I am carrying out a DIF analysis comparing the scores of men and women on a questionnaire. A preliminary ANOVA revealed that when comparing total scores on the test, there is a significant effect of sex, but not of age or an interaction between sex:age.

I was going to have equal numbers of men and women (in total 530 men and 530 women), but I can't decide how to match them. I thought that matching approximately by age can't hurt, even if there is no significant effect of it. But would I be better off matching randomly?

Thank you

  • 1
    $\begingroup$ Why are you matching? Is the goal to control for age in the estimation of the DIF effect of gender? If so, why don't you use a moderated nonlinear factor analysis approach, which doesn't require you to throw away any data? $\endgroup$ – Noah Mar 30 at 17:08
  • $\begingroup$ Thanks for your answer. Yes the goal is to control for age in the DIF analysis. I'm reading up on the moderated nonlinear factor analysis approach looks great, but it looks a little complicated. I already have all my code for my project (a a PCA, simple CFA and a simple DIF analysis), so I'm a little resistant to change everything. Because this is for an undergrad dissertation, the goal is not to carry out very well crafted statistical tests, but to be knowledgable and critical about what you're doing and acknowledging all pitfalls. $\endgroup$ – T August Mar 31 at 8:08
  • $\begingroup$ I see. Yes, MNLFA is definitely overkill for an undergrad thesis. I think looking at DIF by sex is sufficient. You don't need to match or have equal numbers of men and women to perform DIF testing for sex. You can also test for DIF within strata of age. Check out these papers: doi:10.3389/feduc.2017.00039, doi:10.1177/0013164419878861 for some tips on using matching with DIF. $\endgroup$ – Noah Mar 31 at 18:39
  • $\begingroup$ I didn't realise I don't have to match/have equal numbers for DIF, thank you. However, I'm also doing a factor analysis (PCA and CFA) on the exact same data, do you think its a good idea to match for this instead (matching being having 530 in each, all roughly matched by age) $\endgroup$ – T August Apr 4 at 13:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.