In another technically focused episode, co-hosts Jennifer Miller and Ron Landis discuss how to use multiple linear regression to test models involving moderation (or interaction). In episode 18, we discussed multiple linear regression in which we used multiple variables to predict the outcome or criterion variable. But what happens if you have a situation in which the relation between the predictor and outcome variable is actually dependent upon (or is conditional upon) the level of a third variable? In this episode, we deconstruct moderation and some applications of moderation.
In this episode, we had conversations around these questions:
- What is moderation/interaction?
- Why might we want to use multiple linear regression (as opposed to analysis of variance, ANOVA) to test for moderation?
- What are some applications of moderation in People Analytics?
- What's the best way to communicate moderation results?
- What are some of the concerns when presenting visualizations depicting moderation?
Key Takeaways:
- Moderation or interaction involves evaluating with the relation between a predictor and outcome variable is dependent (or conditional) on the level of a third variable. For example, we might be interested in whether employee engagement predicts jobs performance. In this case, we have a simple linear regression. If we add a third variable, such as working environment (I.e., remote or hybrid), we can now ask whether the relation between engagement and job performance is the same across different working environments.
- Moderation and interaction can be used interchangeably. One can use regression based approaches or ANOVA to test for the presence of interactions, though regression allows for the use of continuous predictor variables.
- Moderation is an application of multiple linear regression. In multiple linear regression, the effects are additive meaning that each variable contributes additively to explaining the outcome variable. In moderation, the effects are multiplicative in that a product term needs to be included in the model to examine whether the variance explained in the outcome variable is over and above the when each variable is added independently to the model.
Related Links
