Can I control for covariates with a pca analysis?

[Deborah Jehu]

I am performing a principle component analysis with the between subject variable of fall status (non faller, single faller, recurrent faller) and within subject variable of sex to determine whether fall risk differs by fall status and sex. I would like to control for age and exposure to the exercise intervention in the principle component analysis, but I cannot find any code on how to go about doing this. The following pca code works, but does not control for covariates:

pca BMI_n OverallFallsRiskScore VisualContrast Proprio ReactionTime BestQuad_n EO_Firm_Sway_n EO_Foam_Sway_n Dual_TUG NIA_TotalScore GDS shortFES VAS FCI BarthelTotal_score IADL MMSE MOCA TrailA_n TrailB_n Stroop1_n Stroop2_n Stroop3_n DigitForward DigitBackward DSST_n VerbalFluency if NonSingleMultipleFaller > 1& Sex==0

This is a secondary analysis of RCT data.

Thanks for your help!

[Deborah Jehu]

I am also wondering if I am able to examine the interaction effect of sex as opposed to running separate pca analyses for males and females? Through some reading, some have said that it would be difficult to interpret interaction effects when beta weights differ. Thanks for your feedback.

[Marcos Almeida]

You may wish to try - gsem -, perhaps with a latent class model.

[Deborah Jehu]

Thanks for your quick reply Marcos. Do you suggest gsem because covariates can't be used in pca? I'm not sure that a latent class model is appropriate because we would like to determine which the fall risk factors are characteristic of each group (pca), as opposed to examine the probability that individuals fall into respective groups (gsem with latent class model). Thanks for your help!