Message/Author |
|
|
Hi, Is there a way I can make indicators on separate factors orthogonal to the other factors and their indicators using the monte carlo command? I'd like to produce a data set of an ideal CFA with items loading on intended factors at .6 and no modification indices from each factor to items that aren't supposed to load onto them (correlations between indicators on the same factor would be fine though). I thought adding f1 BY y1-y6*.6 y8-y36@0; f2 BY y7-y12*.6 y1-y6@0 y13-y36@0; f3 BY y13-y18*.6 y1-y12@0 y19-y36@0; f4 BY y19-y24*.6 y1-y18@0 y25-y36@0; f5 BY y25-y30*.6 y1-y24@0 y31-y36@0; f6 BY y31-y36*.6 y1-y30@0; and y1-y6@.6; y7-y12@.6; y13-y18@.6; y19-y24@.6; y25-y30@.6; y31-y36@.6; would do the trick, but I'm still getting small modification indices from every indicator to every factor. Thanks, Mike |
|
|
In any given sample you can get MIs that are not zero even when the population parameter value is zero. It is not clear if your residual variance statements like y1-y6@.6; refer to Model Population or Model. If the former, that is not correct and would be another reason for non-zero MIs. To see how to set up Monte Carlo runs, see the Monte Carlo counterparts to the UG examples which are on the Mplus CD and also on the web site with the UG. |
|
|
Thank you for your support! I've been working at the models using the examples from ch 5 & 11. I've run into an issue that makes me unsure if I'm setting it up right. This is my syntax for a 6-factor (6 indicators each) model with 6 bad cross-loadings each (the .5 loadings). montecarlo: names = y1-y36; nobs = 200; nreps = 1; save = highhighhigh.dat; model population: [y1-y36*0] y1-y36*1; f1 BY y1@1 y2-y6*.6 y7-y12@.5 y13-y36*0; f2 BY y7@1 y8-y12*.6 y13-y18@.5 y1-y6*0 y19-y36*0; f3 BY y13@1 y14-y18*.6 y19-y24@.5 y1-y12*0 y25-y36*0; f4 BY y19@1 y20-y24*.6 y25-y30@.5 y1-y18*0 y31-y36*0; f5 BY y25@1 y26-y30*.6 y31-y36@.5 y1-y24*0; f6 BY y31@1 y32-y36*.6 y1-y6@.5 y7-y30*0; f1-f6@1; f1 WITH f2-f6*.3; f2 WITH f3-f6*.3; f3 WITH f4-f6*.3; f4 WITH f5-f6*.3; f5 WITH f6*.3; y1-y6*.36; model: f1 BY y1@1 y2-y6*.6 y7-y12@.5 y13-y36*0; [repeats, same as above] f1 WITH f2-f6*.3; [repeats, same as above] y1-y6*.36; Analysis is done with another syntax file with simple syntax for the intended simple structure of the model (no cross-loadings, only the 6 intended indicators loading onto each factor). |
|
|
(sorry, the syntax pushed me over the message length limit) Fit indices come out too good (CFI = .99, RMSEA = .026) for this model for me to think that the data are coming out as misspecified as I intend it. Is there an example in the manual/online of setting up models with cross-loadings in the data that are not in the model? Thank you. |
|
|
In your model statement you should not fix the cross-loadings as you do in for instance y7-y12@.5. But then you say that analysis is done with another syntax file and it sounds like that's the one that think gets too good fit indices. We need to see that file. This sounds like it is best handled via support so please send it there. |
|
Back to top |