Message/Author 


I have a question that needs clarification as I am getting two different answers from various sources. I read through your handout in multi group analysis and you compare more constrained model to the less constrained model (chisquare difference (df difference)), rather than the configural (baseline) model. However, I found one book which suggests that no matter which level of the restrictiveness the model has, it has to be compered against the baseline model. Are they just two different approaches, or what I have read is not correct? 


There is no one right way to do this. The configural, metric, and scalar models can be tested against the unrestricted model. Metric and scalar can also be tested against configural. And scalar can be tested against metric. I think it is most common in testing for measurement invariance to test configural against the unrestricted model, and metric and scalar against the configural model. 


Hi Linda, Many thanks for your reply. Just to clarifyisn't configural model you are referring to the most unrestricted model, which is often referred to as "totally free multiple group model (TF)" (Hair et al., 2010). It seems that there is one model that is even less restricted than configural model in your explanation. Please could you elaborate on this? Many thanks in advance. 


Yes, configural is all parameters free across groups. The totally unrestricted model is the model of means, variances, and covariances. 


Sabrina: Please see slides 163 and 164 in the Topic 2 course handout which is on the website. 


Hi, I would like to estimate the ratio of two chisquare values (e.g., from a multigroup model) but more so to get a variance estimate of that ratio. Can Mplus model the chisquare value as a parameter? Thank you! 


I think you would want to use bootstrap to get SE for the ratio. Generate the bootstrap samples outside of Mplus and then estimate the model with multiple data sets using the type=montecarlo option of the data command. Repeat for the other model and you can get the distribution that way. 

Back to top 