
Message/Author 


I have a quick question re: the possibility of overfactoring. I have been running a few EFA's where the model fails to converge at higher levels of factors (i.e., the model is fine until say, 5 or 6 factors). In examining the output, this does not appear due to negative residual variances (in fact, I eliminated one manifest indicator due to this issue which had poor internal consistency anyway). Is it likely that if, say, a fourfactor model appears to account for the lions share of the variance (i.e., strong model fit  RMSEA, CFI, etc.) that the failure to converge at higher factor levels is simply a product of the fact that no higher number of factors can explain the data. I realize there is no 'simple' answer here, but rather am looking for the most likely explanation (as an aside, my sample N = 259, 13 manifest indicators, and am proposing a lower and upper bound of factors at 2 and 6). 


This is probably the result of overfactoring. There is likely no clear 5th or 6th factor. The construction of the items should suggest the correct number of factors to aim for. 


Thanks, Linda. Relevant to the above, am I correct in my understanding that if a failure to converge at higher order factor solutions is due to a negative residual variance of a given manifest indicator, then this indicates that all of the variance in that indicator has been exhausted via the latent factors (i.e., a higher order solution can not be obtained because there is no variance left in a particular indicator?). Or is that only one possible reason? 


I think that is only one possible reason. I think I can imagine cases where with m factors you have a negative residual variance, whereas with m+1 factors you don't because the factor loadings arrange themselves differently. 

Back to top 

