Hi,
I'm conducting a study about sexting and victim blaming beliefs.
I am testing two SEM models in the same dataset: one in which the victim of forwarding of sexts is male, and the other one in which the victim is female.
So respondents answered the same items about victim blaming (a) if the victim is male and (b) if the victim is female. So the variables use in the two models are different.
I want to test whether the relations in these two models significantly differ from each other.
Is this possible? If so, which difference test can I use? I was thinking I can use chi-square difference tests but I am not sure if I can use this if the variables in the two models are different.
Thank you!!
If the models are non-nested then a direct empirical comparison of the models using the chi-square difference test, or preferably, using information criterion indices like the AIC or BIC are not appropriate.
While others would likely disagree (I've had this in reviews of a paper comparing non-nested CFA models), I would argue that you can in a real sense compare two models using the standard fit indices (CFI, TLI, RMSEA, SRMR). For example, assume model 1 displays poor fit according to those indices, and model 2 displays excellent fit. It seems perfectly reasonable to reject model 1 as a poor approximation of the data, and accept 2 as a good representation of the data. As far as I'm concerned a result like that clearly favours model 2 compared to model 1. Furthermore, lets say model 2 was also more parsimonious and had more theoretically expected model parameters. That would lend greater reason to favour model 2 compared to model 1.
If however the two non-nested models both displayed acceptable fit, it is more difficult to compare them. You would then need to rely on issues of parsimony, theoretical consistency, and interpretation of model parameters (lack of multicollinearity, stronger predictive effects etc.) to argue in favour of one model compared to the other.
Thank you for the response! Indeed, those thoughts went through my mind as well. I think it is best then if I test one model, in which I implement the variables of both male and female victims and then use parameter difference tests to compare the relations for male and female victims.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling