Control Variable

In research design, a control variable is defined as a variable that is known to or expected to influence the dependent variable and might also affect the explanatory or independent variable in an analysis, but is not the focus of interest for the researcher. The influence of a control variable may interfere with the main analysis, for instance, by obscuring between-treatment or between-group differences, or creating apparent relationships between variables of interest. Often variables of this type are used to create blocks in an experimental design or stratify a sample. Occasionally, the examination of the influence of control variables is called “elaboration of the analysis” because they are not the variables of main interest in the analysis.

Within the realm of designed experiments, a control variable may be kept constant or controlled for each test or replication of the experiment. In observational studies, control variables are often used in analysis to divide the study population into smaller, more homogeneous groups, to test for the influence of potentially confounding factors. For example, if a researcher wanted to compare death rates between smokers and nonsmokers, he or she would also need to control for gender (males are the largest proportion of smokers, and males also have higher incidences of heart disease). Therefore, death rate differences between smokers and nonsmokers may actually be due to differences between genders. This could be examined by performing a stratified analysis in which data from males and females were analyzed separately. Control variables may also be included in the analysis; for instance, a continuous control variable may be included in a generalized linear model before the independent variables of interest, to separate the amount of variance accounted for by each.

Table 1 Other Names of Control Variables

Control variables have many other names within the literature (see Table 1). Some of the names may [Page 234]be used interchangeably with “control variable”; while other names indicate the specific relationship between the control variable and the independent and dependent variables. Figures 1a through d illustrate the various types of relationships that may be present between control, independent, and dependent variables.

Figure 1 Various Relationships Between Control, Independent, and Dependent Variables

Antecedent or indirect control variables influence the independent variables within an analysis, as illustrated in Figure 1a, which in turn influence the dependent variable(s).

Intervening, mediating, or subordinate control variables are in the causal chain between the independent and dependent variables, as illustrated in Figure 1b. In this case, the independent variable influences the control variable, which in turn influences the dependent variable, so that the effects of the independent variable on the dependent variable are seen only through the influence of the control variable.

If two statistically related variables become statistically independent when a third variable is included in the analysis, then the control variable may be called an incidental variable, as illustrated in Figure 1c. Usually, this type of relationship between the two variables of interest is referred to as spurious because it is due to the influence of the control variable rather than due to any inherent relationship between the variables of interest. A classic example of a spurious relationship is the relationship between ice cream sales and murder rates. These two variables are significantly positively correlated over the calendar year: When ice cream sales increase, so does the murder rate. However, taking into account the third variable of temperature as a control variable, we find the relationship between sales and murder rates is spurious, because both ice cream sales and murders tend to increase when the temperature increases, as in the hottest days of summer.

...