

Evaluating Impact of Nonequivalence A... 

Message/Author 


Hello, Thanks in advance for your continued help. I am trying to evaluate findings from a multiple group comparison study I did where fit statistics indicate partial invariance due to unequal item intercepts for 23 of the 7 items in my construct. I read Oberski's recent paper (2014) on the EPCinterest and was wondering if you new of any people or papers that have calculated this statistic in MPLUS? Alternately, I am wondering if I could demonstrate the impact of this invariance on latent variable means simply by predicting the factor score of my latent construct under two conditions: once with equal item intercepts across groups and once allowing the twothree intercepts that are unequal to vary. (Since I've established factor loading invariance, I'm assuming that I don't need to demonstrate the impact on regression coefficients). Does this approach seem reasonable? Sarah 


Just a follow up on that last question. I meant to ask if I could demonstrate the impact of this invariance on latent variable means by looking at the difference between zero (what the latent means are estimated at when factor loadings and intercepts are held constant) and whatever the difference is when the two intercepts that vary are allowed to vary. Sorry for my confusion! Thanks again for your help. Sarah 


Yes, you can learn about this by comparing the relationships between the estimated factor means with and without allowing for partial measurement intercept invariance. With many groups you can plot the two sets of factor means as in Figure 2 of Asparouhov, T. & Muthén, B. (2014). Multiplegroup factor analysis alignment. Structural Equation Modeling: A Multidisciplinary Journal, 21, 114. DOI: 10.1080/10705511.2014.919210. An earlier version of this paper was posted as web note 18. Mplus input, output, and data files for Web Note 18 Monte Carlo simulations are available here. 


We have not implemented Oberski's approach  I think you refer to the one where you get EPC for other parameters than the one freed. 


Thank you so much! As always, your responses are very very helpful. Sarah 

Back to top 

