Insignificant direct effect and significant indirect effect with -ldecomp- command (Mediating analysis)

[Josep Maria Caroz]

Hello! My name is Josep Maria Caroz and I am doing a research paper about the relationship between employment precariousness (exogenous variable), housing precariousness (mediator variable) and self-perceived stress (dependent variable) using generalized structural equation modelling (gsem). The reason about I wrote this post is that I have strange results that maybe could invalidate some of my conclusions. To put in context, I dichotomized all the variables that I presented above to do my model since I have a reduced sample. So, I was forced to use a gsem model to create the structural modelling since sem is not able to use categorical data.

Then, once I have constructed the model, I use the -ldecomp- command developed by Maarten L. Buis to decompose the direct, indirect and total effects of employment precariousness on self-perceived stress through housing precariousness. In this decomposition, I obtain that the indirect effect (or mediator effect) of employment precariousness through housing precariousness is significant, BUT (and here is the problem) I did not get a significant direct effect between housing precariousness and self-perceived stress (P=0,068) at the 95% confidence level. So, my doubt is that if is statistically valid this significant indirect effect and if I could present these results even they seem to be contradictory.

So much thanks,

Josep

[Richard Williams]

Welcome to Statalist.

it might help to show your code and output. Use code tags so the output is legible. See the Statalist FAQ, especially the section on asking questions effectively.