Independent Variable (In Nonexperimental Research)

Also known as explanatory variable or input variable, an independent variable—or, more precisely, its effect on a dependent variable—is what a researcher analyzes in typical quantitative social science research. Strictly speaking, an independent variable is a variable that the researcher can manipulate, as in experiments. However, in nonexperimental research, variables are not manipulated, so that it is sometimes unclear which among the various variables studied can be regarded as INDEPENDENT VARIABLES. Sometimes, researchers use the term loosely to refer to any variables they include on the right-hand side of a REGRESSION equation, where causal inference is often implied. But it is necessary to discuss what qualifies as an independent variable in different RESEARCH DESIGNS. When data derive from a CROSS-SECTIONAL DESIGN, information on all variables is collected coterminously, and no variables are manipulated. This means that the issue of which variables have causal impacts on other variables may not be immediately obvious. Nonmanipulation of variables in disciplines such as sociology, political science, economics, and geography arises for several reasons: It may not be possible to manipulate certain variables (such as ethnicity or gender); it may be impractical to manipulate certain variables (such as the region in which people live); or it may be ethically and politically unacceptable (such as the causal impact of poverty). Some of these variables (e.g., race, ethnicity, gender, country of origin, etc.) are regarded as “fixed” in the analysis. Contrary to the definition of manipulatibility in experimental research, fixed variables are not at all manipulatible but nevertheless are the ones that can be safely considered and treated as “independent” in nonexperimental research.

[Page 485]With longitudinal designs, such as a PANEL DESIGN, the issue is somewhat more complex, because although variables are still not manipulated, the fact that data can be collected at different points in time allows some empirical leverage on the issue of the temporal and hence causal priority of variables. Strict experimentalists might still argue that the lack of manipulation casts an element of doubt over the issue of causal priority. To deal with this difficulty, analysts of panel data resort to the so-called cross-lagged panel analysis for assessing causal priority among a pair of variables measured at two or more points in time. The line between dependent and independent variables in such analysis in a sense is blurred because the variable considered causally precedent is treated as an independent variable in the first regression equation but as a dependent variable in the second. Conversely, the variable considered causally succedent is treated as dependent in the first equation and as independent in the second. Judging from analyses like this, it appears that the term independent variable can be quite arbitrary; what matters is clear thinking and careful analysis in nonexperimental research.

Given that it is well known that it is wrong to infer causality from findings from a single regression run about relationships between variables, which is what are usually gleaned from nonexperimental research, how can we have independent variables in such research? Researchers employing nonexperimental research designs (such as a SURVEY design) must engage in causal inference to tease out independent variables. Essentially, this process entails a mixture of commonsense inferences from our understanding about the nature of the social world and from existing theory and research, as well as analytical methods in the area that is the focus of attention. It is this process of causal inference that lies behind such approaches as CAUSAL MODELING and PATH ANALYSIS, to which the cross-lagged panel analysis is related.

...