

Negative Variances in Bifactor Modeling 

Message/Author 


Dear Bengt and Linda: My postdoc and I are running some bifactor models and are getting one negative residual variance. Our input looks like this: ANALYSIS: ESTIMATOR IS MLR; Type is Complex; !School is the clustering variable; ITERATIONS = 10000000; MODEL: ASrch by AAIM01_1* AAIM02_1 AAIM04_1 AAIM08_1 AAIM10_1; ACom by AAIM03_1* AAIM05_1 AAIM06_1 AAIM07_1 AAIM09_1 AAIM11_1 AAIM12_1; General by AAIM01_1* AAIM02_1 AAIM03_1 AAIM04_1 AAIM05_1 AAIM06_1 AAIM07_1 AAIM08_1 AAIM09_1 AAIM10_1 AAIM11_1 AAIM12_1; ASrch@1; ACom@1; General@1; General with ACom@0 ASrch@0; The variance for AAIM12_1 is negative and significant. Should we set the negative variance to zero? Is that OK to do? Thanks so much. Seth Schwartz 


You should not fix a significant negative residual variance to zero. This calls for a change in the model. 

Back to top 

