I modeled a 3 factor model with 4 item loadings on 3 a priori (trait) factors and cross loadings for all 12 of my items on a fourth (method) factor. I set the loadings so that the loading on the a prior factor + the loading on the 4th (method) factor (total loading) squared plus the residual variance equaled unity. For example, a total .7071 factor loading results in an a priori factor loading of .4243 and a 4th (method) factor loading of .2828 and a single residual variance of .5. However, there is a problems becuase my model often does not converage (37 of 50 replications) and the coverage statistics are out of the 95% range. Sorry, I would have posted the syntax below, but ran into posting limitations. Thanks.
The covariances between the methods factor and the other factors must be fixed at zero for model identification. This is the same as a general-specific factor model. See short course Topic 1 slides 156 and 157.
Hi Linda. I had my graduate student (Tae) post that question for me b/c I was waiting for my password to get reset but got anxious to pose my question. Anway, I did fix my covariances to zero originally (see MODEL syntax below). Any suggestions?
MODEL: [x1-x12*0]; x1-x12*.5; LV1 BY x1*.4243 x2*.4243 x3*.4243 x4*.4243; LV2 BY x5*.4243 x6*.4243 x7*.4243 x8*.4243; LV3 BY x9*.4243 x10*.4243 x11*.4243 x12*.4243; LV4 BY x1*.2828 x2*.2828 x3*.2828 x4*.2828 x5*.2828 x6*.2828 x7*.2828 x8*.2828 x9*.2828 x10*.2828 x11*.2828 x12*.2828;
Hello, I am having convergence problems estimating a multiple group CFA model. I modeled a 4 factor model with items loadings on 3 a priori factors and cross loadings for all these (continuous) items on a fourth (method) factor that also has five unique (categorical) indicators. Here is my model specification:
MODEL: ERS by ERS1-ERS5 (1) instmot1-instmot4 intmat1-intmat4 scmath1-scmath5 matheff1-matheff8; INTMAT BY instmot1-instmot4 intmat1-intmat4; SCMATH BY scmath1-scmath5; MATHEFF BY matheff1-matheff8; ERS with INTMAT@0; ERS with SCMATH@0; ERS with MATHHEFF@0;
I tried an alternative model: multiple group EFA using orthogonal rotation that converged after a long time (7h 35’) but unfortunately I noticed that only in the first group the correlations between the factors are zero.
My questions are: 1) Is there another way in Mplus to compare group means besides multiple group analyses? f.e is it possible to override the default of factor means=0 in a single level analysis? 2)How can I make sure that the correlations between the factors in the multiple EFA for all the groups are zero (and not only for the first group)? 3)Why doesn’t the multiple group CFA converge? Thanks for your help!