ADVANCED

STATISTICS
Types of Variables and
Levels of Measurement
Roy REVILLA
 Variables and Constants
• The first thing you must be familiar with in statistics is the concept of a variable. A variable is, quite simply,
something that varies. It is a coding scheme used to measure a particular characteristic of interest. It is also a
characteristic that describes people, objects, or places and takes on multiple values in a sample or population.

• Variables sit in contrast to constants, which are characteristics that assume only one value in a sample. A
constant has characteristic that describes people, objects, or places and takes on only one value in a sample or
population.
 Units of Analysis
• It seems rather self-evident, but nonetheless bears explicit mention, that every
scientific study contain something that the researcher conducting the study gathers
and examines. These “somethings” can be objects or entities such as rocks, people,
molecules, or prisons.
• This “something” is called the unit of analysis, and it is, essentially, whatever the
sample under study consists of. It is the object or target of a research study.
• In criminal justice and criminology research, individual people are often the units
of analysis. These individuals might be probationers, police officers, criminal
defendants, or judges. Prisons, police departments, criminal incidents, or court
records can also be units of analysis. Larger units are also popular; for example,
many studies focus on census tracks, block groups, cities, states, or even countries.
 Independent Variables and Dependent Variables
• Researchers in criminal justice and criminology typically seek to examine relationships between two or more
variables. Observed or empirical phenomena give rise to questions about the underlying forces driving
them. To be empirical means to have the qualities of being measurable, observable, or tangible. Empirical
phenomena are detectable with senses such as sight, hearing, or touch.
• Researchers undertaking quantitative studies must specify dependent variables (DVs) and independent
variables (IVs). Dependent variables are the empirical events that a researcher is attempting to explain.
They are phenomena that a researcher wishes to study, explain, or predict. Independent variables are factors
a researcher believes might affect the DV. They are also factors or characteristic that is used to try to explain
or predict a dependent variable. The designation of a certain phenomenon as an IV or a DV depends on the
nature of the research study.
 Relationships Between Variables: A Cautionary
Note
• It is vital to understand that independent and dependent are not synonymous with cause and effect,
respectively.
• A particular IV might be related to a certain DV, but this is far from definitive proof that the former is the
cause of the latter. To establish causality, researchers must demonstrate that their studies meet three criteria.
 The Three Criteria

• First is temporal ordering, meaning that the IV must occur prior to the DV. The causality requirement holding
that an independent variable must precede a dependent variable. It would be illogical, for instance, to predict
that adolescents’ participation in delinquency will impact their gender; conversely, it does make sense to predict
that adolescents’ gender affects the likelihood they will commit delinquent acts.
• The second causality requirement is that there be an empirical relationship between the IV and the DV. The
causality requirement holding that the independent and dependent variables possess an observed relationship
with one another. This is a basic necessity—it does not make sense to try to delve into the nuances of a
nonexistent connection between two variables. For example, if a researcher predicts that people living in high-
crime areas are more likely to own handguns for self-protection, but then finds no relationship between
neighborhood-level crime rates and handgun ownership, the study cannot proceed.
• The last requirement is that the relationship between the IV and the DV be nonspurious, or that the causality
requirement holding that the relationship between the independent variable and dependent variable not be the
product of a third variable that has been erroneously omitted from the analysis. This third criterion is frequently
the hardest to overcome in criminology and criminal justice research (indeed, all social sciences) because
human behavior is complicated, and each action a person engages in has multiple causes. Disentangling these
causal factors can be difficult or impossible.
 Spuriousness and Omitted Bias
• The reason spuriousness is a problem is that there could be a third variable that explains the DV as well as,
or even better than, the IV does. This third variable might partially or fully account for the relationship
between the IV and DV. The inadvertent exclusion of one or more important variables can result in erroneous
conclusions because the researcher might mistakenly believe that the IV strongly predicts the DV when, in
fact, the relationship is actually partially or entirely due to intervening factors.

• Another term for this problem is omitted variable bias. This is an error that occurs as a result of
unrecognized spuriousness and a failure to include important third variables in an analysis, leading to
incorrect conclusions about the relationship between the independent and dependent variables. When omitted
variable bias (i.e., spuriousness) is present in an IV–DV relationship but erroneously goes unrecognized,
people can reach the wrong conclusion about a phenomenon.
 Examination of Aggregate Trends
• A final caution with respect to causality is that statistical analyses are examinations of aggregate trends.
• Uncovering an association between an IV and a DV means only that the presence of the IV has the tendency to be related to either
an increase or a reduction in the DV in the sample as a whole—it is not an indication that the IV–DV link inevitably holds true for
every single person or object in the sample.
• For example, victims of early childhood trauma are more likely than nonvictims to develop substance abuse disorders later in life
(see Dass-Brailsford & Myrick, 2010). Does this mean that every person who was victimized as a child has substance abuse
problems as an adult? Certainly not! Many people who suffer childhood abuse do not become addicted to alcohol or other drugs.
Early trauma is a risk factor that elevates the risk of substance abuse, but it is not a guarantee of this outcome. Associations present
in a large group are not uniformly true of all members of that group.
• In sum, you should always be cautious when interpreting IV–DV relationships. It is better to think of IVs as predictors and DVs as
outcomes rather than to view them as causes and effects. As the adage goes, correlation does not mean causation. Variables of all
kinds are related to each other, but it is important not to leap carelessly to causal conclusions on the basis of statistical associations.
 Levels of Measurement

• Every variable possesses a level of measurement. Levels of measurement are ways of

classifying or describing variable types. They are a variable’s specific type or classification.
There are four types: nominal, ordinal, interval, and ratio.
• There are two overarching classes of variables: categorical (also sometimes called
qualitative) and continuous (also sometimes referred to as quantitative).
• Categorical variables comprise groups or classifications that are represented with labels.
They are variables that classify people or objects into groups. There are two types: nominal
and ordinal.
• Continuous variables are made of numbers that measure how much of a particular
characteristic a person or object possesses. They are variables that numerically measures the
presence of a particular characteristic. There are two types: interval and ratio.
 The Categorical Level of Measurement: Nominal
and Ordinal Variables
• Categorical variables are made up of categories. They represent ways of divvying up people and objects
according to some characteristic. Categorical variables are subdivided into two types: nominal and ordinal.
The nominal level of measurement is the most rudimentary of all the levels. It is the least descriptive and
sometimes the least informative. It is a classification that places people or objects into different groups
according to a particular characteristic that cannot be ranked in terms of quantity. Race is an example of a
nominal-level variable.
 Nominal Variable
• In Table 2.2, the variable representing the races of stopped drivers is nominal because races are groups into
which people are placed. The labels offer descriptive information about the people or objects within each
category. Data are from the Police–Public Contact Survey (PPCS).
 Nominal Variables
• Gender is another example of a nominal variable. The table below (table 2.3) displays the gender breakdown
among people who reported that they had sought help from the police within the past year.
 Nominal Variables
• Much information is missing from the nominal variables in Tables 2.1 through 2.3. For instance, the question
about knowing a local police officer does not tell us how often respondents talk to the officers they know or
whether they provide these officers with information about the area. Similarly, the race variable provides
fairly basic information. This is why the nominal level of measurement is lowest in terms of descriptiveness
and utility. These classifications represent only differences; there is no way to arrange the categories in any
meaningful rank or order. Nobody in one racial group can be said to have “more race” or “less race” than
someone in another category—they are merely of different races. The same applies to gender. Most people
identify as being either female or male, but members of one gender group do not have more or less gender
relative to members of the other group.
 Mutually Exclusive and Exhaustive
• One property that nominal variables possess (and share with other levels) is that the categories within any
given variable are mutually exclusive and exhaustive.
• They are mutually exclusive because they have the property of all levels of measurement whereby each unit
in the data set (person, place, and so on) can fall into only one category and that there is no overlap between
the categories within a variable. They are exhaustive because they have the property of all levels of
measurement whereby all units have a category that applies to them and the categories or range within a
variable capture all possible values.
• For example, a variable measuring survey respondents’ criminal histories that asks them if they have been
arrested “0–1 time” or “1–2 times” would not be mutually exclusive because a respondent who has been
arrested once could circle both answer options. This variable would also violate the principle of
exhaustiveness because someone who has been arrested three or more times cannot circle any available
option because neither is applicable. To correct these problems, the answer options could be changed to, for
instance, “no arrests,” “1–2 arrests,” and “3 or more arrests.” Everyone filling out the survey would have one,
and only one, answer option that accurately reflected their experiences.
 Ordinal Variables
• Ordinal variables are one step up from nominal variables in terms of descriptiveness because they can be
ranked according to the quantity of a characteristic possessed by each person or object in a sample. They are a
classification that places people or objects into different groups according to a particular characteristic that
can be ranked in terms of quantity.
• University students’ class level is an ordinal variable because freshmen, sophomores, juniors, and seniors can
be rank-ordered according to how many credits they have earned. Numbers can also be represented as ordinal
classifications when the numbers have been grouped into ranges like those in Table 2.4, where the income
categories of respondents to the General Social Survey (GSS; see Data Sources 2.2) are shown. Table 2.4
displays another variable from the PPCS. This survey question queried respondents about how often they
drive. Respondents were offered categories and selected the one that most accurately described them.
 Ordinal Variables
• Ordinal variables are useful because they allow people or objects to be ranked in a meaningful order. Ordinal
variables are limited, though, by the fact that no algebraic techniques can be applied to them. This includes
ordinal variables made from numbers such as those in Table 2.4. It is impossible, for instance, to subtract
<$1,000 from $15,000–$19,999. This eliminates the ability to determine exactly how far apart two
respondents are in their income levels. The difference between someone in the $20,000–$24,999 group and
the ≥$25,000 group might only be $1 if the former makes $24,999 and the latter makes $25,000. The
difference could be enormous, however, if the person in the ≥$25,000 group has an annual family income of
$500,000 per year. There is no way to figure this out from an assortment of categories like those in Table 2.4.
The same limitation applies to the variable in Table 2.4. A glance at the table reveals general information
about the frequency of driving, but there is no way to add, subtract, multiply, or divide the categories to
obtain a specific measurement of how more or less one person in the sample drives relative to another.
 Ordinal Variables
• Table 2.6 contains another example of an ordinal variable. This comes from the GSS and measures female
respondents’ educational attainment. You can see that this variable is categorical and ranked. Someone who
has attended junior college has a higher educational attainment than someone who did not complete high
school. As before, though, no math is possible. Someone whose highest education level is junior college
might be in her first semester or quarter, putting her barely above someone in the high-school-only category,
or she might be on the verge of completing her associate’s degree, which places her nearly two years above.
 The Continuous Level of Measurement: Interval and Ratio Variables

• Continuous variables differ from categorical ones in that the former are represented not by categories but
rather by numbers. Interval variables are numerical scales in which there are equal distances between all
adjacent points on those scales and are quantitative variables that numerically measures the extent to which a
particular characteristic is present or absent
• Ambient temperature is a classic example of an interval variable. This scale is measured using numbers
representing degrees, and every point on the scale is exactly one degree away from the nearest points on each
side. Twenty degrees Fahrenheit, for instance, is exactly 1 degree cooler than 21 degrees and exactly 4
degrees warmer than 16 degrees.
• An example of an interval-level variable is the GSS’s scoring of respondents’ occupational prestige. The GSS
uses a ranking system to assign each person a number representing how prestigious his or her occupation is.
Figure 2.2 displays the results. The scores range from 16 to 80 and so are presented as a chart rather than a
table.
Figure 2.2 Respondents’ Occupational
Prestige Scores
 Ratio Variables
• Ratio variables are the other subtype within the continuous level of measurement. They are a quantitative
variable that numerically measures the extent to which a particular characteristic is present or absent and has
a true zero point. The ratio level resembles the interval level in that ratio, too, is numerical and has equal and
known distance between adjacent points. The difference is that ratio-level scales, unlike interval ones, have
meaningful zero points that represent the absence of a given characteristic.

• Temperature, for instance, is not ratio level because the zeros in the various temperature scales are just
placeholders. Zero does not signify an absence of temperature. Likewise, the data presented in Figure 2.2 and
Tables 2.7, as discussed previously, do not have meaningful zero points and therefore cannot be multiplied or
divided. You could not, for instance, say that someone who scores a 4 on the punitiveness scale is twice as
punitive as a person with a 2 on this measure. Ratio-level data permit this higher level of detail.

• Criminal justice and criminology researchers deal with many ratio-level variables. Age is one example.
Although it is strange to think of someone as having zero age, age can be traced close enough to zero to make
analytically reasonable to think of this variable as ratio. A 40-year-old person is twice as old as someone who
is 20.
Table 2.7 Respondents’ Political Views
 Interval and Ratio Variables
• In the real world of statistical analysis, the terms interval and ratio variables are often used interchangeably.
It is the overarching categorical-versus-continuous distinction that usually matters most when it comes to
statistical analyses. When a researcher is collecting data and has a choice about level of measurement, the
best strategy is to always use the highest level possible. A continuous variable can always be made categorical
later, but a categorical variable can never be made continuous.

• Level of measurement is a very important concept. It might be difficult to grasp if this is the first time you
have been exposed to this idea; however, it is imperative that you gain a firm understanding because level of
measurement determines what analyses can and cannot be conducted. This fundamental point will form an
underlying theme of this entire book, so be sure you understand it. Do not proceed with the book until you
can readily identify a given variable’s level of measurement. Table 2.11 summarizes the basic characteristics
that define each level and distinguish it from the others.