I am running a simple logit regression, where my outcome variable is 1 when individuals indicate they will vote in the election and zero otherwise. Individuals psychological health such as depression, locus of control, anxiety, etc. which are measured by continuous scores ranging from -5 till 5 are among my independent variable.
I have significant effects of the anxiety, and locus of control in the original regression, and I am calculating the marginal effects at different scores of anxiety/locus of control using the margin command.
I have two questions:
1- If I want to find out whether the average probability of voting at different scores of the anxiety/locus of control are different from each other, should I use the pwcompare or contrast option.
I particular:
Code:
logit vote anx loc dep $dem margins, at(anx=(-5(1)5)) pwcompare(effects)
Code:
margins, at(anx=(-5(1)5)) contrast(effects)
2- When I use pairwise comparision, with bonferroni or other adjustment methods, all the comparisons become insignificant while the unadjusted p-values indicate significant difference between points. This I guess is due to large number of hypothesis being tested. But I am wondering wether adjustment is needed or not?
Thank you very much fro taking the time.
Emma
0 Response to margins, pwcompare vs. margins, contrast
Post a Comment