Interaction

In most research contexts in the biopsychosocial sciences, researchers are interested in examining the influence of two or more predictor variables on an outcome. For example, researchers might be interested in examining the influence of stress levels and social support on anxiety among first-semester graduate students. In the current example, there are two predictor variables—stress levels and social support—and one outcome variable—anxiety. In its simplest form, a statistical interaction is present when the association between a predictor and an outcome varies significantly as a function of a second predictor. Given the current example, one might hypothesize that the association between stress and anxiety varies significantly as a function of social support. More specifically, one might hypothesize that there is no association between stress and anxiety among individuals reporting higher levels of social support while simultaneously hypothesizing that the association between stress and anxiety is strong among individuals reporting lower levels of social support. Data consistent with these joint hypotheses would be suggestive of a significant interaction between stress and social support in predicting anxiety.

Hypothetical data consistent with this interaction are presented in Figure 1. The horizontal axis is labeled Stress, with higher values representing higher levels of stress. The vertical axis is labeled Anxiety, with higher values representing higher levels of anxiety. In figures such as these, one predictor (in this case, stress) is plotted along the horizontal axis, while the outcome (in this case, anxiety) is plotted along the vertical axis. The second predictor (in this case, social support) forms the lines in the plot. In Figure 1, the flat line is labeled High Social Support and represents the association between stress and anxiety for individuals reporting higher levels of social support. The other line is labeled Low Social Support and represents the association between stress and anxiety for individuals reporting lower levels of social support. In plots like Figure 1, as the lines depart from parallelism, a statistical interaction is suggested.