitted to
using cannabis at least once in their lifetime. We wish to construct a 95% confidence interval for the
proportion of lifetime cannabis users in the sampled population of psychiatric hospital admissions.
SOURCE: D.C. Mathers, A.H. Ghodse, A.W. Caan, and S.A. Scott, “Cannabis Use in a Large Sample of Acute Psychiatric
Admissions,” British Journal of Addiction, 86 (1991), 779-784 as printed in Wayne Daniel, Biostatistics: A Foundation for Analysis in
the Health Sciences” 6e (1995).
Solution:
1. Calculate the point estimate of the population proportion,
2. Determine the corresponding reliability factor for the desired confidence interval. In this case,
the reliability factor that should be used is 1.96.
3. Construct the confidence interval for p using the formula
√
Interpretation (practical and probabilistic): We are 95% confident that the population
proportion p is between .3069 and .3835, since, in repeated sampling, about 95% of the
intervals constructed in the manner of the current single interval would include the true p. On
the basis of these results we would expect, with 95% confidence, to find somewhere between
30.69% and 38.35% of psychiatric hospital admissions to have a history of cannabis use.
There are occasions when we might be interested in the magnitude of the difference between
two proportions. For example, we may want to compare men and women, two age groups, two
socioeconomic groups, or two diagnostic groups with respect to the proportion possessing some
characteristic of interest. A 100(1 – α) percent confidence interval for P1 – P2 is given by
( ⁄ )
√
Sample problem:
Borst et al. investigated the relation of ego development, age, gender, and diagnosis to
suicidality among adolescent psychiatric inpatients. Their sample consisted of 96 boys and 123 girls
between the ages of 12 and 16 years selected from admissions to a child and adolescent unit of a
private psychiatric hospital. Suicide attempts were reported by 18 of the boys and 60 of the girls. Let us
assume that the girls behave like a simple random sample from a population of similar girls and that
the boys likewise may be considered a simple random sample from a population of similar boys. For
these two populations, we wish to construct a 99% confidence interval for the difference between the
proportions of suicide attempters.
SOURCE: Sophie R. Borst, Gil G. Noam, and John A. Bartok, “Adolescent Suicidality: A Clinical-Development Approach,” Journal
of the American Academy of Child and Adolescent Psychiatry, 30 (1991), 796-803 as printed in Wayne Daniel, Biostatistics: A
Foundation for Analysis in the Health Sciences” 6e (1995).
Solution:
1. Calculate the sample proportions for the girls and boys
2. Determine the corresponding reliability factor for the desired confidence interval. In this case,
the reliability factor that should be used is 2.576.
3. Construct the confidence interval for p1 – p2 using the formula
( ⁄ )√
Interpretation (practical): We are 99% confident that for the sampled proportions, the
proportion of suicide attempts among girls exceeds the proportion of suicide attempts among
boys by somewhere between .1450 and .4556.
NAME:__________________________________________________________ DATE:______________
A. Software Application (35 points)
Obtain a template of the spreadsheet that you will be using for this exercise from your
instructor. Using the NORM.S.INV and T.INV.2T functions of Microsoft Excel 2010 (NORMSINV and TINV
for earlier versions), design an algorithm that will return the lower and upper limits of the confidence
interval when the necessary data are inputted into the cells of the spreadsheet.
B. Problem Solving (55 points)
Use the spreadsheet that you have just designed to help you answer the following questions:
1. We wish to estimate the mean serum indirect bilirubin level of 4-day-old infants. The mean for
a sample of 16 infants was found to be 5.98 mg/100 cc. Assuming bilirubin levels in 4-day-old
infants are approximately normally distributed with a standard deviation of 3.5 mg/100 cc find:
a. The 90% confidence interval for µ (2 points)

Lower limit: ______________
Upper limit: ______________
b. The 95% confidence interval for µ (2 points)
Lower limit: ______________
Upper limit: ______________
c. The 99% confidence interval for µ (2 points)
Lower limit: ______________
Upper limit: ______________
2. In an investigation of the flow and volume dependence of the total respiratory system in a
group of mechanically ventilated patients with chronic obstructive pulmonary disease (COPD),
Tantucci et al. collected the following baseline values on constant inspiratory flow (L/s): .90, .97,
1.03, 1.10, 1.04, 1.00. Assume that the six subjects constitute a simple random sample from a
normally distributed population of similar subjects.
SOURCE: C. Tantucci, C. Corbeil, M. Chassé, J. Braidy, N. Matar, and J. Milic-Emili, “Flow Resistance in Patients with
Chronic Obstructive Pulmonary Disease in Acute Respiratory Failure,” American Review of Respiratory Disease, 144
(1991), 384-389 as printed in Wayne Daniel, Biostatistics: A Foundation for Analysis in the Health Sciences” 6e (1995).
a. What is the point estimate of the population mean? (1 point)

________________________
b. What is the standard deviation of the sample? (2 points)
________________________
c. What is the estimated standard error of the sample mean? (2 points)
________________________
d. Construct a 95% confidence interval for the population mean constant inspiratory flow
(2 points)
Lower limit: ______________
Upper limit: ______________
e. What is the precision of the estimate? (2 points)
_________________________
f. State the probabilistic interpretation of the confidence interval you constructed. (2 points)
___________________________________________________________________________
___________________________________________________________________________
___________________________________________________________________________
g. State the practical interpretation of the confidence interval you constructed. (2 points)
___________________________________________________________________________
___________________________________________________________________________
___________________________________________________________________________
3. Zucker and Archer state that N-NITROSOBIS (2-oxopropyl)amine (BOP) and related β-oxidized
nitrosamines produce a high incidence of pancreatic ductular tumors in the Syrian golden
hamster. They studied the effect on body weight, plasma glucose, insulin, and plasma
glutamate-oxaloacetate transaminase (GOT) levels of exposure of hamsters in vivo to BOP. The
investigators reported the following plasma glucose levels for 8 treated and 12 untreated
animals:
Subject Group Sample Mean Sample Standard Deviation

Untreated 101 mg/dl 5 mg/dl
Treated 74 g/dl 6 mg/dl
SOURCE: Peter F. Zucker and Michael C. Archer, “Alterations in Pancreatic Islet Function Produced by Carcinogenic
Nitrosamines in the Syrian Hamster,” American Journal of Pathology, 133 (1998), 573-577 as printed in Wayne Daniel,
Biostatistics: A Foundation for Analysis in the Health Sciences” 6e (1995).
a. State the necessary assumptions for this problem (2 points)
___________________________________________________________________________
___________________________________________________________________________
___________________________________________________________________________
Construct the following confidence intervals:
b. The 90% confidence interval for µ1 – µ2 (2 points)

Lower limit: ______________
Upper limit: ______________
c. The 95% confidence interval for µ1 – µ2 (2 points)
Lower limit: ______________
Upper limit: ______________
d. The 99% confidence interval for µ1 – µ2 (2 points)
Lower limit: ______________
Upper limit: ______________
4. The average length of stay of a sample of 20 patients discharged from a general hospital was 7
days with a standard deviation of 2 days. A sample of 24 patients discharged from a chronic
disease hospital has an average length of stay of 36 days with a standard deviation of 10 days.
Assuming normally distributed populations with unequal variances, construct the following
confidence intervals:
a. The 90% confidence interval for µ1 – µ2 (2 points)

Lower limit: ______________
Upper limit: ______________
b. The 95% confidence interval for µ1 – µ2 (2 points)
Lower limit: ______________
Upper limit: ______________
c. The 99% confidence interval for µ1 – µ2 (2 points)
Lower limit: ______________
Upper limit: ______________
5. Rothberg and Lits studied the effect on birth weight of maternal stress during pregnancy.
Participants were 86 Caucasian mothers with a history of stress who had no known medical or
obstetric risk factors for reduced birth weight. The investigators found that 11 of the mothers
in the study gave birth to babies satisfying the criterion for low birth weight.
SOURCE: Alan D. Rothberg and Bernice Lits, “Psychosocial Support for Maternal Stress During Pregnancy: Effect on
Birth Weight,” American Journal of Obstetrics and Gynecology, 165 (1991), 403-407 as printed in Wayne Daniel,
Biostatistics: A Foundation for Analysis in the Health Sciences” 6e (1995).
a. What is the point estimate of the population proportion? (1 point)

________________________
b. What is the standard deviation of the sample? (2 points)
________________________
c. What is the estimated standard error of the sample proportion? (3 points)
________________________
d. Construct a 99% confidence interval for the population proportion of mothers with a
history of stress that gives birth to underweight babies (2 points).
Lower limit: ______________

Upper limit: ______________
e. What is the precision of the estimate? (2 points)
_________________________
f. State the probabilistic interpretation of the confidence interval you constructed. (2 points)
___________________________________________________________________________
___________________________________________________________________________
___________________________________________________________________________
g. State the practical interpretation of the confidence interval you constructed. (2 points)
___________________________________________________________________________
___________________________________________________________________________
___________________________________________________________________________
6. Research by Lane et al. assessed differences in breast cancer screening practices between
samples of predominantly low-income women aged 50 to 75 using county-funded health
centers and women in the same age group residing in the towns where the health centers are
located. Of the 404 respondents selected from the community at large, 59.2% agreed with the
following statement about breast cancer: “Women live longer if the cancer is found early.”
Among the 795 in the sample of health center users, 44.9% agreed with the statement.
SOURCE: Etta Williams, Leclair Bissell, and Eleanor Sullivan, “The Effects of Co-dependence on Physicians and
Nurses,” British Journal of Addiction, 86 (1991), 37-42 as printed in Wayne Daniel, Biostatistics: A Foundation for
Analysis in the Health Sciences” 6e (1995).
a. State the assumptions that you think are appropriate for calculating the interval estimate:
(2 points)
___________________________________________________________________________
___________________________________________________________________________
___________________________________________________________________________
Construct the following confidence intervals:
b. The 90% confidence interval for P1 – P2 (2 points)

Lower limit: ______________
Upper limit: ______________
c. The 95% confidence interval for P1 – P2 (2 points)
Lower limit: ______________
Upper limit: ______________
d. The 99% confidence interval for P1 – P2 (2 points)
Lower limit: ______________
Upper limit: ______________
In the previous lesson, the term hypothesis has already been introduced. To review, a
hypothesis is a belief concerning a parameter. As we have indicated previously, a parameter may be a
population mean, proportion, variance, standard deviation, etc. We also mentioned that there are two
forms of hypothesis: null hypothesis and alternative hypothesis. The null hypothesis is the prevalent
opinion, previous knowledge, basic assumption, or prevailing theory while the alternative hypothesis is
the rival opinion. The null hypothesis is assumed to be true as long as we find evidence against it. If a
sample gives strong enough evidence against the null hypothesis, then the alternative hypothesis
comes into force.
Below are different examples of null and alternative hypotheses. Study how these hypotheses
were formulated and try to identify the differences between the two in the manner of how these two
hypotheses were formulated.
H0: The mean height of males equals 174 cm.

HA: The mean height of males is taller than 174 cm.
H0: Half of the population is in favor of the nuclear power plant operation.
HA: More than half of the population is in favor of the nuclear power plant operation.
H0: The amount of overtime work is equal for males and females in Community X.
HA: The amount of overtime work is not equal for males and females in Community X.
H0: There is no correlation between the interest rate and gold price in the market.
HA: There is a correlation between the interest rate and gold price in the market.
The risk that our hypothesis is wrong is like asking the risk of a judge giving a wrong judgment
to a case. There may be a 50% probability that the hypothesis we give is wrong, just like a judge
giving a trialled person a not guilty verdict when the defendant is in fact guilty. What we need to
remember is that the null hypothesis remains valid until it is proven otherwise. Sometimes it happens
that an innocent person is proven guilty, so this is the same case that may happen to our hypothesis
testing. We may reject a null hypothesis although it is true because there is always a risk of being
wrong when we reject a null hypothesis. The occurrence of this risk is due to what we call sampling
error.
The basis of our decision on whether we should reject or accept our hypothesis is through a p
value (probability value). Let us start with the basic assumption that our null hypothesis is true. The
p-value is the probability of getting a value equal to or more extreme than the sample result, given
that the null hypothesis is true. Our decision rule enables us to see whether we should reject or not
reject the null hypothesis and whether we should accept or reject our alternative hypothesis. If the
p-value is less than 5%, then we reject the null hypothesis and if the p-value is 5% or more, then the
null hypothesis remains valid. In any case, one must give the p-value as a justification for our decision
in hypothesis testing.
Hypothesis Testing
1. State the null hypothesis, H0; and the alternative hypothesis, HA or H1;
2. State the level of significance, α;
3. Choose the test statistic and determine its sampling distribution;
4. Determine the critical region;
5. Calculate the test statistic;
6. Make a statistical decision. (Note: This is where you make the decision of whether or not to
reject the null hypothesis);
7. Draw the conclusion about the population.
Notes on the Decision Rule:
Take note, if the p-value is less than your α which may be 0.01 with a 99% level of significance;
0.05 with a 95% level of significance; or 0.1 with a 90% level of significance) then reject the null
hypothesis. Otherwise, the null hypothesis remains valid. In any case, you must give the p-values as a
basis for your decision.
Let us look at the steps of hypothesis testing in detail.
The hypotheses that we must formulate must be something that we can “test” so that it opens
the way for a statistical assessment. There are three general statements to keep in mind when framing
the null and alternative hypotheses. These three general statements are as follows:
 The null hypothesis is the hypothesis of “no difference.” This means that the null hypothesis is
a statement of equality.
 H1 is usually the research hypothesis, meaning it is the hypothesis the investigator believes in.
 H0 is always framed in hopes of being able to reject it so that H1 could be accepted.
The formulation of our hypothesis is dependent on whether we want to perform a one-tailed

hypothesis test or a two-tailed hypothesis test. Obviously, when we indicate that we are performing a
one-tailed hypothesis test, we must use either the left tail or the right tail of the normal distribution
curve. Meanwhile, when we want to perform a two-tailed hypothesis test, we are interested in using
both the left and right tails of the normal distribution curve.
In a one-tailed test, we must know beforehand that only deviations to one direction are
possible. An example would be to say that our Ho : Pa = Pb, while the alternative hypothesis takes the
form of being either “less than” or “greater than” the other parameter or statistic. This is either
expressed as HA: Pa > Pb, or Pa < Pb.
In a two-tailed test, the assumption is to use both the left and right tails of the normally
distributed curve. We typically use the two-tailed test if we do not have knowledge regarding the
direction of the data set. In the two-tailed test, the deviations from the null hypothesis are in both
directions of the normally distributed curve. In formulating the alternative hypothesis in a two-tailed
test, it takes the form “different than.” An example of our null hypothesis is H0: Pa is equal to Pb while
an example of the alternative hypothesis is HA: Pa is not equal to Pb.
In this step, we determine our probability level. It is denoted by the Greek letter α (alpha) and
is conventionally taken as 5%, 1%, or 10%. The significance level that the researcher chooses is the risk
that he/she is willing to take of making the wrong decision of dismissing the null hypothesis as being
very unlikely and to favor the null hypothesis instead.
There are two types of errors: Type 1 error or α error and Type II error or β error. We commit a
Type 1 error if the null hypothesis is in fact true and it is just unfortunate that our sample yields an
unlikely result that we reject the null hypothesis. This occurs when there is really no difference
between the population parameters being tested, but the investigator is misled by chance differences
in the sample data. Therefore, the type I error is the error of rejecting a true hypothesis. On the other
hand, the Type II error is the error that we commit when we do not reject a false null hypothesis. It
occurs when there really is a difference between the population parameters being tested, but the
investigator misses the difference. It can result from either too much sampling variability or the
insensitivity of the test employed, or both, and depends on the number of observations (sample size)
included in the study.
Condition of Null Hypothesis

Possible Action
True False
Correct action Type II error
Fail to reject H0
Probability = 1 – α Probability = β
Type I error Correct action
Reject H0
Probability = α Probability = 1 – β
In this step, we select the appropriate tool to test our particular hypothesis. When we choose
the test statistic, it has its own sampling distribution that we use to assess the probability of occurrence
of sample results under our null hypothesis. There is a wide array of statistical tests that we can use to
test the hypothesis before we can make any decision of rejecting the null hypothesis or not rejecting it.
What is the criterion for test selection?
The choice of a particular test statistic depends on several criteria including the types of
variables (qualitative or quantitative), level of measurement (nominal, ordinal, interval, or ratio),
whether samples are dependent (related or before-after measurements) or independent (different
groups).
Hypothesis Testing
There is a wide array of reasons for doing the tests. We may want to determine whether the
sample could have come from a population with a stipulated mean or proportion, or from a population
of some pre-specified distribution (one sample case). It could be that we want to do the test because
we want to compare two means or two proportions (two sample cases). We may also want to use the
test because we are interested in comparing more than two means or proportions (k sample cases).
Better yet, we may want to determine whether a relationship exists between the variables that we are
studying. The assumption of selecting the test to use also depends on whether we want to do a
parametric test or a non-parametric test. A parametric test is appropriate when the data we have is
obtained through the random selection of the sample. The normal distribution of the population to
which the samples were drawn exists, and when more than one population is sampled, there is
equality of variances (homoscedasticity). Also, the numerical data measured must either be on an
interval or a ratio scale. The non-parametric test is a type of a distribution-free test. This is a test in
which no hypothesis is made about the specific values of population parameters. If the researcher
doubts the validity of whether the study satisfies the parametric assumptions, then the non-parametric
test should be employed. All sets of data, which are not truly numerical, are tested through non-
parametric tests.
In this step, the critical region is our region of rejection. The critical region is the set of values
of the test statistic which leads to the rejection of the null hypothesis. The critical region indicates the
values whose probability of occurrence is less than or equal to the level of significance. The critical
region is usually found at the tail end of the distribution. It is also similar to what comprises the region
of acceptance. The size of the critical region is determined by the researcher’s chosen level of
significance. The location of the critical region is also determined by the nature of the alternative
hypothesis and whether the researcher opted to do a one-tailed or a two-tailed test of hypothesis.
In this step, the test statistic chosen is calculated to help us decide on whether or not the null
hypothesis should be rejected. It is presumed that the level of significance, test statistic, and the
critical region have been determined prior to doing this step. The formula to be used by the
researcher varies depending on what test statistic is chosen.
In this step, we either decide whether to reject or not reject the null hypothesis. We reject the
null hypothesis if the computed value of our test statistic falls within the critical region. Otherwise, it is
not rejected. When the null hypothesis is rejected, the results are statistically significant and the
observed difference may not be attributed to sampling variation. If the hypothesis is not rejected, the
results are not statistically significant and the sampling variation is the likely explanation of the
observed differences.
The rejection of the null hypothesis leads to a conclusion stated in the form of the alternative
hypothesis. If a statistical decision is to not reject the null hypothesis, we do not necessarily accept the
null hypothesis. Instead, we say that there is no sufficient evidence to conclude whatever is stated in
the alternative hypothesis. The table below shows an example of how we draw conclusions based on
our statistical decision.
Statistical Decision Conclusion

(We state the alternative hypothesis)
The proportion of students who obtain a

H0 rejected grade of 2.0 or better among those using
modules for demonstration is greater
than the proportion among those using
cadavers.
(“We do not have sufficient evidence to
say that (state the alternative hypothesis)”
NOTE: We do not “accept” the null
hypothesis, we can only “reject” it.)
H0 not rejected There is no sufficient evidence to say that

the proportion of students who obtain a
grade of 2.0 or better among those using
modules for demonstration is greater
than the proportion among those using
cadavers.
 The null hypothesis (H0) which cannot be rejected is not necessarily “accepted” especially if the
sample size is small. We simply do not have enough evidence to reject it.
 Statistical significance is not the same as practical or clinical significance.
 Statistical inference is not valid for badly designed studies. Therefore, when we do a study,
before we collect the data, we have to make sure that our methods are accurate and reliable.
The sources of biases should be removed or minimized.
 Statistical analysis is not applicable for studies involving the total population.
 One of the purposes of hypothesis testing is to assist administrators and clinicians in making
decisions. However, the outcome of statistical tests is only one piece of evidence that should
influence the administrative or clinical decision. The statistical decisions must be interpreted
along with all the other relevant information to the decision maker.
Normally, when we talk about qualitative data, they are not numerical data. The type of data
may not be possible to be applied in basic mathematical operations. There is a need to categorize this
kind of data and determine the frequency per category. Working with the frequencies of qualitative
data in different categories is the most common way of presenting the data or the results of a study.
When working with the frequencies, the usual process is to compare the observed and the expected
frequencies or relate the observed frequency to the total number of observations to get a sample
proportion to which is now comparable to a pre-specified population proportion. A common example
of qualitative data is gender, where we have males and females. Another example could be blood type
where it is categorized into different categories like blood type A, B, AB, and O.
Due to the nature of the data that we have oftentimes we are limited in terms of methods or
tests that we can use to deal with this kind of data. The common statistical tests that we will be taking
up and are indicated in this module are: z test for 1 proportion; z test for 2 proportions; and the chi
square test of homogeneity.
In most qualitative tests that we employ, the data that we have is usually categorized into what
we call as a binomial population. A binomial population is one in which the elements of our data set
belong to either one of the two mutually exclusive and collectively exhaustive categories. There are 2
categories that are mutually exclusive because the elements can only belong to one and only one of
the categories. Just like when we talk of gender, it is either a subject is categorized as a male or a
female. We cannot have one belonging in both categorizations. Collectively exhaustive as what we
indicated earlier means that we literally exhaust the population after distributing the elements to the
categories to which they all belong. Looking again at our parameter and statistic, let us subdivide the
population into those who possess the characteristic and those who do not. This is shown on the table
below.
Parameter Statistic
Proportion who possess the
characteristic
Proportion who do not possess the
characteristic
Standard error of proportions √ √
The concept of estimation of the population parameters focuses on either the point estimation
or the interval estimation. Point estimate for a population parameter is the corresponding statistic. The
point estimate of a population proportion, possessing a characteristic of interest as denoted by P, is the
sample proportion, p. This proportion is expressed below:
The point estimate is used if we are definitely sure that it will hit the parameter but if not, we
can use the interval estimate. Through the interval estimate, we can set bounding values of the
statistic, which will include the parameter with a specified degree of confidence. This is just like
showing the interval or the possible values where the statistic will lie depending on the degree of
confidence. For example, if we know that approximately 95% of the possible values of a statistic, let’s
say proportion lie within the + 1.96 standard deviations from the parameter proportion. The 1.96 here
is the z-deviate score of our 95% degree of confidence.
Take note that the common z-deviate scores for the following commonly used degree of
confidence assuming a 2-tailed test is performed are shown on the table below:
Degree of Confidence Z-deviate score

90% 1.64
95% 1.96
99% 2.58
If we want to do a 1-tailed test, the following are the degree of confidence and the z-deviate
scores.
Degree of Confidence Z-deviate score

90% 1.28
95% 1.64
99% 2.33
Meanwhile, the interval estimate is expressed on the illustration below:

Analysis of Qualitative Data
The 2 points that are + 1.96 standard deviations from P are:

√
The equation is expressed as:

√
Since the P is unknown, we substitute its point estimator, p in equation as:

√
Let us look at an example that uses the concept of estimation. Let’s say that a survey was
conducted and that study dealt with the dental health practices of children in a certain school. Of the
300 school children interviewed, 123 school children indicated that they had a regular dental check-up
twice a year. Now, the problem is this: (1) What percent of the school children in the sample had
regular dental check-ups? (2) Give an estimate of the school children population who had regular
dental check-ups? (3) Compute and interpret the 95% confidence interval estimate of the population
proportion.
First, let us answer the first problem. In this problem, we are asked what percent of the school
children in the sample had regular check-ups. Thus, we are interested in getting the proportion of
school children with regular check-up over the total number of school children included in the study.
We can express this using the formula below:
p = Number of school children with regular check-up X 100

Total number of school children examined
p = 123/300 X 100
p = 41%
The proportion of school children with regular check-up over those school children examined
is 41%. Let’s proceed with the next problem. We are now asked to give an estimate on the school
children population who had regular dental check-ups. The population proportion P is unknown, so
we use the point estimator, p to estimate for the school children population who had regular dental
check-ups. In this case, the point estimate of P is 41%. For the third problem, we are asked to
compute and interpret the 95% confidence interval estimate of the population proportion of school
children with regular dental check-up. The 95% confidence interval estimate is calculated using the
formula below.
√
In this problem, we can interpret the values we’ve obtained by saying that we are 95%
confident that the proportion of school children in the population who submit themselves to regular
dental check-ups is anywhere in between 35.4% and 46.6%.
In the determination for interval estimate, 90%, 95% and 99% are the commonly used
confidence coefficients. The confidence coefficient is a degree of certainty that the parameter being
estimated is within the computed confidence interval.
Using the same problem indicated above, compute and interpret the 90% and 99% confidence
interval estimates. How would these confidence interval estimates compare with the 95% confidence
interval estimate? Using the same formula but changing the confidence coefficients, we can calculate
for the p1 and the p2 for both the confidence coefficients. Try to see how the succeeding values were
derived. For the 90% confidence interval estimate, the values are in between 36.3% and 45.6%
whereas in the 99% confidence interval estimate, the values are in between 33.7% and 48.3%. Take
note that as we have a higher confidence coefficient, the wider our estimates are. Hence, the less
precise, is our confidence interval.
The aim of performing statistical tests for 1 sample is to determine whether a sample comes
from a standard population or norm. The norm or standard population is either known based on the
past experiences, or it may be specified as the way one wants or claims the population to be, or it may
obey a basic law. Let us look at some examples of what we call as a standard population or norm.
 How the standard population gets known through past experience:

- Example1: It is known through previous Operation Timbang Projects that 30% of the
children less than 6 years of age are malnourished.
- Example2: It is known that through previous study, 90% of the individuals in the
community suffered from dengue.
- Example 3: It is known through previous surveys that common intestinal worms affect 90%
of preschoolers.
 How the standard populations are specified or claimed by interested parties:

- Example 1: The aim of the Local Government is to cover 90% of the target population for
vaccination of the ailment X.
- Example2: A newly developed drug for a particular disease is claimed by the manufacturer
to be 95% effective.
 Obeys a basic law:

- Example1: The basic law states that the probability of a male birth is equal to the
probability of a female birth is true, then we should expect 50% of births to be males and
50% females.
In our previous lesson on hypothesis testing, we had discussed the z test for single proportions.
Recall the z- test formula for estimating population proportions. The formula is shown below:
What if we have a problem and our problem is about the Operation Timbang conducted in a
community. The Operation Timbang was conducted to monitor and help out the children from
malnutrition. The survey was conducted among 200 randomly selected children to determine if the
aim of the Health Office to cover 80% of the target population was attained. It was found that 176 out
of the 200 children surveyed were provided with feeding. Did the Health office meet their objective?
To answer the problem, let us first analyze what type of test we can use to answer the problem.
Looking at it, we can say that it requires a test of hypothesis for a single proportion. We want to find
out if the data collected from a sample of 200 children support the hypothesis that 80% of the
population is addressed. The sample proportion p and the sample size requirement that both np and
nq must be greater than or equal to 5 must be initially satisfied if we want to employ the z statistic. If it
satisfies the condition, then we can proceed with the statistic. Let us now employ the steps of
hypothesis testing that we have learned from the previous lesson.
Steps:
1. Stating the hypothesis:
Null hypothesis: P is equal to 0.8
Alternative hypothesis: P is not equal to 0.8
2. Identify level of significance:

Level of significance = 0.05
3. Choose Test Statistic:

The test statistic is the z test for estimating population proportions.
√
4. Determine the critical region:
Since the level of significance is 0.05 and we want to perform a 2-tailed test, our critical region
will be:
z > 1.96 and z < -1.96
5. Do the computations:
p = 176/200 = 0.88
Substitute the computed p and the given P and compute for the z value using the test statistic
indicated above. The z value will be equal to 28.28
6. Make a Decision:
The calculated z value is 28.28 and this value is greater than the 95% confidence z value, which
is 1.96.
z = 28.28 > z0.05 = 1.96
From this, we can reject our null hypothesis and accept our alternative hypothesis.
In making inferences for two proportions, we basically consider that there are two
independent samples to make inferences for two proportions. There is a need to randomly select the
subjects from each of the two populations or we randomly allocate the volunteers to two groups to
come up with two independent samples. From the two independent samples, we can estimate the
two proportions and estimate the differences between the two proportions by the use of point
estimate and interval estimate.
The point estimate of the difference between two population proportions, , is the
difference between the sample proportions, . If the sample is sufficiently large, the
distribution is approximately normal with the mean equal to and the standard error equal to
Since P1 and P2 are unknown, we use the estimates p1 and p2 instead. From the general
formula for computing confidence interval estimates, we derive the formula for the confidence interval
estimate of the difference between two proportions as shown below:
( ⁄ )√
Let us look at an example where we can apply the given formula above. Suppose that a study
was conducted and it determined the prevalence of cholera in the 6 barangays of Manila City. Only the
results for the Barangays 1 and 2 are given. Of the 414 respondents in Barangay 2, 46 (11.1%) were
positive for cholera as compared to 62 (15.1%) of the 410 respondents in Barangay 1. The following
are the problems: (1) What is the point estimate for the difference in the proportion of positive for
cholera in Barangays 1 and 2? (2) Construct the 90%, 95% and 99% confidence interval estimates of the
difference in the proportions of positive in cholera in Barangays 1 and 2. (3) How does the confidence
coefficient affect the interval estimate calculated?
Let us begin by solving the first problem. This problem is asking us to determine the point
estimate for the difference between those who are positive for cholera in Barangays 1 and 2. The point
estimate is determined by getting the difference between the two sample proportions, p1 and p2. The
point estimate is expressed as . We can assume that p1 is the proportion of respondents who
are positive for cholera in Barangay 1 and p2 as the proportion of respondents who are positive for
cholera in Barangay 2. Getting the difference will give us a value of 4%. We will obtain a different
point estimate if we assign p1 as the proportion of respondents who are positive for cholera in
Barangay 2 and p2 as the proportion of respondents who are positive for cholera in Barangay 1. In this
case, the point estimate is -4%. Note that the negative sign just indicates that the proportion of
respondents who are positive for cholera is higher in Barangay 1.
To solve the second problem, we are asked to construct the 90%, 95% and 99% confidence
interval estimates of the difference in the proportions of the respondents who are positive for cholera
in Barangays 1 and 2. Remember the z-deviate scores for the 90%, 95% and 99% confidence levels
because these values will be substituted into the formula. The 90% confidence interval can be
calculated using the formula below:
The z-deviate score for a 90% two-tailed hypothesis is 1.64. This value has to be changed if the
confidence level is 95% or 99%. Using the formula above, the confidence estimate will be -7.8% and
-0.2%. The solution is shown below:
Try to solve for the 95% and the 99% confidence interval estimates. It is expected that the 95%
confidence interval estimate will be -8.6% and 0.6% while the 99% confidence interval estimate will be
-10.0% and 2.0%. For the third problem, you are asked how the confidence coefficient correlates with
the interval estimate calculated. So, what do you think? If your answer is that the higher the confidence
coefficient, the wider the confidence interval estimate becomes, then you got it right!
Hypothesis testing is not limited to testing a single proportion but can also be applied for two
proportions similar to testing single means and the difference between two means. In the test of
hypothesis for two proportions, the most commonly tested hypothesis for population proportions is:
Under the null hypothesis, , two possible conditions may arise:

a) ;
b) , where P is the common value to which both P1 and P2 are equal.
The pooled estimate P indicated above is more precise than either one of the 2 distinct
estimates. Because of the condition set in (b), the p1 and p2 are values that are both estimating P. We
can combine the samples to derive a pooled estimate of P (p), which is more precise than either one of
the two distinct estimates. The formulas are shown on the next page.
̅
The z-statistic is therefore calculated as:
̅̅ ̅̅
√
where ̅ ̅
The sample size requirement for the application of the previous equation is, that both n1pq and
n2pq should be greater than or equal to 5.0. Let us look at an example. A nutritionist screened 465
males and 656 females and classified them as to whether they had normal or elevated blood uric acid
(BUA) levels. Results showed that 143 males and 133 females had elevated BUA. Do these findings
suggest that there is a higher proportion of males with elevated BUA? Set alpha to 0.10.
Given:
• n (♂) = 465; p (♂ w/ hyperuricemia) = 143/465 or 30.75%
1 1
• n (♀) = 656; p (♀ w/ hyperuricemia) = 133/656 or 20.27%
2 2
H0: p = p
1 2
HA: p > p
1 2
Level of Significance (α) = 0.10
Test statistic: z-test for 2 proportions
̅
̅̅ ̅̅
√
Critical Region (α = 0.1): > 1.28 (at the ‘tail’)
Calculations: p = (143 + 133)/(465 + 656) = 276/1121 = 0.2462 or (24.62%)
q = 100 – 24.62 = 75.38%
√ √
Decision: Since 4.0129 is within the critical region, reject Ho.
“The proportion of males with hyperuricemia is greater than the proportion of females with hyperuricemia.”
Take note, that the alternative hypothesis shows that proportion 1 is greater than proportion 2.
The calculations and the hypothesis of going about it are shown above. Now, let us look at how the
hypothesis testing is to be undertaken if we say that P1 is not equal to P2. The calculations and the
steps are shown on the next page:
Given:
• n (♂) = 465; p (♂ w/ hyperuricemia) = 143/465 or 30.75%
1 1
• n (♀) = 656; p (♀ w/ hyperuricemia) = 133/656 or 20.27%
2 2
H0: p = p
1 2
HA: p = p
1 2
Test statistic: z-test for 2 proportions
̅
̅̅ ̅̅
√
Critical Region (α/2 = 0.05): z > 1.64 and z < -1.64 (at the ‘tails’)
Calculations: p = (143 + 133)/(465 + 656) = 276/1121 = 0.2462 or (24.62%)
q = 100 – 24.62 = 75.38%
√ √
Decision: Since 4.0129 is within the critical region, reject Ho.
“The proportion of males with hyperuricemia is not equal to the proportion of females with hyperuricemia.
It is significantly higher.”
At the same level of significance, a one-tailed hypothesis test has a higher chance of rejecting
the null hypothesis compared to a two-tailed hypothesis test. In one-tailed tests of hypotheses, the
possibility of rejecting the null hypothesis increases as you increase the value of α.
The chi-square test is another useful tool for the analysis of qualitative data. The chi-square
distribution is the most frequently employed statistical technique for the analysis of count or frequency
data. The test compares whether the observed and expected frequencies fall under the same category
if the null hypothesis were true. The chi-square test has three types: the goodness of fit test, the test of
independence, and the test of homogeneity. We will discuss the test of independence and test of
homogeneity in more detail on Chapter 13: Investigating Relationships Between Variables.
NAME:__________________________________________________________ DATE:______________
For each of the following problems, carry out the steps in hypothesis testing at the designated level of
significance.
1. A researcher conducted a study to examine the reasons why occupational therapists have left the
field of occupational therapy. Her sample consisted of female certified occupational therapists who
had left the profession either permanently or temporarily. Out of 696 participants who responded
to the data-gathering survey, 63% had planned to take time off from their jobs to have and raise
children. On the basis of these data, can we conclude that, in general, more than 60% of the
subjects in the sampled population had planned to take time off to have and raise children? Let
.
a. H0 (1 point): ____________________________________________________________________
b. HA (1 points): _________________________________________________________________
c. Level of Significance:
d. Test statistic (2 points):
e. Critical region (2 points):

__________________________________________________________
f. Calculation of the test statistic (5 points):
g. Statistical Decision (2 points): ______________________________________________________

h. Conclusion (2 points): ____________________________________________________________
2. Research has suggested a high rate of alcoholism among patients with primary unipolar
depression. A study further explored this possible relationship. In 210 families of females with
primary unipolar major depression, they found that alcoholism was present in 89. Of 299 control
families, alcoholism was present in 94. Do these data provide sufficient evidence for us to conclude
that alcoholism is more likely to be present in families of subjects with unipolar depression? Let
a. H0 (1 point): ____________________________________________________________________
b. HA (1 points): _________________________________________________________________

__________________________________________________________

In this section, emphasis is applied on the hypothesis testing of means. Recall that before we
performed hypothesis testing, we initially estimated for population means by either calculating the
point estimate or constructing the interval estimate. Remember that in point estimation, a single
numerical value is used to estimate for the corresponding population parameter whereas in interval
estimation, two numerical values define an interval ranging the degrees of confidence of the
corresponding population parameter.
In a test of hypothesis of the population mean, our objective is to determine whether the
results obtained from the sample supports long established norms or whether it is consistent with
what is claimed to be the existing population value. Just like in hypothesis testing, we start by defining
the null and alternate hypotheses. The test statistic that we use depends on whether or not the
population variance is known.
Known variance:
√
Unknown variance:
√
The critical region of the computed z value depends on the level of significance that we
establish in the beginning. Meanwhile, the critical region of the t test depends on both the level of
significance and the degrees of freedom. The computed t values, which fall within the critical region
gives us reason to reject our null hypothesis. Let us look at an example of testing for the hypothesis of
a single mean. The average number of persons per household for the whole country is based on a
census made and was found out to be 5.6. If a random sample of 25 households in a survey showed a
mean household size of 5.2 persons with a standard deviation of 1.56, does this result indicate that
there has been a change in the mean household size in the Philippines since the last census? In this
example, let us assume that we want to use a 90% level of significance.
H0:
HA:
Test statistic: t-test
√
Critical Region: t(α/2,df) with df = n – 1 = 24; t(0.05, 24) = 1.711.
Therefore, the critical region is t ≥ 1.711 and t ≤ -1.711
Calculations:
√
Decision: Since the computed value does not fall within the critical region, do not reject H0.
Conclusion: There is no sufficient evidence to conclude that there has been a change in the mean
household size in the Philippines since the last census.
Remember that two samples may either be independent or related. If we want to do a test of
hypothesis for two means, these means must be obtained from two independent samples. To perform
the test of hypothesis for two means, there is a need to know the population variances. If we come
across a situation where the population variances for both groups are unknown, we can assume them
to be equal before we can perform the test. The selection of our statistical test here depends on
whether or not the population variances or standard deviation is known. If it is known, then we
perform the z test and if it is unknown then we can assume that they are equal and perform the t test.
Let us show an example. In this case, assume that the null and alternate hypothesis for this problem is:
Hypothesis:
H0: or
HA: if two-tailed: simply a statement of inequality between the two groups
if one-tailed: specify direction of relationship

or
Known population variances for both groups
̅ ̅
√
Analysis of Quantitative Data
Unkown population variances for both groups but assumed to be equal
 in practice, population variances are rarely known

 when the population variance is unknown but the sample size is satisfactorily large for
both groups (n ≥ 30), the sample variance (s) may be substituted for the population
variance (σ) in the z-test.
̅ ̅
 when the sample size is small for both groups (n < 30), the independent t-test is the
appropriate test statistic
o population variances can be estimated by the sample variances and
o when variances are assumed to be equal, a pooled variances, can be computed
o the test statistic follows the t-distribution with df = n1 + n2 – 2
̅ ̅
Sample Problem:
A study aims to determine the association of salt intake to the blood pressure of persons aged
15 years and over. The mean systolic blood pressure (SBP) of 20 subjects with low salt diet was
compared to that of an equal number of subjects with a high salt diet. The following data were
obtained:
Statistics High salt diet Low salt diet

Mean SBP 138 mmHg 120 mmHg
SD of SBP 11.9 mmHg 12.2 mmHg
Is there a difference between the mean SBP of subjects with high and low salt diets? Let α = 0.05
Solution:
Hypotheses:
H0: µH = µL (There is no difference between the mean SBP of subjects with high and low salt diets.)
HA: µH ≠ μL (There is a difference between the mean SBP of subjects with high and low salt diets.)
Level of significance: α = 0.05
Test statistic:
̅ ̅
We chose this test statistic because both population variances are unknown and both groups
have a small sample size.
Critical region: tα/2,(20+20-2) = t α/2,38
t ≤ -2.021 or t ≥ 2.021
Calculation of the test statistic:
√ (√ )
Statistical Decision: Since 4.72 > 2.021, reject H0

Conclusion: There is a significant difference between the mean SBP of subjects with a high salt diet and
those with a low salt diet.
If we are performing a test of hypothesis for two means that are paired or matched, we must
keep in mind that matching may only be achieved if we use the same subjects in the both samples
such as in “before and after studies” or “brand X and brand Y” comparisons. The pairing of the subjects
must be with respect to any extraneous variables in order to minimize the unwanted effects of these
variables (an example of which is using twins in a study). Paired or matched sampling is used to
overcome the difficulty imposed by extraneous differences between two groups when we are testing
the differences between two means.
Hypotheses:
H0:
where d is the “difference”
Example:
Sample
Before X1 X2 X3 … Xn
After Y1 Y2 Y3 … Yn
Difference d1 d2 d3 … Dn
HA: if two tailed:

if one-tailed: or
Test Statistic:
̅
⁄
√
with df = n – 1
where:
 ̅ = mean difference
 = standard deviation of difference
∑ ⁄
∑
√
 = number of pairs
Critical region:
If two-tailed: |t| ≥ t(α/2, df)
If one-tailed: t ≥ t(α, df) or t ≤ -t(α, df)
Decision rule:
If two-tailed: reject H0 if |t| ≥ t(α/2, df)
If one-tailed: reject H0 if t ≥ t(α, df) or if t ≤ -t(α, df)
Sample Problem:
The women’s weight before and after the 12 weeks of treatment with a very low calorie diet
(VLCD) are shown below. We wish to know if these data provide sufficient evidence to conclude that
the treatment is effective in causing weight reduction in obese women.
Before 117.3 111.4 98.6 104.3 105.4 100.4 81.7 89.5 78.2
After 83.3 85.9 75.8 82.9 82.3 77.7 62.7 69.0 63.0
Hypotheses:
H0: µB = µA (There is no significant difference in the mean weights of women before and after the diet.)
HA: µB > µA (The mean weights of the women before the diet is greater than their mean weights after
the diet.)
Level of significance: α = 0.05 (if level of significance is not mentioned, assume at 5%)
Critical region: t(α,df) = t(0.05,8) = 1.860; Reject H0 if t > 1.860
Test statistic:
̅
⁄
√
with df = 9 – 1 = 8
Before 117.3 111.4 98.6 104.3 105.4 100.4 81.7 89.5 78.2
After 83.3 85.9 75.8 82.9 82.3 77.7 62.7 69.0 63.0
Difference 34.0 25.5 22.8 21.4 23.1 22.7 19.0 20.5 15.5
Sum of differences: 22.70
Standard deviation of differences: 5.10
⁄
√
Statistical Decision: Since 13.35 > 1.86, reject H0

Conclusion: The mean weight of the women before the diet is greater than their mean weights after
the diet. We therefore conclude that the treatment is effective in causing weight reduction in obese
women.

o



 To do a 2-tailed z-test, use the following syntax:

=2*MIN(Z.TEST(array,x,sigma),1-Z.TEST(array,x,sigma))
o Use NORM.S.INV (NORMSINV for Excel 2007 and earlier) on the Z.TEST result to
obtain the value of the test statistic.
 T.TEST – returns the probability associated with a Student’s t-test (TTEST for Excel 2007 and
earlier)
o Syntax: =T.TEST(array1,array2,tails,type)
 array1 – the first data set
 array 2 – the second data set
 tails – If tails = 1, uses the one-tailed t distribution. If tails = 2, uses the two-
tailed t distribution
 type – the kind of t-test to perform.
 If type = 1, performs a paired t-test
 If type = 2, performs an independent t-test (variances assumed
equal or homoscedastic)
 If type = 3, performs an independent t-test (unequal variances or
heteroscedastic)
o Use T.INV.2T (TINV for Excel 2007 and earlier) on the T.TEST result to obtain the
value of the test statistic.
NAME:__________________________________________________________ DATE:______________
For each of the following problems, carry out the steps in hypothesis testing at the designated level of
significance. When possible, use Microsoft Excel to calculate the value of the test statistic. Write the
syntax to substitute for calculations by hand. Submit a copy of your spreadsheet to your instructor.
1. Suppose it is known that the IQ scores of a certain population of adults are approximately normally
distributed with a standard deviation of 15. A simple random sample of 25 adults drawn from this
population had a mean IQ score of 105. On the basis of these data can we conclude that the mean
IQ score for the population is not 100? Let the probability of committing a type I error be 0.05.
a. H0 (1 point): ____________________________________________________________________
b. HA (1 points): _________________________________________________________________

__________________________________________________________

2. A researcher conducted a study to examine prospectively collected data on gentamicin
pharmacokinetics in three populations over 18 years of age: patients with acute leukemia, patients
with other nonleukemic malignancies, and patients with no underlying malignancy or
pathophysiology other than renal impairment known to alter gentamicin pharmacokinetics.
Among other statistics reported by the investigators were a mean initial calculated creatinine
clearance value of 59.1 with a standard deviation of 25.6 in a sample of 211 patients with
malignancies other than leukemia. We wish to know if we may conclude that the mean for a
population of similar subjects is less than 60. Let α = 0.10.
a. H0 (1 point): ____________________________________________________________________
b. HA (1 points): _________________________________________________________________

__________________________________________________________

3. Does sensory deprivation have an effect on a person’s alpha-wave frequency? Twenty volunteer
subjects were randomly divided into two groups. Subjects in group A were subjected to a 10-day
period of sensory deprivation, while subjects in group B served as controls. At the end of the
experimental period the alpha-wave frequency components of the subjects’
electroencephalograms were measured. The results were as follows:
Group A: 10.2, 9.5, 10.1, 10.0, 9.8, 10.9, 11.4, 10.8, 9.7, 10.4
Group B: 11.0, 11.2, 10.1, 11.4, 11.7, 11.2, 10.8, 11.6, 10.9, 10.9
a. H0 (1 point): ____________________________________________________________________
b. HA (1 points): _________________________________________________________________

__________________________________________________________

4. A researcher conducted a study to test the hypotheses that weight loss in apneic patients results in
decreases in upper airway critical pressure (Pcrit) and that these decreases are associated with
reductions in apnea severity. The study participants were patients referred to the Johns Hopkins
Sleep Disorder Center and in whom obstructive sleep apnea was newly diagnosed. Patients were
invited to participate in either a weight loss program (experimental group) or a “usual care”
program (control group). Among the data collected during the course of the study were the
following before and after Pcrit (cm H2O) scores for the weight-loss participants:
Before -2.3 5.4 4.1 12.5 0.4 -0.6 2.7 2.7 -0.3 3.1 4.9 8.9 -1.5
After -6.3 0.2 -5.1 6.6 -6.8 -6.9 -2.0 -6.6 -5.2 3.5 2.2 -1.5 -3.2
May we conclude on the basis of these data that the weight-loss program was effective in
decreasing upper airway Pcrit? Let α = 0.10
a. H0 (1 point): ____________________________________________________________________
b. HA (1 points): _________________________________________________________________

__________________________________________________________

It is nearly impossible to examine all members of a population in order to come up with
population parameters, which is why random sampling and sampling distributions are essential
statistical analysis. Whereas the statistical analysis to be performed is highly dependent on the study
design, it is also reliant on the sample size. A statistical test can be deemed powerful if the sample size
is sufficient.
 Sample Size – the number of elementary units that must be examined in order to arrive at data
that may be truly representative of the population. The researcher’s discretion must be based
on pragmatism, that is, a sample size that is too large may lead to excessive cost, while a
sample size that is too small may lead to erroneous results.
 Power – the power of any statistical test is a measure of the sensitivity of the statistical test,
whether it can really determine the difference between two means, two proportions, etc. It is
calculated as Power = 1 – β, where beta is Type II error (error of not rejecting a null hypothesis
that is false). Most often, power is assigned at 80% or 0.80 in the same manner that the
significance level (α) is usually 5% or 0.05.
In research proposals, the study design must include the calculated sample size, assigned
significance level (α) and occasionally, the power. The method of sample size calculation depends on
the type of statistical analysis to be used, the type of variable to be examined, parameters or published
data, and even the actual study design. The sample size can be calculated through a (1) formula, (2)
published sample size tables or (3) software.
Based on Whitley, E. and Ball, J. (2002) Statistics review 4: Sample size calculations. Critical Care, Vol.
6(4): 335-341. Available online: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC137461/
 For Hypothesis Testing for the Difference Between Two Means
Where: n = number of subjects required in each group

d = standardized difference = (the target difference is assigned by
the researcher and the standard deviation is based on previous similar studies)
cp,power = a constant defined by the values chosen for the P value and power
P value Power (%)
50 80 90 95
0.05 3.8 7.9 10.5 13.0
0.01 6.6 11.7 14.9 17.8
*Table from Whitley and Ball (2002).
 For Hypothesis Testing for the Difference Between Two Proportions

[ ]
Where: n = number of subjects required in each group

cp,power = a constant defined by the values chosen for the P value and power
The use of published sample size tables is especially important in surveys or observational
studies. The use of sample size tables may be based on the total population, significance level, p value
or standardized difference assigned by the researcher. The World Health Organization has published a
public document entitled “Tables of Minumum Sample Size” available for download at
http://whqlibdoc.who.int/publications/9241544058_%28p23-p80%29.pdf
Sample size calculation with power is usually performed by researchers using statistical
software such as Stata (licensed) and the free, java-run, online and downloadable Piface
(http://homepage.cs.uiowa.edu/~rlenth/Power/).
Sample Size and Power
Whichever method a student researcher prefers to use for sample size calculation, it is
important to remember that the calculated sample size is usually a minimum sample size and that it
should not limit the researcher from using more than the calculated sample size. Typically, in the case
of surveys on human subjects, an additional number of samples may be studied to account for possible
non-response. As long as the sample size is sufficient and is at least equal to the sample size
calculated, then the researcher can be confident in the implementation of his/her study design.
In this section, we are interested in (1) establishing association between two qualitative
variables; (2) establishing correlation between two quantitative variables; and (3) predict whether a
relationship exists between two or more variables. This section will introduce you to the chi square
test of association to answer our first interest; the correlation coefficient for the second; and the
regression analysis for the last interest.
We want to investigate the relationships of our variables because we want to know if they are
associated or correlated with each other. We want to determine their strength of association and we
want to predict a given outcome based on the relationship of the variables.
The choice of our statistical tool depends on the number of variables we are testing, the type
of variables we are investigating, and what the intention of the researcher is. There are two
approaches that we can use to show relationships between variables: the graphical approach and the
statistical test. In the graphical approach, we are interested in demonstrating the nature of the
relationship between the variables. Here we can use several graphical representations to show the
relationship of the variables. For example, when we want to show the relationship between two
variables that are both qualitative, we can construct a comparative bar graph. On the other hand, if we
want to show the relationship of two variables that are both quantitative, we can construct a scatter
plot diagram. There are endless possibilities of graphical approaches that we can use to depict the
relationship of our variables. Usually, the graphical approach is merely a descriptive manner of showing
the relationship between variables. It does not put one in position to make generalizations based on
the graphs showed. The graphical approach is oftentimes used as the first step or guide in choosing
the best statistical tool to analyze the data.
In this section, we will discuss the chi-square test of independence (also called the chi-square
test of association) as well as the chi-square test of homogeneity. The test of independence is useful for
testing the relationship between two categorical variables while the test of homogeneity allows the
investigator to determine whether several groups of samples are homogeneous with respect to a
particular classification. Both tests use the same mathematical concepts and are carried out by
following the steps in hypothesis testing. However, the first step in performing a chi-square test is to
construct the contingency table.
What is a contingency table? A contingency table is a cross-tabulation consisting of r number

of rows and c number of columns and gives frequency of observations that fall under each category.
The table below is an example of a contingency table. In this table, note that the O11 is the observed
frequency in the C1 category of the variable of interest of the sample stratum 1. The n1 indicated on the
table signifies the row totals while the n.1 is the column total.
Samples Categories Total

C1 C2 C3
S1 O11 O12 O1n n1
S2 O21 O22 O2n n2
Sn On1 On2 Onn nn
Total n.1 n.2 n.n n.. = n
In the contingency table, take note that the categories are assumed to be exhaustive as well as
exclusive. The samples of the strata are presumed to be independent of one another. If we are to
apply the chi-square test to test a hypothesis, we must follow the steps in hypothesis testing as we
have done previously. We still start the hypothesis testing by stating the null and alternative
hypotheses. In this case, let us look at an example of how the null and the alternative hypotheses are
formulated.
H0: The proportion of elements falling in each category of the variable of interest is the same for all
strata.
HA: There are differences between strata in the proportion of elements falling in each category of the
variable.
After the hypotheses have been stated, the next step is to set the level of significance. Then,
identify the appropriate test statistic. In this case, let us assume that we are going to use the chi-square
test as our test statistic. Take note that the shape of the chi-square distribution depends on the
number of degrees of freedom (df) where df = (r – 1)(c – 1). We then proceed to determining the
critical region. This is determined by the number of the degrees of freedom and the level of
significance. The chi-square values are obtained from the chi-square distribution table listed as Table E
at the end of this module.
The test statistic for the chi-square test is:
∑[ ]
with (r – 1)(c – 1) degrees of freedom
where:
 Oi is the observed frequency for the ith category; and
 Ei is the observed frequency given H0 is true for the ith category
 The quantity X2 is a measure of the extent to which pairs of observed and expected
frequencies agree.
 When there is close agreement between observed and expected frequencies, it is small,
and when the agreement is poor, it is large.
 Only a sufficiently large value of X2 will cause rejection of the H0.
Investigating Relationships Between Variables
The Decision Rule
 The computed value of X2 is compared with the tabulated value of χ2 with k – r degrees of
freedom.
 The decision rule is: Reject H0 if X2 is greater than or equal to the tabulated χ 2 for the
chosen value of α.
We have learned in lesson 5 (basic concepts in probability) that if two events are independent,
the probability of their joint occurrence is equal to the product of their individual probabilities. For
example, under the assumption of independence, we calculate the probability that one of the o
observations represented in Table 10-4 will be counted in row 1 and column 1 (that is, cell S1C1) of the
table by multiplying the probability that the subject will be counted in row 1 by the probability that
the subject will be counted in column 1. Using the notation of the table, the desired calculation is
( )( )
To obtain the expected frequency, we multiply this probability by the total number of subjects,
n. Therefore, the expected frequency for cell S1C1 is given by
( )( )
Since the n in one of the denominators cancels into numerator n, this expression reduces to
In general, to obtain the expected frequency for a given cell, multiply the total of the row in
which the cell is located by the total of the column in which the cell is located and divide the product
by the grand total.
What is the rationale for the computation of expected frequencies?
The rationale on the computation of expected frequencies lies on the assumption that
populations are represented in a contingency table that is homogeneous with respect to the variable
of interest we are studying. It also depends on the expected frequency in the ith sample that is
obtained by multiplying the pooled estimate by the total number of subjects in the sample that we
have. Remember that the data in a contingency table is only applicable if the expected frequencies are
sufficiently large. In a 2 x 2 table the requirement is for all expected frequencies to be greater than or
equal to 5. If it does not satisfy the requirement, then the chi-square test cannot be used. Instead, one
can resort to the use of the Fisher’s exact probability test. In the case of larger tables, the requirement
is for all expected frequencies to be greater than or equal to 1 and not more than 20% of the cells
should have expected frequencies less than 5. Again, if the condition is not met the researcher must
merge the adjacent categories to meet the condition before the chi-square test can be applied.
The tabulated values of the chi-square distribution are provided in Table E at the end of this
module. Note that that in the chi-square test, α is not divided by two. To obtain the desired critical
value, simply align the value of α that you have set with its corresponding degrees of freedom.
Alternatively, you may use Microsoft Excel 2010’s CHISQ.INV.RT function which returns the inverse of
the right-tailed probability of the chi-square distribution. The appropriate syntax is
=CHISQ.INV.RT(probability,deg_freedom) where probability represents α and deg_freedom represents
the degrees of freedom given by (r – 1)(c – 1). For those using earlier versions of Excel, the analogous
function is CHIINV which uses similar arguments.
The most frequently used chi-square test, it tests the null hypothesis that two criteria of
classification, when applied to the same set of entities, are independent.
 Example: If the socioeconomic status and area of residence of the inhabitants of a certain
city are independent, we would expect to find the same proportion of families in the low,
medium, and high socioeconomic groups in all areas of the city.
 We say that two criteria of classification are independent if the distribution of one criterion
is the same no matter what the distribution of the other criterion.
Sample Problem
The purpose of a study by Vermund et al. (1991) was to investigate the hypothesis that HIV-
infected women who are also infected with human papillomavirus (HPV), detected by molecular
hybridization, are more likely to have cervical cytologic abnormalities than are women with only one or
neither virus. The data shown in Table 9.2 were reported by the investigators. We wish to know if we
may conclude that there is a relationship between HPV status and stage of HIV infection.
HIV
HPV Total
Seropositive, Symptomatic Seropositive, Asymptomatic Seronegative
Positive 23 4 10 37
Negative 10 14 35 59
Total 33 18 45 96
SOURCE: Sten H. Vermund, Karen F. Kelley, Robert S. Klein, Anat R. Feingold, Klaus Schreiber, Gary Munk, and Rubert D. Burk,
“High Risk of Human Papillomavirus Infection and Cervical Squamous Intraepithelial Lesions Among Women with
Symptomatic Human Immunodeficiency Virus Infection,” American Journal of Obstetrics and Gynecology, 165 (1991), 392 – 400,
as printed in Biostatistics: A Foundation for Analysis in The Health Sciences 6e by Wayne W. Daniel (1995)
Hypotheses:
H0: HPV status and age of HIV infection are independent
HA: The two variables are not independent
Significance Level: α = 0.05
Test Statistic:
∑[ ]
With (r – 1)(c – 1) degrees of freedom = (2 – 1)(3 – 1) = 2
Critical Region:
X2 ≥ χ2α, df = (r-1)(c-1)
Reject H0 if the computed value of X2 is greater than or equal to 5.991
Statistical Decision: Reject H0 since 20.60081 > 5.991

Conclusion: We conclude that H0 is false, and that there is a relationship between HPV status and stage
of HIV infection.
The chi-square test of homogeneity finds out whether two or more populations have the same
proportions for the different categories of another variable. When only two populations are
considered and the variable of interest only has two categories, it can be used interchangeably with
the z test for two proportions. If this is the case, the data is usually casted in a 2x2 contingency table.
In the chi-square test for independence, the total sample is assumed to have been drawn
before the observations were classified according to the two criteria of classification. That is, the
number of observations falling into each cell was determined after the sample was drawn.
Consequently, the row and column totals are chance quantities not under the control of the
researcher. We may think of the sample drawn under these conditions as a single sample drawn from a
single population. However, there may be instances when either the row or column totals are under
control of the researcher. This means that the researcher may specify that independent samples be
drawn from each of several populations. In the case of the test of homogeneity, one set of the marginal
totals is said to be fixed while the other set, corresponding to the criterion of classification applied to
the samples, is said to be random. The test of independence and test of homogeneity not only involve
different sampling methods but also lead to different questions and null hypotheses. The test of
independence is concerned with the question: “Are two criteria of classification independent?” while the
test of homogeneity is concerned with the question: “Are the samples drawn from populations that are
homogeneous with respect to some criterion of classification?” In the test of homogeneity, the null
hypothesis states that the samples are drawn from the same population. Despite these differences in
concept and sampling method, the two tests are mathematically identical as we will demonstrate in
the following example:
Sample Problem:
Kodama et al. studied the relationship between age and several prognostic factors in
squamous cell carcinoma of the cervix. Among the data collected were the frequencies of histologic
cell types in four age groups. The results are shown in Table 10-6. We want to know if the populations
represented by the four age-group samples are not homogeneous with respect to cell type.
Hypotheses:
H0: The four populations are homogeneous with respect to cell type.
HA: The four populations are not homogeneous with respect to cell type.
Significance Level: α = 0.05
Test Statistic:
∑[ ]
With (r – 1)(c – 1) degrees of freedom = (4 – 1)(3 – 1) = 6
Critical Region:
X2 ≥ χ2α, df = (r-1)(c-1)
Reject H0 if the computed value of X2 is greater than or equal to 12.592
Statistical Decision: We are unable to reject H0 since 4.444 < 12.592

Conclusion: We conclude that the four populations may be homogeneous with respect to cell type.
3. Arrange the data into arrays where the actual or observed frequencies are separated from
the expected frequencies.
4. If using Microsoft Excel 2010, use the CHISQ.TEST function with the syntax
=CHISQ.TEST(actual_range,expected_range). The analogous function for earlier versions of
excel is CHITEST with the same form of arguments.
5. After typing “=CHISQ.TEST(“ [without quotation marks] into an empty cell, select the range
of actual frequencies, type “,” then do the same for the expected frequencies. Finish the
syntax by typing “)” and press Enter.
6. The CHISQ.TEST function returns the probability value (p). You may directly compare this to
α to make your statistical decision. To get the value of the X2 statistic, use the CHISQ.INV.RT
function (CHIINV for Excel 2007 and earlier).
This test measures the strength and direction of the relationship between two or more
variables. No distinction between the two variables like on how they vary jointly is determined by this
test. This test uses a correlation coefficient to show both the strength and the direction of the
relationship. If the data is parametric, the Pearson’s correlation coefficient is normally used while if the
data is non-parametric, the Spearman’s rank correlation coefficient should be used.
What are the assumptions of the Pearson’s correlation coefficient?
The Pearson’s correlation coefficient assumes that for each value of the variable X, the
corresponding sub-population of values for the variable Y is normally distributed. It also assumes that
for each value of the variable Y, the corresponding sub-population of values for the variable X is
normally distributed. The joint distribution of the variables X and Y is also assumed to be normally
distributed.
The value of our Pearson’s correlation coefficient r or rho ranges from -1 to +1. The positive
and negative signs show the direction of the relationship. The minimum absolute value is 0 while the
maximum absolute value is 1. If r or the rho value is 0, it means that no relationship exists between the
variables of interest. Meanwhile, if the value is nearer the absolute value of 1, then the stronger the
relationship that exists between the variables of interest. A positive sign indicates a direct relationship
while a negative sign indicates an inverse relationship. For example, r=0.72 and r=-0.72 have the same
magnitude or strength of relationship. Both values only differ in the direction of the relationship. The
Pearson’s correlation computational formula is shown below:
∑ ∑ ∑
√ ∑ ∑ ∑ ∑
The strength and direction of the relationship are expressed by means of a correlation
coefficient which is mathematically defined as:
S XY SCP
r 
S X SY ( SSX )(SSY )
The sum of cross products of deviations
( X i )( Yi )
SCP   ( X i  X )(Yi  Y )   X i Yi 
n
The sum of squared deviations for X
( X i ) 2
SSX   ( X i  X )   X 2
i
2
n
The sum of squared deviations for Y
( Yi ) 2
SSY   (Yi  Y )   Yi 
2 2
n
The Pearson’s correlation coefficient r
SSX   X 
 X  2 i
2
i
n
SCP
r Y ) ( 2
SSY   Y  2 i
( SSX )( SSY ) n
i
 X Y   n
X Y i i
SCP  i i
This test depends on a dependent variable being affected by an error-free independent

variable. This test is primarily concerned with using the relationship between two variables for the
purpose of predicting one variable based on the knowledge of the other. Remember that correlation
analysis is primarily concerned with discovering whether or not a relationship exists in the first place,
and then specifying the strength and direction of the relationship. In regression analysis, we can still
determine whether a relationship exists, specify the strength and the direction of the relationship, and
in addition, predict the outcome of the variable with the knowledge of the other variable. In other
literatures, this test is also referred to as the least squares method. Regression analysis can be
performed either as a simple regression analysis or as a multiple regression analysis. In simple
regression analysis, we have a single Y and a single X, whereas in multiple regression analysis, we have
a single Y and multiple X variables. The simple linear regression equation is given below:
Yˆ  b0  b1 X
Where X = given data
b0 = intercept of the regression line
b1 = slope of the regression line
Graphically, this is expressed in the figure below:
In regression analysis, one can also perform a linear regression analysis or a non-linear
regression analysis. In linear regression analysis, we assume that for every incremental change in Y,
there is a corresponding incremental change in X whereas in non-linear regression analysis, we assume
that for every incremental change in Y, there is no corresponding incremental change in X, or that the
change in X may not be a linear increment change as expressed by our Y variable.
The coefficient of regression can be calculated using the formulas indicated below:
SCP
b1  b0  Y  b1 X
SSX
SCP   ( X i  X )(Yi  Y )   X iYi 

(  X )(Y )
i i
( X i ) 2
SSX   ( X i  X )   X  2
i
2
b1 
 ( X  X )(Y  Y )
i i
(X  X ) i
2
The coefficient of determination is a measure that is commonly used to describe how well the
sample regression line fits the observed data. This is expressed in the formula below:
R2 
SSR

b1
2
 ( X i  X ) 2
SST  i
(Y  Y ) 2
The range of the coefficient of determination is: 0 < R2 < 1. A value of 0 indicates the poorest
fit while a value closest to 1 indicates the best fit for our regression model.
This test allows us to determine whether significant differences exist between three or more
groups. In this section, we will only cover the one-way ANOVA test. If we are to do a one-way ANOVA,
there should be more than two levels of our single independent variable. Remember that in
performing this test, the variables need to be grouped and the variable must consist of a number of
levels. After grouping the independent variables, we can now set the level of significance, sample size,
and hypotheses. The null hypothesis here indicates that any observed difference between three or
more groups will be attributable to random sampling errors whereas in the alternate hypothesis it
indicates that the difference is not attributable to random sampling errors.
The assumptions in performing a one-way ANOVA are the following:
1. Samples should be independent.

2. Each of the k populations should be normal
3. The k samples should have equal variances
To evaluate the Fobserved with the F distribution, we need to remember that the distribution of F
ratios is not normally distributed as it is positively skewed. The F distribution depends on the number
of degrees of freedom in the numerator and the denominator. The Fcritical is the F value that must be
equalled or exceeded to classify a difference among the group means as statistically significant. The
Fcritical depends on the degrees of freedom in the numerator and the denominator and the chosen level
of significance, all of which are listed in Table F at the end of this module.
If we want to compute for the ANOVA F statistic manually, an example of performing the test is
shown on the next page:
WITHIN BETWEEN
difference: difference
group data - group mean group mean - overall mean
data group mean plain squared plain squared
5.3 1 6.00 -0.70 0.490 -0.4 0.194
6.0 1 6.00 0.00 0.000 -0.4 0.194
6.7 1 6.00 0.70 0.490 -0.4 0.194
5.5 2 5.95 -0.45 0.203 -0.5 0.240
6.2 2 5.95 0.25 0.063 -0.5 0.240
6.4 2 5.95 0.45 0.203 -0.5 0.240
5.7 2 5.95 -0.25 0.063 -0.5 0.240
7.5 3 7.53 -0.03 0.001 1.1 1.188
7.2 3 7.53 -0.33 0.109 1.1 1.188
7.9 3 7.53 0.37 0.137 1.1 1.188
TOTAL 1.757 5.106
TOTAL/df 0.25095714 2.55275
Overall mean: 6.44 F = 2.55275/0.25095714 = 10.172
Once the ANOVA indicates that the groups do not all have the same means, we can compare
them two by two by performing a multiple comparisons test or post hoc test. A post hoc test
determines which specific pairs of means are significantly different. It is a follow up test that is only
performed once we have determined a significant ANOVA. The post hoc test is like a collection of little
t-tests but with control over the overall type I error. There are different types of post hoc tests that one
can use. Examples include the Bonferroni procedure, Duncan multiple range test, Dunnett’s multiple
comparison test, Newman–Keuls test, Scheffe’s test, and Tukey’s test. No single test is found to be best
in all situations, and a major difference between them lies in the manner in which they control the
increase in Type I error due to multiple testing. Remember that in hypothesis testing, using the same
statistical test repeatedly results to an increase in type I error. The post hoc test may not have as much
power as the omnibus test, which is that of the ANOVA. The purpose of this test is just to identify the
locus of the effect or which means are significantly different. Among the post hoc tests, the Tukey’s
HSD (honestly significant difference) test is the commonly used. The Tukey’s HSD test uses the formula
below:
HSD  q MS within
n
The HSD determines the magnitude of the mean difference that must exist to claim that the
levels are significantly different. The q is the studentized ranged statistic that depends on the number
of levels to be compared and the dfwithin and the level of significance. The q value is obtained from a
table. The Mean Square within is taken from the ANOVA summary table while n is the number of
subjects in each group.
SOURCE: Biostatistics: A Foundation for Analysis in the Health Sciences 6e by Wayne W. Daniel (1995)
SOURCE: Biostatistics: A Foundation for Analysis in the Health Sciences 6e by Wayne W. Daniel (1995)
SOURCE: Biostatistics: A Methodology for the Heatlh Sciences 2e by van Belle et al. (2004)
Acacio-Claro, PJ (2008). Lectures in Biostatistics. University of the Philippines Manila, Manila.
Albert, J. (1996, November 18). The Subjective Interpretation of Probability. Retrieved October 30,
2012, from http://www-math.bgsu.edu/~albert/m115/probability/subject.html
Albert, J. (1996, November 18). The Relative Frequency Interpretation of Probability. Retrieved
October 30, 2012, from http://www-math.bgsu.edu/~albert/m115/probability/relfeq.html
Arsham, H. (n.d.). Combinatorial Mathematics: How to Count Without Counting. Retrieved October
30, 2012, from http://home.ubalt.edu/ntsbarsh/Business-stat/otherapplets/ComCount.htm
Capistrano, TA (n.d.). Course Notes in Stat 114 (Descriptive Statistics). University of the Philippines
Diliman, Quezon City.
Centers for Disease Control and Prevention. (1992). Principles of Epidemiology.
Chernick, MR & Friis, RH (2003). Introductory Biostatistics for the Health Sciences. Hoboken, New Jersey:
John Wiley & Sons, Inc.
Daniel, W. (1995). Biostatistics: A Foundation for Analysis in the Health Sciences (6th ed.). Hoboken, New
Jersey: John Wiley & Sons, Inc.
Doug, J. (2010). Class Notes in Medical Statistics (Fall 2010): Summary Statistics. University of Illinois
at Urbana-Champaign.
Fairfax County Department of Neighborhood and Community Services. (2012, April). Overview of
Sampling Procedures. Fairfax, Virginia, United States of America. Retrieved November 2, 2012,
from http://www.fairfaxcounty.gov/demogrph/pdf/samplingprocedures.pdf
Kuzma, JW & Bohnenblust, SE (2005). Basic Statistics for the Health Sciences International Ed.
McGraw-Hill (Asia): Philippines.
Le, CT (2003). Introductory Biostatistics. Hoboken, New Jersey: John Wiley & Sons, Inc.
Mapua, CA (2010). Lectures in Molecular Epidemiology and Biostatistics. St. Luke's College of
Medicine, Quezon City.
Mendoza, OM; Borja, MP; Sevilla, TL; Ancheta, CA; Saniel, OP; Sarol Jr, JN and Lozano, JP (2009).
Foundations of Statistical Analysis for the Health Sciences. Department of Epidemiology and
Biostatistics, College of Public Health, University of the Philippines Manila: Manila.
National Institute of Standards and Technology. (n.d.). Percentiles. Retrieved October 30, 2012, from
Engineering Statistics Handbook: http://itl.nist.gov/div898/handbook/prc/section2/prc252.htm
Samuels, ML & Witmer, JA (1999). Statistics for the Life Sciences 2nd Ed. Prentice-Hall, Inc.: New Jersey.
Schröder, B. (n.d.). Slides in Sample Spaces and Events. Louisiana Tech University, College of
Engineering and Science.
van Belle, G; Fisher, LD; Heagerty, PJ; & Lumley, T (2004). Biostatistics: A Methodology for the Health
Sciences (2nd ed.). Hoboken, New Jersey: John Wiley & Sons, Inc.

File Course Module in Biostatistics

Uploaded by

Copyright:

Available Formats

File Course Module in Biostatistics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

File Course Module in Biostatistics

Uploaded by

Copyright:

Available Formats

Biostatistics is coined from two words, bios, which means life, and statistics, which means the

Type of Data Method of

Reason for choosing the best

Prevalence of drug and

Average heavy metal

It is systematic because it is akin to a problem-solving technique that follows the scientific

1. Identification of the research problem

Biologists investigate research problems related to their specific expertise or field of

1. The Null Hypothesis

2. The Alternative Hypothesis

 Constant versus Variable

 General Types of Variables

 Nominal Level Data

 Ordinal Level Data

 Interval Level Data

o Numerical observations of a quantitative variable, in which 0 (zero) is not absolute

 Ratio Level Data

Research designs may be generally divided into observational and experimental.

 Basic components of an experiment:

The experimental unit may be subjected to various conditions (independent variable or

 Variation due to the effects of the independent variable/s

An experimental research design must be formulated in a manner wherein variation in

a. Blood glucose level in mg/100ml constant variable _________________________

b. Blood type constant variable _________________________

c. Avogadro’s number constant variable _________________________

d. Dialect constant variable _________________________

e. Cases of pneumonia constant variable _________________________

f. Nutritional status constant variable _________________________

2. Specify the level of measurement of the given variables. (10 points)

a. Patient Name ____________________________________

b. Age in years and months ____________________________________

d. Digital Systolic Blood Pressure ____________________________________

e. History of Surgery ____________________________________

f. Pain Scale ____________________________________

h. Students Enrolled ____________________________________

i. Distance Traveled ____________________________________

j. Calories per Serving ____________________________________

3. Answer the questions related to the given research topics.

i. What general type of research design is described above?

ii. Independent Variable __________________________________________________

iii. Dependent Variable ___________________________________________________

iv. Variable Type of the Dependent Variable and Level of Measurement:

v. Data Acquisition Techniques Employed in the Design:

i. Formulate an appropriate alternative hypothesis ___________________________

ii. What general type of research design is described above?

iii. Dependent Variable ___________________________________________________

iv. Variable Type of the Dependent Variable and Level of Measurement:

Basic methods of data presentation:

58 82 151 121 100 68

5.9 6.1 6.1 5.7 5.9 6.0

a) What is your calculated range?

Life Expectancy Life Expectancy

Group Number of Species

City Dinorado (%) Sinandomeng (%) Thai Jasmine (%)

Month Ave. Rainfall (cm)

 Convey information regarding the average value of a set of values.

1. To calculate the numerator, sum the individual observations

2. For the denominator, count the number of observations: n = 10

 The mean or average of a set of data measured on a logarithmic scale.