Significant path becomes non sig. bec. non sig. control var?

[Bart uva]

Dear all,

For my master's thesis I am using PLS SEM.

My model has three control variables firm age, work experience at current company and overal job experience.

When I run my model without control variables I have two significant relationships (2 LV's). My three moderators have no significant effect. However, when I add my control variables, they show no significant paths, but they decrease the path's of my two LV's and turn them into non-significant relationships.

It would be a bummer if my control variables having no significant effect, still mess up the relationships. Should I therefore exclude the control variables? My sample is only N=58 implying that it might be difficult to include control variables.

Could somebody give me some advice on what I should do?

Thank you very much!

Regards,

[jmbecker]

With such a small sample size, it sounds very likely that adding additional control variables (that might also be slightly correlated with your focal variables) reduces statistical power and therefore turns your significant into non-significant relations. There might also be a case of suppression effects (which can happen even if the included variable is itself not significant).

For a master thesis, I would report the behavior of your model when including the controls and try to explain it. This shows that you know what you are doing. Just skipping the control variables would not be a good approach.
Additionally, you might consider collecting more data. 58 observations are in any case not enough to generate stable model results and draw generalizable conclusions.