I am running a SEM model using lavaan package in R. The idea is to check different mediation effects of social class of origin, cognitive abilities, education and personality on the log-income of the offsprings. Basically, this is the path-scheme I designed:

path_male <- '
#latent variable 
ability =~ c_cgwri_dv + c_cgwrd_dv + c_cgs7ca_dv + c_cgvfc_dv + c_cgna_dv

ability ~ origin_class
education ~ ability + agree + conscientiousness + extrov + neurotic + openness + origin_class 
log_income ~ education + origin_class + ability + agree + conscientiousness + extrov + neurotic + openness + job_type 

c_cgwri_dv ~~ c_cgwrd_dv
c_cgs7ca_dv ~~   c_cgna_dv
ability  ~~ origin_class
openness~~ origin_class
neurotic~~ origin_class
extrov~~ origin_class
conscientiousness ~~neurotic
conscientiousness ~~extrov

The resulting estimates are controversial. Indeed, all the measures to check the fit are quite good: CFI= 0.978; TLI=0.96; RMSEA=0.028; SRMR=0.020;
but the p-value of the chi-squared is 0.000 suggesting that the model is a poor-fit.

I also used the modification indices to check what additional paths/correlations are necessary to improve the fit, but those with high values (around 20/30) are not theoretically meaningful. What could be the reason of this contradiction between chi-squared and the other measures of fit?



1 Answer 1


I just go through the Saris-Satorra (2009) approach focusing on the MI and EPC. Therefore, with not relevant misspecification detected with the command:

 modificationindices(fit, sort. = TRUE, power = TRUE, delta = .4, op = "~~") 

(and repeat it for loadings), I can accept with confidence the model even if the p-value on the chi-squared is less than 0.05.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.