Intervening Versus Confounding Variables

In exercises 3 and 4, you explored possible confounding variables. As exercise 3 illustrated, if the original relationship between the independent and dependent variable vanishes when the possible confounding variable is controlled for, then you should conclude that the control variable really is a confounding variable and that the original relationship was a spurious one.

In this exercise, you examined a possible intervening variable. Because the original relationship between the independent and dependent variable became much weaker when the possible intervening variable was controlled for, you concluded that the control variable really was an intervening variable that helped to explain the link between the independent and dependent variables.

It is important to understand the difference between the analysis in this exercise compared to the analyses in the previous two exercises. In those exercises, the control variable was a potential confounding variable, one that might produce a spurious association between the independent and dependent variables. In such a situation, if the original relationship disappears when the control is applied, then we would conclude that the original relationship was spurious. In this exercise, we are controlling for a potential intervening variable. If the original relationship disappears, or even greatly weakens, then we would conclude that we have identified a mechanism or path through which the independent variable affects the dependent variable.

If controlling for a variable makes the original relationship disappear or greatly weaken, how do you know whether the control variable is a confounding variable or an intervening variable? The answer is that you have to theoretically interpret the results. You have to decide whether the control variable is affected by the independent variable (which would make the control variable an intervening variable) or whether it affects the independent variable (which would make it a confounding variable). It is not just a matter of reading the table percentages. It also is a matter of interpreting the results, which requires you to think about how the variables might affect each other.