Dear community,
I am conducting a series of stepwise linear regression to examine if any IV from the five IVs I am interested in predicts the DV. After controlling for three covariates in the first step, I entered one IV each time in the second step of the regression analysis, so I conducted five stepwise regression in total.
I am concern if this will increase type I error due to the number of regression analyses. If so, how can I control the type I error and ensure the results are valid?
This kind of model is knows as a hierarchical regression (which should not be confused with hierarchical linear modeling). You can do a Google search for this in combination with bonferroni correction to get some directions on how to do this.
That said, I have seen many articles that use a hierarchical regression approach, but none of them ever offered a correction for the multiple analyses. Given that consensus, I think you could go ahead with out it, and then re-consider the issue if one of your reviewers raises this point.
If you are doing stepwise please be advised that stepwise doesn't work please see below. Best, David Booth btw particularly see the Austin and Tu reference it can be found by searching on Research gate. DB
David Eugene Booth I agree that stepwise is a poor choice if the goal is to decide which variables to include in a model. In contrast, hierarchical regression (or hierarchically ordered regression) adds blocks of variables according to a set of theoretical justifications.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis