Skip to main content icon/video/no-internet

Causality assumes that the value of an interdependent variable is the reason for the value of a dependent variable. In other words, a person’s value on Y is caused by that person’s value on X, or X causes Y. Most social scientific research is interested in testing causal claims. In fact, most theoretically derived hypotheses implicitly (or explicitly) assume causal relationships. However, causality is very difficult to prove. In fact, some believe that causality can never be demonstrated with finality and that the best researchers can do is to generate increasingly compelling evidence that is consistent with causality.

There are three widely accepted preconditions to establish causality: first, that the variables are associated; second, that the independent variable precedes the dependent variable in temporal order; and third, that all possible alternative explanations for the relationship have been accounted for and dismissed. In what follows, these preconditions will be considered, causality will be discussed in the environment of multivariate effects, and the role of survey and experimental research in testing causality will be explained.

Association

For one variable to cause another, they must be theoretically and empirically linked. Because unscrupulous researchers can capitalize on chance associations to make dubious claims, it is necessary that any causal relationship can be theoretically proposed prior to statistical analysis. If a researcher says “based on my theoretical expectations, X should cause Y,” and X and Y are strongly associated with each other (e.g., highly correlated), then the researcher can conclude that the association is consistent with, but not dispositive proof of, the causal assumption of the theory. If theory does not precede statistical analysis, a researcher may be capitalizing on chance by finding whatever associations are present in the data and reasoning backward to develop a plausible causal explanation. This practice is referred to as data mining and is inappropriate.

If there is a reasonable theoretical expectation that one variable causes another, a researcher can test the plausibility of this claim by investigating the empirical relationship. For example, if a researcher predicts that violent video games will make children exhibit more interpersonal aggression and finds that, upon investigation, those who frequently play violent video games are no more or less interpersonally aggressive than those who rarely or never play violent video games, the variables cannot be said to be associated and the theory will have failed first test of causality. However, if the researcher finds that those who frequently play violent video games exhibit more interpersonal aggression, the variables are associated and the first test of causality is passed. The researcher can say that the association is consistent with causality, though it does not prove causality.

Temporal Order

There are a number of reasons that two variables may be associated, and one is that X causes Y. If all the researchers know is that two variables are associated, it is also possible that Y causes X. For example, children that exhibit a great deal of interpersonal aggression may play more violent video games because these games are another outlet for aggression. In other words, there may be good theoretical reasons to suspect that X causes Y but also that Y causes X. Demonstrating that X and Y are associated does not resolve this “chicken and egg” conundrum.

...

  • Loading...
locked icon

Sign in to access this content

Get a 30 day FREE TRIAL

  • Watch videos from a variety of sources bringing classroom topics to life
  • Read modern, diverse business cases
  • Explore hundreds of books and reference titles

Sage Recommends

We found other relevant content for you on other Sage platforms.

Loading