My dataset is the following:
- 1 IV (independent variable), which ranges from 1 to 10 and I treat as continuous
- 1 M (mediating variable), which is dichotomous
- 1 DV (dependent variable), which is dichotomous as well
- 10 control variables, which are all categoric. Hence, when split up in dummies, I have around 50 dummy variables. 36 of them describe the country of origin.
- 3100 different firms. Important: these firms have responded to the survey in consecutive waves. I have pooled their responses. As some firms have participated in plenty of consecutive waves, I have 3500 of such pooled observations. I cluster the standard deviations around the firm id to account for potential heterogeneity caused by this.
My code is the following
Code:
sem (IV -> M, ) (IV-> DV, ) (controls-> M, ) (controls-> DV, ), vce(cluster firm_id) nocapslatent
However: I am now interested in goodness-of-fit statistics.
-
Code:
estat gof
Code:model was fit with vce(cluster); only stats(residuals) valid.
- Am I right to assume that this perfect fit is, probably, caused by the many dummy variables in the model?
- If so, what would you suggest I do?
- More importantly: I would like to have a chi-squared gof and the other gof-statistics, which are now not generated. Any idea how I can calculate these given the above mentioned constraint that the model was fit with vce(cluster)?
- I am yet to find an empirical paper which follows this methodology, that is mediation with SEM when there are no latent variables. If you know one, please share!
0 Response to SEM mediation analysis - goodness-of-fit
Post a Comment