STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 1

February 22, 2009

Chapter 2: Descriptive Statistics

In order to determine what a set of data has to tell us, we must extract the information in the data.
Typically a first step in a statistical analysis of data is to simply describe the data using descriptive
statistics and look at the data using appropriate graphing techniques.
In this chapter we will introduce some basic descriptive statistics (means, medians, standard deviations,
etc.) and some common graphical techniques for looking at the data.
We shall use SAS to obtain the desired statistics and graphs.
To illustrate ideas, we return to our previous example with pelican eggs:

Example. The PCB concentration (ppm) and thickness (mm) of 65 Anacapa pelican eggs (Risebrough
1972) were measured and the values are listed below:

452 0.14 204 0.28 261 0.34 199 0.46

184 0.19 89 0.28 150 0.34 216 0.46
315 0.20 320 0.28 143 0.35 236 0.47
115 0.20 138 0.29 229 0.35 206 0.49
139 0.21 198 0.29 132 0.36 237 0.49
177 0.22 191 0.29 175 0.36
214 0.22 193 0.29 173 0.36
356 0.22 316 0.29 220 0.37
246 0.23 265 0.29 212 0.37
177 0.23 122 0.30 236 0.37
289 0.23 305 0.30 119 0.39
166 0.23 203 0.30 144 0.39
175 0.24 396 0.30 147 0.39
296 0.25 250 0.30 171 0.40
205 0.25 230 0.30 232 0.41
260 0.26 214 0.30 216 0.41
188 0.26 256 0.31 164 0.42
208 0.26 46 0.31 185 0.42
324 0.26 204 0.32 216 0.42
109 0.27 218 0.34 87 0.44

Measures of Location

What is the central or typical value for the eggshell thicknesses? What we would like is a measure of
central location. If we plot the eggshell thicknesses on a number line, what is the central location of the
thicknesses? The three most popular statistics for measuring location are the sample mean, the median,
and the mode. The mode is simply the observation that occurs most frequently.

The Sample Mean. If we let denote the average pelican egg thickness for the entire population, we can
estimate using the sample mean (or arithmetic mean) which is simply the average of the observations
in the data set. Letting y1 , y2 , . . . , yn denote the data values for the observed egg thicknesses, the sample
mean is defined and denoted as: n
1X
y = yi . (1)
n i=1
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 2

P
The symbol simply means to add up the values. For this particular example we have:
n
1X
y = yi
n i=1
1
= [y1 + y2 + y3 + + y65 ]
65
1
= [0.14 + 0.19 + 0.20 + + 0.49]
65
= 0.316

Computing y by hand is quite tedious for a sample size of n = 65. Instead we shall illustrate how to use
SAS to compute this for us in a moment. One way to think of the sample mean is that it is the center
of gravity of all the data points. If the points are lined up on a number-line seesaw, the seesaw would
balance at exactly the sample mean.

The Sample Median. Another very commonly used measure of location is the sample median. The
median is simply the middle value of the data after it has been arrange in ascending order. In the pelican
egg data set above note that the observations are listed in ascending order according to eggshell thickness.
If there are an odd number n of observations, then there is a unique middle value, the (n + 1)/2th value.
In the pelican example, n = 65 is odd, so the median is the (65 + 1)/2 = 33rd value which is 0.30. If the
sample size n is even, then the median is taken as the average of the n/2 and the n/2 + 1 values, i.e. the
two middle values.
Note that the sample median is an estimator of the population median which is the median value for the
entire population.

Question: Which measure of location, the mean or median, should we use in practice? The answer
depends on the type of data set under investigation. If the distribution of data points are symmetric about
the mean, then the mean and median will be equal. A symmetric distribution is one where the relative
position of data points are the same on each side of the mean. If the distribution of observations for the
entire population are symmetric about the population mean, then samples obtained from the population
should be approximately symmetric about the sample mean. In practice, if the population is symmetric
about the mean, the sample mean and median will be close in value but not necessarily exactly equal to
each other. This will become clearer once we plot some data sets later.
One often encounters skewed distributions with particular types of data, such as toxin levels in soil or
water, home prices, or salaries. For strongly skewed distributions, the median tends to be a better measure
of location than the mean. As an extreme example, consider the following hypothetical example:

Example. Suppose a small neighborhood consists of 5 modest homes and one very large mansion. The
home prices (in thousands of dollars) are:

112, 115, 116, 120, 122, 564.

The mean home price is $191,500 which is clearly not representative of the neighborhood because its value
is greater than all the home prices except the mansion. The median for this example is (116+120)/2 = 118
(since we have an even number of observations). Clearly $118,000 is a better measure of central location
than the mean in this example.
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 3

Measures of Spread
In order to adequately describe a data set, the mean (or median) is generally not sufficient by itself. To see
why, consider the following simple example: Suppose the annual amount of rain in a sample of 10 United
States cities was obtained in the years 1900 and 2000 yielding the following values (in inches):

1900 : 34 38 39 40 40 40 41 41 43 45
2000 : 21 21 33 33 36 38 39 50 60 70

The mean amount of rainfall in the 10 cities for both years is 40.1 inches. However, the rainfall amounts
in 2000 are much more spread out than in 1900. This means that in the year 2000 one is much more
likely to have seen droughts and floods than in 1900, even though the mean amount of rainfall is the
same for the two years. This example illustrates that the mean (or average) is not sufficient by itself
to describe a population (or sample). What is needed is a statistic for measuring the spread in the
data. A simple measure of spread is the range which is the difference between the largest and smallest
observation. However, the most commonly used measure of spread is the variance and its square root, the
standard deviation. Given a set of observations y1 , y2 , . . . , yn with sample mean y, the idea is to look at
the deviations about the mean:
(yi y).
However, the deviations always average out to be zero. Instead, we look at the average of the squared
deviations and define the sample variance, denoted s2 as follows:
Pn
2 i=1 (yi y)2
Sample Variance : s = . (2)
n1
The (positive) square root of the variance is called the sample standard deviation:
v
u n
uX
Sample Standard Deviation : s= s = t (yi y)2 /(n 1).
2 (3)
i=1

To illustrate the computation, we compute the sample variance and standard deviation for the two sets of
rainfalls. Letting s21 and s22 denote the sample variances for 1900 and 2000 we compute

s21 = {(34 40.1)2 + (38 40.1)2 + (39 40.1)2 + (40 40.1)2 + (40 40.1)2 +
(40 40.1)2 + (41 40.1)2 + (41 40.1)2 + (43 40.1)2 + (45 40.1)2 }/(10 1)
= 8.54

and

s22 = {(21 40.1)2 + (21 40.1)2 + (33 40.1)2 + (33 40.1)2 + (36 40.1)2 +
(38 40.1)2 + (39 40.1)2 + (50 40.1)2 + (60 40.1)2 + (70 40.1)2 }/(10 1)
= 248.99.

The standard deviations for tests 1 and 2 are

s1 = 8.54 = 2.923
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 4

and
s2 = 248.99 = 15.779.
Note that the standard deviation of the rainfall amounts in the year 2000 is much larger than for the year
1900 indicating much more variability in rainfall amounts in the year 2000.
Just as the sample mean y was an estimator for the population mean , the sample variance s2 is an
estimator for the population variance which is denoted by the Greek letter 2 (sigma-squared). Recall that
the sample variance is the average of the squared deviations of the observations about the sample mean.

Question: Why do we divide by n 1 then instead of n in (2)? The answer is that if we divide by n
instead of n 1, the estimator s2 will be too small on average. In other words, dividing by n will tend to
make s2 systematically smaller than the true population variance 2 on average. This is known as bias
in statistics. Often we prefer to use estimators that are unbiased. A biased estimator will tend to be
systematically too large or too small. Dividing by n 1 in (2) makes s2 unbiased.
Note that the standard deviation (or variance) will be zero if and only if all the observations have exactly
the same value (no variability).

Using SAS to Obtain Descriptive Statistics.

Ordinarily, statistical software is used to do the tedious computations needed to obtain the values of
descriptive statistics. We shall illustrate how to do this using SAS. The appendix to these notes has more
detail about specifics of using SAS. Generally in SAS, to obtain statistical output of interest, one uses one
of several Procedures which are abbreviated PROC in SAS. In order to obtain simple statistics like the
mean and standard deviation, one could use PROC MEANS. In the SAS program below, we use PROC
UNIVARIATE which generates a large number of descriptive statistics, most of which we will ignore for
now.

/***********************************************************
The PCB concentration (ppm) and thickness (mm) of
65 Anacapa pelican eggs (Risebrough 1972) were measured.
This SAS program will illustrate PROC UNIVARIATE for obtaining
some basic statistics.
***************************************************************/
data pelican;
infile c:\stt630\notes\pelican.dat;
input pcb thick;
run;
proc print data=pelican;
run;
proc univariate plot;
var thick;
run;

Here are some notes about the SAS program above:

It is a good idea to put lots of comments in your SAS programs so that you will know what the
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 5

program is suppose to do. In SAS, comments are added by using / and the /. Anything that
appears between these symbols is ignored by SAS.

Each command line in SAS must end with a semi-colon ;. Please remember this I once spend a
couple hours trying to debug a SAS program when I first started using SAS and my error was that
I forgot a semi-colon.

When you read a data set into SAS, we need to give it a name. The data pelican; statement creates
an internal SAS data set and we have named it pelican.

In order to analyze a data set, the program must read in the data. There are two ways to do
this in SAS. In this program, we read in the data from an external file which in this case is called
pelican.dat. The data is read in using the infile command. Following the infile command, you
must give the location of the data set on your computer hard drive in single quotes. Note that data
files often have the .dat extension so that one will know that the file is a data file. The other way
of reading in data is to type the data lines directly into the program. We shall see examples of this
later.

The next line of the program is the input statement. This statement gives names to each of
the variables in the data set. The way data sets are organized is that each row corresponds to an
observation and each column corresponds to a variable. In this data set there are two columns for the
two variables which I have names pcb for PCB concentration and thick for eggshell thickness.

After each command in a SAS program, it is a good idea to have a run statement which directs
SAS to run the command. Thus, the run statement following the infile and input statements directs
SAS to go ahead and read in the data.

The first PROC statement in this program is the PROC PRINT statement which simply prints out
the data in the output window. Note that we are specifying SAS to print out the pelican data set
by writing data=pelican. This statement is not needed here because there is only one SAS data
set in this program. In more complicated programs, one can work with several SAS data sets in a
single program and it is useful to specify which one to apply a procedure to. If you do not specify a
data set using the data=, then SAS automatically uses the most recently created data set.

The main purpose of the above program is to obtain some basic descriptive statistics. This is done
using PROC UNIVARIATE which produces a lot of output, most of which we are not interested in
at the moment. However, below we see some of the pertinent output from PROC UNIVARIATE
from the output window. The plot command in the proc univariate line instructs SAS to produce
some plots. We discuss plots below.

The var thick tells SAS to compute descriptive statistics for the variable egg thickness only. If we
leave off the var statement, the SAS will automatically produce statistics for all the variables (i.e.
each column in the data set).

Moments

N 65 Sum Weights 65
Mean 0.31630769 Sum Observations 20.56
Std Deviation 0.08047844 Variance 0.00647678
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 6

Skewness 0.31932748 Kurtosis -0.501788

Uncorrected SS 6.9178 Corrected SS 0.41451385
Coeff Variation 25.4430857 Std Error Mean 0.00998212

Basic Statistical Measures

Location Variability

Mean 0.316308 Std Deviation 0.08048

Median 0.300000 Variance 0.00648
Mode 0.300000 Range 0.35000
Interquartile Range 0.11000

From this SAS output, we see that the sample size is 65, the sample mean is 0.3163, the median is 0.30,
and the sample standard deviation is 0.080478. Several other statistics are produces such as the skewness.
Note that the standard error of the mean is the sample standard deviation divided by the square root of
the sample size. The coefficient of variation is 100 (standard deviation/mean )

Linear Transformations When collecting data by taking measurements, the scale that the measurements
are made is somewhat arbitrary: inches or centimeters, Fahrenheit or Celsius, etc. It is quite common
to change the measurement scale when analyzing the data. Many transformations are linear. That is,
suppose our original measurement is yi and we change the scale by considering instead xi = a + byi where
a and b 6= 0 are constants. What happens to the mean and variance if we perform a linear transformation?
Consider first adding a constant to each measurement xi = a + yi . For instance, suppose yi represents the
salary for an employee of a large company and the owner gives each employee a $1000 raise. Then the new
salaries are xi = a + yi where a = 1000. The average (or mean) salary will increase by 1000. However, if
we add a constant to each observation, does that change the spread? The answer is no. The variance of
the xi s will be the same as the yi s. This is easy to demonstrate algebraically.
If you make a transformation by multiplying each observation by a non-zero constant b, xi = byi , then the
sample mean will change by a factor of b and the sample variance changes by a factor of b2 :

x = by and s2x = b2 s2y

where s2x is the sample variance for the x-measurements. Note that the sample standard deviation for the
xi s is |b|sy (b could be negative but the standard deviation cannot be negative).
In general, if xi = a + byi , then the mean of the xi s is x = a + by and the standard deviation for the xi s
is sx = |b|sy .

Graphical Methods

Descriptive statistics are very useful for understanding a data set. Complementing the use of descriptive
statistics are visual aids ways of graphing the data so one can get a picture of the distribution of the
data points. The shape of the distribution can tell us a lot about the population of interest as we shall
see with some examples. We shall introduce a few very common graphical methods.
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 7

Stem-and-leaf plots. In PROC UNIVARIATE in the SAS program above, because we specified plot
in the program, SAS produced a stem-and-leaf plot and a boxplot. The advantage to the stem-and-leaf plot
is that it is easy to draw by hand. In the stem-and-leaf plot, each data point is broken into two pieces: a
stem and a leaf. For example, if one considers a data set consisting of test scores in a range of 0 to 100,
the stem could be the 10s digit and the leaf could be the units digit. There may be several observations
with a common stem (say test scores in the 70s). To draw the plot, begin by drawing a vertical line. On
the left side, write out the stem values in order. On the right side, write down all the leaf values for each
stem value (e.g. for the stem value of 7 corresponding to test scores in the 70s, to the left of the vertical
line, write down all the leaf values for each test score in the 70s. The stem-and-leaf plot produced by SAS
in proc univariate simply uses 0s for the leaf values, but one should use the actual leaf values.
Box plots. The box plot is drawn by simply drawing a box. The bottom and top of the box correspond
to the lower quartile (i.e. the 25th percentile) and upper quartile (i.e. the 75th percentile) of the data sets
respectively. A line is drawn through the middle of the box corresponding to the median of the data set
(note that the median is the 50th percentile). Also, + sign is drawn in the box to indicate the sample
mean. If the distribution is roughly symmetric the mean and median will be approximately equal in which
case the line through the middle of the box and the + sign will coincide. If the distribution is skewed to
the left or right, the mean will be pulled in the direction of skewness away from the median. Next, vertical
lines are drawn from the top and bottom of the box these lines are called whiskers. The whisker lines
extend to the most extreme value in the data set within the range of

upper quartile + 1.5 [IQR],

and
lower quartile 1.5 [IQR],
where
IQR = Inter-Quartile Range = upper quartile lower quartile.
From the SAS output above, the IQR for the pelican eggshell thickness is 0.11. Observations that lie
beyond the whisker lines are considered outlying values and these are plotted in the box plot beyond the
whiskers. Any observation lying more than 3 times the IQR above or below the upper or lower quartiles
are considered extreme outliers. One should examine these data points to see if they correspond to a
typographical error. Sometimes extreme outliers occur because the observation may come from a different
population than the one under consideration.
Box plots are particularly useful when comparing two or more distributions by drawing the box plots for
the different distributions side-by-side. We will see examples of this later as well.
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 8

Below is the SAS stem-and-leaf and box plot produced from proc univariate in SAS:

Stem Leaf # Boxplot

48 00 2 |
46 000 3 |
44 0 1 |
42 000 3 |
40 000 3 |
38 000 3 |
36 000000 6 +-----+
34 00000 5 | |
32 0 1 | |
30 000000000 9 *--+--*
28 000000000 9 | |
26 00000 5 +-----+
24 000 3 |
22 0000000 7 |
20 000 3 |
18 0 1 |
16 |
14 0 1 |
----+----+----+----+
Multiply Stem.Leaf by 10**-2

In the box plot, the mean and median coincide. The stem-and-leaf plot shows that the shape of the
distribution appears to be slightly skewed to the right (towards larger values) and unimodal a single
mode. The box plot does not show any observations as being outliers.
Histograms. One of the most useful plots for capturing the shape of a distribution is the histogram.
The idea is to divide the observations into groups. In the test score example, one can create groups of
scores as 90100, 8089, 7079, etc. Draw a horizontal number line and mark the group boundaries on
this number line. Next, draw rectangles for each group whose base is on the number line and whose height
corresponds to either the frequency (i.e. number of observations) in the group or the relative frequency
(i.e. the percentage of observations) in the group. The histogram on the next page, produced by SAS,
is a relative frequency histogram. Again, we see a unimodal distribution of eggshell thicknesses for the
pelicans that has a slight hint of being skewed to the right.
It is possible to write SAS code to obtain this high-resolution histogram. However, SAS for windows has
a built-in menu driven way of producing these plots (as well as obtaining basic statistical output) without
having to write any programs. To obtain this histogram in SAS, on the top toolbar, click with your mouse
on Solutions followed by Analysis followed by Analyst:
Solutions Analysis Analyst.
A new window will open up. Under the file menu, choose Open by SAS name. Double-click on the
Work file which contains all the SAS data sets in operation. Next, double-click on pelican. You will
see the pelican data set open up in the Analyst window. At the top along the toolbar you will see several
options. Choose Graphs and click on histogram. Click on thick followed by Analysis followed by
OK. This will produce the histogram on the next page.
STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 9

Figure 1: Histogram of pelican eggshell thicknesses, produced by SAS.

 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 10

Figure 2: Four common shapes of data distributions: top-left: symmetric and bell-shaped (or mound-
shaped), top-right: skewed to the right, bottom-left: skewed to the left, bottom-right: bimodal

Distribution Shapes. Part of the problem with constructing a histogram is determining the number
of groups for dividing the data into. If the number of groups is too large, then there will not be enough
observations in each group. If the number of groups is too small, then it will be difficult to get a picture
of the shape of the distribution. If we could obtain larger and larger sample sizes from the population of
interest, then we could create histograms with a larger number of groups over smaller and smaller regions.
Imagine then that in the limit as the sample size becomes arbitrarily large, the histogram will become finer
and finer and will look like a smooth curve. These smooth plots are known as density curves and they tell
us the shape of the distribution.
Figure 2 shows four different shapes commonly encountered with real data sets. The bell-shaped distribu-
tion (top-left panel of Figure 2) was constructed based on facial measurements of a sample of adult men.
This symmetric bell-shape distribution is actually quite common in many applications. The bell-shaped
curves are often seen with data from homogeneous populations where the frequency of measurements on
either side of the mean drop off symmetrically the further you move away from the mean.
A study on mumps was conducted on children and the amount of mumps antibodies in the blood was the
variable measured. The data from this study produced a histogram with the shape shown in the plot of
the bottom-right of Figure 2. The shape of this distribution is bimodal. Why is this? The answer is that
some children had already had mumps and therefore had the mumps antibodies in their blood while other
children had not yet had mumps. Bimodal distributions appear quite frequently in practice (e.g. data on
males and females). Of course, one can encounter multi-modal distributions as well in practice.
The distributions plotted in the top-right and bottom-left panels of Figure 2 show skewed distributions.
Data on home prices and salaries tend to be skewed to the right due to some very expensive homes and
high incomes. If you look at test scores from an easy exam, then you would expect to see a distribution
skewed to the left.
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 11

Figure 3: Mixture distribution of molar widths for two species of voles.

Mixtures. Sometimes a population may contain homogeneous subgroups. If we collect data from such
a population, then the resulting histogram may reveal multiple modes as in the bottom-right plot in
Figure 2. A study was conducted in order to distinguish closely related species of voles (Airoldi et al
1996). For example, the left panel of Figure 3 shows a smoothed histogram of a sample of voles where
the variable measured was molar widths. This smoothed histogram shows a clear bimodal pattern. The
reason why is that the sample contains two closely related species of vole: microtus mutiplex and microtus
subterraneus. The right panel of Figure 3 shows the individual smoothed histograms for each species. A
distribution that consists of several homogeneous sub-populations is known as a mixture distribution. One
of the statistical problems in this framework is to estimate the parameters (means, standard deviations,
etc.) for the separate components of the mixture. The problem is often further complicated in cases where
there is no information in the data set as to which sub-population a particular observation belongs.
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 12

Figure 4: Smoothed histograms of weekly HAM-D scores for phenelzine subjects (dashed histogram) and
placebo subjects (solid histogram).

Example. A clinical trial was conducted to study the effectiveness of phenelzine in treating depression.
In order to evaluate the effectiveness of the drug, a control is needed in the study. If a patient gets better,
can we claim that it is due to the drug, or due to a placebo effect, or due to spontaneous improvement? In
this particular study, subjects were randomized to either receive the study drug phenelzine or a placebo.
The subjects were then evaluated for six weeks. At each weekly examination, the patients depression was
recorded on the Hamilton depression (HAM-D) scale. Lower scores indicate lower levels of depression.
Figure 4 shows smoothed histograms of the HAM-D distributions for patients on phenelzine (dashed
curve) and placebo (solid curve) respectively. At baseline (week 0), the HAM-D distributions for patients
on phenelzine and placebo both appear to be approximately bell-shaped. However, as the weeks go by,
the distribution for phenelzine treated subjects moves to the left towards improvement. That is, the mean
HAM-D is shifting to lower values each week. However, notice how the distribution becomes more and
more skewed each week for phenelzine treated subjects. The reason this happens is that the distribution
has a sharp boundary of zero on the left, the lowest score possible. Patients tend to gravitate towards this
score during the course of treatment. When there is a sharp boundary for a variable of interest and data
values fall near this boundary, the result is often a skewed distribution away from this boundary.
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 13

Empirical Rule

34% 34%

13.5% 13.5%
2.5% 2.5%

2 + + 2

Figure 5: An illustration of the empirical rule

The Empirical Rule

We end this chapter with a very useful rule for interpreting standard deviations. Recall from Figure 2
that many populations exhibit a bell-shape when plotted (in histograms say). For these types of data
sets, the information can typically be summarized by the mean and standard deviation without much loss
of information. The following rule, known as the empirical rule, allows us to interpret the mean and
standard deviation of a distribution:

Approximately 68% of the observations will lie within 1 standard deviation of the mean: .

Approximately 95% of the observations will lie within 2 standard deviations of the mean: 2.

Approximately 99.7% of the observations lie within three of the mean. That is, practically all
observations will lie within 3 standard deviations of the mean.

If 68% of the distribution lies within one standard deviation of the mean, then we can divide this 68%
evenly in two (68%/2 = 34%) with 34% lying between and . The other 34% lies between and
+ . Further breakdown of the distribution is illustrated in Figure 5.
The empirical rule also holds for sample data possessing a bell-shaped by simply replacing by y and
by s in the above rule.
According to the empirical rule, practically the entire distribution (99.7%) will lie in a range of 6 (from
3 to + 3).

A useful way to interpret an observation is in terms of its z-score which is the relative standing of the
observation in terms of the mean and standard deviation. If x is a data point, the z-score is computed by
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 14

standardizing the value by

x x x
Population z-score : z = Sample z-score : z = . (4)
s

Batting Averages Example. No major league baseball player has had a batting average of .400 or
better since Ted Williams in 1941. A players batting average is basically the number of hits the player
has divided by the number of times the player goes to bat. Many baseball fans have debated the reason for
the extinction of .400 hitting in baseball for years. If we look at the data, the empirical rule demonstrates
why .400 hitting is non-existent in major league baseball. Below are dotplots for league batting averages
in the years 1920 and 1997 the dots represent batting averages for all players with at least 250 plate
appearances for the season. The dotplots show that the spread of batting averages in 1920 is greater than
in 1997. The dotplots also show approximately bell-shaped batting average distributions.

.
. : : .:: .
:. :.:::::: ::
. .... :::::::::::::: :... :
. : :.:.:::::::::::::::::::::::::.: .. :. .. .
---+---------+---------+---------+---------+---------+---1920
. . : :
: : ..: .:.
.:: :::::. :::.
::: ::::::::::: .
. :::::::::::::::::
. ::::::::::::::::::: : .
..: :::::::::::::::::::: : :.
. :::::::::::::::::::::::::::::: . ...
---+---------+---------+---------+---------+---------+---1997
0.200 0.240 0.280 0.320 0.360 0.400

The mean and standard deviation for league batting averages in 1920 are 1 = 0.28553 and 1 = 0.03769
and for 1997, the mean and standard deviation are 2 = 0.27490 and 2 = 0.02874. In 1920, a .400 batting
average was about three standard deviations above the mean: z1 = (.400 .28553)/0.03769 = 3.037. Thus,
a .400 batting average in 1920 is pretty extreme but it would not be too unusual to observe one player
batting .400 or better and that is what happened in 1920. However, in 1997, a .400 batting average is
about 4.35 standard deviations above the mean: z2 = (.400 .27490)/0.02874. According to the empirical
rule, practically all players will have batting averages within three of the mean. In 1997, a .400 batting
average was 4.35 standard deviations above the mean indicating that obtaining a .400 batting average that
year would have been extraordinarily rare. One of the differences in the statistics between 1920 and 1997
is that the standard deviation of batting averages is smaller in 1997. In fact, the standard deviations of
batting averages became smaller and smaller from the early 1900s to the end of the 20th century while the
overall league batting averages tended to fluctuate up and down. To understand why .400 hitting became
extinct, one needs to figure out why the variability in batting averages (and thus the standard deviations)
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 15

declined as the years went by. Stephen Jay Gould provided a discussion of this in his book Full House as
well as his explanation for the decreasing variability in batting averages.

Diversity Indices. An important class of statistics for measuring the ecological health of an area is given
by diversity indices. Diversity indices are used for nominal scale data where observations are recorded
depending upon which category they fall in (e.g. data on birds may record how many birds are observed as
belonging to particular species of bird). Suppose you are examining the ecological health of a stream that
is populated by 5 different species of fish. Let p1 , p2 , . . . , p5 denote the proportion of fish that fall into the 5
P
categories of fish ( pi = 1.) If p1 = 1 and all the other pi s equal zero, then there is very little diversity of
fish. Typically the ecological health of an area corresponds to high levels of diversity. If p1 = p2 = = p5 ,
then there is an even spread of fish in the stream or high diversity. A couple of popular diversity indices
are
k
X
Shannons Index: = pi log pi
i=1
k
X
Simpsons Index: = 1 p2i
i=1

where k = the number of categories.

1 Problems
SAS code and data can be found in the Appendix of this chapter.

1. n = 7 sediment samples are obtained from a lake yielding the following PCB concentrations (in ng/g
dry wt.):
0.4, 0.7, 1.3, 2.8, 3.9, 6.0, 6.1.

a) What is the sample mean PCB concentration?

b) What is the median PCB concentration?
c) What is the sample standard deviation from these measurements?
d) Suppose each of the n = 7 observations are re-scaled by multiply each by 100. What is the new
sample mean and sample standard deviation for the re-scaled data?

2. Data from the Skyline 50k Ultra-Marathon (August 4, 2002) is contained in the data file ultra.dat.
For this assignment, run the SAS program ultra.sas to obtain basic univariate statistics for the
ultra-marathon times. The goal is to use the output from this program to compare the performances
of for men and women in the ultra-marathon race. For this problem, write a short report (no more
than one page not including graphics) with the following parts:

a) Write an introductory paragraph.

b) Compare men and women in terms of their average times and their standard deviations.
c) Include a plot in your report that compares the times for men and women and comment on the
plot.
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 16

d) Write a concluding paragraph summarizing your results.

3. The systolic blood pressure was recorded (mm Hg) for 6 adult males between the ages of 30-40 while
standing yielding the following data:

105, 124, 102, 114, 96, 106.

a) (2 point) What is the population of interest?

b) What is the variable?
c) What is the sample mean from this data?
d) What is the sample median?
e) Compute the sample standard deviation (without using a computer)?

4. Run the nortemp.sas program and report the average pulse rates for men and women and their
standard deviations. From these 130 observations, can we infer that in the population of healthy
adults that women have a higher pulse rate on average than men? Just give your thoughts on this
question (a sentence or two) without any formal statistical analysis (which we havent covered yet).

5. Data was collected on soil samples in the Florida Everglades and the the amount of phosphorus in
the soil samples in parts per million were measured. As is often the case with this type of data, the
distribution of phosphorus is skewed to the right. The table below shows SAS output from PROC
MEANS for the phosphorus data and the natural logarithm of the phosphorus data. A logarithm
transformation is often used for skewed right data so that the resulting distribution becomes more
bell-shaped.

n mean standard deviation min max

phosphorus 287 302.06 161.220 76.82 740.94
log(phosphorus) 287 5.56 0.568 4.342 6.608

Suppose we change the phosphorus data unit of measurement from parts per million to parts per
billion by multiplying each phosphorus value by 1000. Let x1 , x2 , . . . , xn denote the phosphorus data
in parts per million. Then yi = 1000xi is the resulting value in parts per billion. Answer the following
questions:

a) What will be the mean phosphorus level in parts per billion (ppb)? Show this algebraically.
b) What is the standard deviation of phosphorus in ppb? Show this algebraically.
c) What is the mean of the natural log-transformed phosphorus values in ppb? Show this alge-
braically.
d) What is the standard deviation of the natural-log transformed values of phosphorus in ppb?
Show this algebraically.

6. The SAS program butterclams.sas contains data on the ratio (length/width) for a sample of native
butter clam (Saxidomus giganteus) shells from the Puget Sound. The lengths and widths were
measured in centimeters. Run the program and do the following parts:

a) What is the sample mean and median for the data?

b) What is the sample standard deviation?
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 17

c) Describe the shape of the distribution from the histogram.

d) The boxplot identifies an observation as an outlier. What is the value of this observation? How
many standard deviations is this observation from the mean?

7. (5 points) The distribution of weights for newborn full term babies is bell-shaped (not necessarily nor-
mal) with an average birth weight of 7.2 pounds and standard deviation 0.5 pounds. Approximately
what percentage of full term babies have weights below 6.7 pounds? (Justify your answer.)

8. Suppose babies learn to walk on average at age 12 months with standard deviation 1.8 months and
that the distribution of ages at which babies learn to walk is bell-shaped. A mother is concerned
that her 13 month-old baby is not walking yet. Using the empirical rule, explain to the mother (in
one or two sentences) why she should not be concerned.

9. The editor of Mad Magazine stated recently in the New York Times that the average age of Mad
Magazine readers is 26 and the median age is 19. Which of the following do you think best describes
the shape of the age distribution for readers of Mad Magazine?
(a) symmetric (b) skewed right (c) uniform (d) discrete

10. Bicep girth (in cm) was measured in a sample of 6 adults yielding the following measurements:

32.5, 34.4, 33.4, 31.0, 32.0, 33.0

a) Find the sample mean for these 6 measurements.

b) Compute the sample variance for these 6 measurements.
c) What is the median for these 6 measurements?

11. The volume of solution in 50 test tubes is measured and the average volume was found to be 5.2
ml with standard deviation 0.3 ml. Exactly 1.2 ml of additional solution is added to each test
tube. Which of the following is the standard deviation for the volumes of solutions after adding the
additional solution? (Circle one)
(a) 0.3 ml (b) 1.5 ml (c) 0.3286335 ml (d) 0 ml (e) 7.4 ml (f) 0.36 ml

References

Airoldi, J. P., Flury, B. and Salvioni, M., (1996), Discrimination between two species of Microtus
using both classified and unclassified observations. Journal of Theoretical Biology, 177, p 247262.

Gould, S. J. (1996), Full House : The Spread of Excellence from Plato to Darwin, Harmony Books.

2 Appendix: SAS Programs and Data

Ultra-Marathon Data

/********************************************************
Data from the Skyline 50k Ultra-Marathon (August 4, 2002)
column 1: hours,
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 18

column 2: minutes
column 3: seconds,
column 4: Code for sex 0=male, 1=female,
column 5 = age of runner
**************************************************************/
options ls=80;
data ultra;
input hours minutes seconds sex age;
time = hours*60 + minutes + seconds/60;
datalines;
3 59 15 0 34
4 05 08 0 31
4 09 56 0 36
4 13 55 0 44
4 23 33 0 22
4 28 14 0 46
4 28 59 0 51
4 29 50 0 36
4 33 14 0 41
4 33 21 1 27
4 36 04 0 42
4 41 46 1 29
4 44 20 0 40
4 46 08 0 44
4 49 40 0 47
4 51 07 0 54
4 51 15 1 29
4 51 37 0 42
4 52 16 0 48
4 52 47 0 58
4 52 47 0 38
4 53 31 0 42
4 54 28 0 45
4 59 35 1 39
4 59 59 0 33
5 00 59 0 48
5 03 31 0 48
5 04 00 0 35
5 05 19 0 28
5 06 04 0 55
5 06 53 1 38
5 07 09 0 50
5 07 36 0 44
5 08 04 0 49
5 08 21 0 42
5 09 05 0 39
5 09 13 0 58
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 19

5 11 00 1 49
5 11 27 0 28
5 11 38 0 45
5 14 07 0 38
5 14 07 0 34
5 14 41 0 40
5 16 17 0 47
5 17 59 0 56
5 18 28 0 30
5 18 57 0 39
5 19 01 1 35
5 20 50 0 54
5 21 08 0 42
5 21 37 0 56
5 21 42 0 42
5 23 00 0 48
5 23 32 0 48
5 23 55 1 36
5 24 06 1 39
5 25 19 0 43
5 25 29 0 53
5 25 50 0 34
5 26 05 0 55
5 27 06 1 34
5 29 45 0 28
5 32 03 1 47
5 33 23 0 42
5 34 06 0 36
5 34 48 0 48
5 37 53 0 39
5 39 29 1 38
5 40 40 1 37
5 41 13 0 29
5 42 57 0 50
5 46 16 0 27
5 46 39 1 48
5 46 39 0 39
5 47 32 0 60
5 48 29 0 50
5 49 42 0 51
5 50 15 0 40
5 50 26 0 47
5 50 43 1 31
5 51 04 0 51
5 51 15 0 48
5 52 28 1 38
5 54 40 0 31
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 20

5 55 57 0 32
5 58 41 0 38
5 58 49 0 45
5 59 59 0 35
6 00 43 1 56
6 01 26 1 40
6 03 11 0 47
6 03 56 0 48
6 04 07 0 48
6 04 15 0 34
6 04 29 1 26
6 05 05 0 41
6 05 18 1 39
6 05 19 1 43
6 05 20 0 53
6 06 31 0 31
6 06 37 0 35
6 06 44 1 45
6 09 56 0 49
6 10 35 0 49
6 12 13 0 41
6 12 20 1 61
6 12 56 1 52
6 12 59 0 62
6 13 14 0 64
6 13 18 1 44
6 14 40 1 43
6 15 25 0 47
6 15 32 0 43
6 17 26 0 55
6 17 27 0 33
6 17 32 0 48
6 18 39 0 45
6 18 40 0 53
6 19 41 0 50
6 19 50 1 32
6 19 50 1 37
6 19 50 0 37
6 22 27 1 29
6 22 29 0 29
6 22 34 1 56
6 23 56 0 56
6 24 34 1 41
6 24 35 0 65
6 25 46 1 44
6 25 46 1 52
6 26 47 1 47
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 21

6 27 17 0 55
6 27 36 1 44
6 28 06 1 61
6 28 39 0 51
6 28 42 0 60
6 29 05 1 52
6 31 10 1 24
6 31 10 0 30
6 31 13 1 45
6 31 13 1 50
6 35 00 0 64
6 35 26 0 46
6 35 47 0 39
6 36 09 0 63
6 36 09 0 53
6 36 09 0 48
6 38 46 0 49
6 38 53 0 40
6 38 53 1 35
6 39 01 1 55
6 39 01 1 51
6 39 27 0 54
6 40 24 0 52
6 40 25 0 57
6 41 25 1 50
6 42 08 0 52
6 42 36 0 56
6 45 10 0 43
6 46 15 1 37
6 46 16 0 39
6 47 42 1 55
6 48 06 0 46
6 48 22 0 34
6 50 12 0 51
6 50 17 1 41
6 50 58 0 34
6 51 00 0 38
6 52 25 0 33
6 53 04 0 65
6 53 17 0 59
6 57 17 0 56
6 57 37 0 45
6 58 05 0 58
6 58 50 0 51
6 59 22 1 61
6 59 55 0 54
7 00 54 0 48
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 22

7 01 40 1 60
7 02 34 0 40
7 06 35 0 55
7 08 23 0 56
7 11 46 1 44
7 16 48 1 38
7 20 08 1 55
7 21 59 1 57
7 21 59 0 35
7 23 21 0 57
7 29 08 1 50
7 35 26 1 79
7 35 31 0 64
7 41 29 0 61
7 44 03 0 59
7 55 19 1 45
7 55 44 1 44
;
run;
proc corr; var time age; run;
proc print;
run;
proc sort;
by sex;
run;
proc means;
by sex;
run;
proc univariate plot normal; * obtain summary statistics on running times;
var time;
by sex;
run;
proc univariate plot normal;
var time;
run;
quit;

Butterclams.sas

/**************************************************************
The data below gives the ratio = length/width for a sample of
native butter clam (Saxidomus giganteus) shells from the Puget
Sound. The lengths and widths were measured in centimeters.
Data is from the Quantitative Environmental Learning Project
webpage: http://www.seattlecentral.edu/qelp/Data_MathTopics.html#Linear
**************************************************************/
data butterclams;
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 23

input ratio;
datalines;
1.31
1.36
1.29
1.27
1.21
1.67
1.33
1.29
1.36
1.23
1.30
1.33
1.21
1.40
1.32
1.35
1.31
1.31
1.19
1.36
1.25
1.31
1.34
1.31
1.28
1.27
1.33
1.23
1.32
1.29
1.26
1.29
1.28
1.33
1.36
1.25
1.32
1.38
1.33
1.29
1.32
1.27
1.29
1.31
1.33
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 24

1.24
1.30
1.30
1.38
1.27
1.25
1.24
1.32
1.24
1.29
1.23
1.31
1.17
1.25
1.41
1.28
1.28
1.19
1.34
1.28
1.30
1.22
1.37
1.21
1.21
1.43
1.23
1.21
1.22
1.37
1.19
1.18
1.25
1.17
1.36
1.28
1.28
1.32
1.17
1.27
1.22
1.20
1.15
run;
proc univariate plot normal;
run;
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 25

Body Temperature and Pulse Data

/**********************************************************************
SOURCE:
Mackowiak, P. A., Wasserman, S. S., and Levine, M. M. (1992), "A Critical
Appraisal of 98.6 Degrees F, the Upper Limit of the Normal Body
Temperature, and Other Legacies of Carl Reinhold August Wunderlich,"
_Journal of the American Medical Association_, 268, 1578-1580.

Column 1: Temperature (F)

Column 2: Gender (1 = male, 2 = female)
Column 3: Pulse rate
**********************************************************************/
options linesize=76 nodate;
data bodytemp;
input temp gender pulse;
datalines;
96.3 1 70
96.7 1 71
96.9 1 74
97.0 1 80
97.1 1 73
97.1 1 75
97.1 1 82
97.2 1 64
97.3 1 69
97.4 1 70
97.4 1 68
97.4 1 72
97.4 1 78
97.5 1 70
97.5 1 75
97.6 1 74
97.6 1 69
97.6 1 73
97.7 1 77
97.8 1 58
97.8 1 73
97.8 1 65
97.8 1 74
97.9 1 76
97.9 1 72
98.0 1 78
98.0 1 71
98.0 1 74
98.0 1 67
98.0 1 64
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 26

98.0 1 78
98.1 1 73
98.1 1 67
98.2 1 66
98.2 1 64
98.2 1 71
98.2 1 72
98.3 1 86
98.3 1 72
98.4 1 68
98.4 1 70
98.4 1 82
98.4 1 84
98.5 1 68
98.5 1 71
98.6 1 77
98.6 1 78
98.6 1 83
98.6 1 66
98.6 1 70
98.6 1 82
98.7 1 73
98.7 1 78
98.8 1 78
98.8 1 81
98.8 1 78
98.9 1 80
99.0 1 75
99.0 1 79
99.0 1 81
99.1 1 71
99.2 1 83
99.3 1 63
99.4 1 70
99.5 1 75
96.4 2 69
96.7 2 62
96.8 2 75
97.2 2 66
97.2 2 68
97.4 2 57
97.6 2 61
97.7 2 84
97.7 2 61
97.8 2 77
97.8 2 62
97.8 2 71
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 27

97.9 2 68
97.9 2 69
97.9 2 79
98.0 2 76
98.0 2 87
98.0 2 78
98.0 2 73
98.0 2 89
98.1 2 81
98.2 2 73
98.2 2 64
98.2 2 65
98.2 2 73
98.2 2 69
98.2 2 57
98.3 2 79
98.3 2 78
98.3 2 80
98.4 2 79
98.4 2 81
98.4 2 73
98.4 2 74
98.4 2 84
98.5 2 83
98.6 2 82
98.6 2 85
98.6 2 86
98.6 2 77
98.7 2 72
98.7 2 79
98.7 2 59
98.7 2 64
98.7 2 65
98.7 2 82
98.8 2 64
98.8 2 70
98.8 2 83
98.8 2 89
98.8 2 69
98.8 2 73
98.8 2 84
98.9 2 76
99.0 2 79
99.0 2 81
99.1 2 80
99.1 2 74
99.2 2 77
 STT 430/630/ES 760 Lecture Notes: Chapter 2: Descriptive Statistics 28

99.2 2 66
99.3 2 68
99.4 2 77
99.9 2 79
100.0 2 78
100.8 2 77
;
run;
proc means;
by gender;
run;
proc univariate normal plot;
var temp;
run;