#hierarchical regression

[homegrown-hot-takes]

Tired of reading academic papers where the analytic technique is way overcomplicated. I do not say this because I don’t understand- trust me that I have the qualifications to understand the analysis- but honestly using a trendy analysis is often useless, especially for some of the social sciences. Your analysis should fit the data you have. You should not run an advanced model because “machine learning is cool and people will trust my shoddy research if my analysis is above most of the field’s head.” If you ran a factorial experiment and you run a majority rules classifier on your sample of 300, I will kill you. Furthermore, if your hierarchical regression features 4 levels, you belong in hell

Thoughts?

Agree

Disagree

Voting ended onJun 4

[statsineducation]

Hierarchical Regression

Q: I did a hierarchical regression with two predictors, with each entered in as a separate step. I ran this regression two ways. I first entered predictor 1 (X1), and it accounted for 8% of the variance in the outcome. Then I added predictor 2 (X2). Then I started over again. I entered X2 first and examined the R2 change attributable to X1. This time X1 only accounted for 2% of the unique variance in Y. Why are they not the same?

A: Actually, the fact that they change is exactly WHY many researchers do analyses in this way. From the numbers you’ve given me here, I can see that X1 would account for 8% of the variance in Y if it was the only predictor. I can also see that 6% of the total variance in Y could be explained by either X1 or X2. So if you put X2 in first, it will explain that 6% of the variance in Y, leaving only the remaining 2% left for X1 to explain.