Noncausal Covariation

Although correlation is a necessary condition for causation, it is not a sufficient condition. That is, if X and Y can be shown to correlate, it is possible that X may cause Y or vice versa. However, just because correlation is established between the two variables, it is not certain that X causes Y or that Y causes X. In instances when X and Y are correlated but there is no empirical evidence that one causes the other, a researcher is left with a finding of noncausal covariation. A researcher [Page 514]can speculate that one variable causes the other, but unless there is empirical evidence demonstrating an internally valid casual relationship, the researcher has no solid ground upon which to claim the relationship is causal.

In survey research, researchers rarely have valid evidence upon which to base conclusions of causation. Many researchers forget this and often interpret and report their results as though a causal relationship does exist between variables. For example, a researcher may find a correlation between minority status and the willingness to cooperate in a survey when sampled. However, merely finding that minority status is correlated with someone's response propensity is not sufficient to claim that being a racial or ethnic minority person “causes” one to be less likely to participate in surveys. Instead, it is likely that some other variables that are correlated with both being a minority and not being as willing to participate in surveys, such as educational attainment, are the real causal agents.

To demonstrate a causal relationship using a research design with strong internal validity, a true experiment is necessary. Experiments require that random assignment of respondents be carried out with exposure to different levels of the independent variable that the researcher controls. Then, in its simplest form, the experiment will show whether the group assigned to one level of the independent variable shows statistically different levels of the dependent variable than does the group exposed to the other level of the independent variable. If it does, then a causal relationship has been identified. For example, if survey respondents were randomly assigned to one of two levels of prepaid incentives ($5 or $10), then the researcher could determine whether the difference in incentives changed the response rate of the group getting the higher incentive. If it did, then the researcher has evidence of causation, not merely correlation.

Unfortunately, there are many relationships that survey researchers are interested in studying that do not readily lend themselves to experimentation. Although there are other statistical techniques that can be used to investigate whether a correlational relationship is likely to also represent a casual relationship, without an experimental design, a researcher cannot be as confident about drawing cause-and-effect conclusions and often must resign herself or himself to acknowledging that the relationship is one of noncausal correlation.