State politics research frequently involves correlated variables. This article examines multicollinearity (MC) using simulated data to reveal its hidden dangers and how simply dropping variables introduces bias.
Understanding MC Issues
The problem of highly related explanatory variables is often misunderstood, leading researchers astray when attempting to isolate effects in regression models.
Better Solutions Than Omission
Contrary to common practice, the article argues that omitting correlated variables isn't justifiable. Instead, alternative approaches should be considered.
MC and Interaction Terms
The analysis extends to multiplicative interaction models—a key area where MC is often mistakenly addressed by variable removal.
Using legislative initiatives as a concrete example of policy responsiveness research illustrates these crucial points.
Researchers face clear challenges when dealing with MC. This paper offers practical guidance on navigating this statistical pitfall.
| What to Do (and Not Do) with Multicollinearity in State Politics Research was authored by Kevin Arceneaux and Gregory Huber. It was published by Sage in SPPQ in 2007. |
