I am running a SEM model examing how exercise avoidance (EAM) mediates the association between weight stigma (SSI and WBIS) and physical activity (PA).
There are also multiple covariates in the model.
EAM and PA are two latent variables in the model, and the other variables are all observed.
MODEL: EAM BY eam_gym eam_thin eam_emb eam_now; PA BY pa_vig pa_mod pa_walk;
PA ON ssi wbis EAM dif_wgt bmi male age univ hinc tss; EAM ON ssi wbis dif_wgt bmi male age univ hinc tss;
For publication, we received a request to include "weight change (dif_wgt)" in the model as a covariate.
The problem is, after adding it in the model, one item for PA, "pa_mod (moderate physical activity)", becomes non-significant. It was significant without it in the model.
The estimated indirect effects after 5,000 bootstrapping iterations are still identical to the previous estimates, but I am worried about the non-significance of the item in the final model.
FYI, Cronbach's alpha of three variables for PA is >.60.
I would really appreciate it if you can provide some advice on this issue. Maybe I should exclude it from the measurement model? Or would it be ok if I report the current results with non-significant item for PA latent variable in the model?