Message/Author 


I'm testing some nested models, using either the Wald test (using model test: ), or using the Satorra method described here: http://statmodel.com/chidiff.shtml . However, I'm getting quite different results. With the Wald test, I get a chisquare difference of about 12 (with 4 df), with the chisquare difference test, I get 32. (I just swap "model test:" for "model constraint:", to make sure the models are the same). Is this because the constraint is fixing a correlation to 1.00, and therefore I'm having a boundary problem? Or is it something else? Which (if either) of these results should I trust? Thanks, Jeremy 


I can't answer this without seeing exactly what you are doing. You should get the same results using the Wald test versus testing two nested models. I have seen examples where these are the same. So perhaps there are some defaults that change. If you send the inputs, data, outputs, and your license number to support@statmodel.com, I can take a look at it. 

Kihan Kim posted on Wednesday, September 22, 2010  3:42 pm



I was going to run Wald chisquare difference test with one degree of freedom by constrainting one path (i.e., cc on itv) to 0. I used the following syntax, but I get the same chisquare value as the model without the MODEL TEST command. Is there anything wrong in my syntax? Model: dv3 on dc cc; dc on itv ts pd cent form12; cc on itv(p1) ts pd cent form12; dc with cc; itv with ts cent form12 pd; ts with cent form12 pd; cent with form12 pd; form12 with pd; Model Test: p1 = 0; 


The test of model fit does not change when MODEL TEST is used. What you are testing in MODEL FIT is the same test that is done using the ztest from the regular output. The following statement cc on itv(p1) ts pd cent form12; should be cc on itv(p1) ts pd cent form12; I think if you look at your output you will find everything after (p1) is ignored. 


Hello, I would like to know if I can use the MODEL TEST command to compare the strength of two different path coefficients predicting the same outcome. I am worried that if the two predictors have a different scale of measurement, then constraining them to equality may not be meaningful, if the constraint applies to unstandardized path coefficients. But if the test allows to compare the strength of the standardized coefficients for the two paths, then this strategy seems OK. Thanks for your help. 


You would need to compare the standardized coefficients in this case. You can use MODEL CONSTRAINT to define the standardized coefficients and then test their difference. 

Back to top 