Message/Author 


I am running a cfa and have found that two of the scales have both positive and negative loadings. I have examined the data set, checked the wording of the items, etc. Additionally, these scales have been used in prior research and so, from my perspective, should be valid. Do you have any thoughts as to why this might be occurring? Thank you! 


Are you saying that two of the factor indicators load on more than one factor and that on one the factors the loadings are positive and on the other they are negative? 


In an initial attempt to identify factors in the data set I am using, I did a PCA using SPSS, and the 5 indicators for the factor (teacher/student relationship) were positive. When doing the CFA using Mplus, two of these factor indicators remain positive but three are now negative. This occurred for another factor, as well. Additionally, I did an EFA using Mplus and the results were quite different from the SPSS results. I guess I am just wondering if, being a new Mplus user, I have made an error that is causing these differing results. Thank you. 


It sounds like you are not reading your data correctly. Do a TYPE=BASIC with no MODEL command and see if your sample statistics look correct and your sample size is correct. You may have blanks in your data. This is not allowed with free format data. 


Problem solved! Thanks so much for your feedback! 


Hi, I am running a multilevel CFA with 12 categorical indicators (6 binary and 6 ordinal with 3 categories). In the level 1 I have specified four factors and in level two there are two factors. After running the model, I received positive loadings for all the 12 items in the level 1. But in the Level 2, for one factor the model is yielding negative loading for three items. I am a bit confused why in the level 2 the loadings turned into negative when the corresponding level 1 pathways are positive. In addition, the residual variances of 12 indicators are positive too. Please help! Thanks in advance, Zabir 


Maybe you have few clusters. To explore in a rough way the between factor structure, you can do a factor analysis of the cluster means for the variables. 


I have 346 clusters. But for two of the binary indicators which flip signs number of positive responses are few 3% and 5%. Is that problematic? 


Could be a problem but doesn't have to be. You may want to discuss model fitting strategies on SEMNET. 

Erin Taylor posted on Monday, September 10, 2018  10:58 am



Hi, I am running a CFA for a latent construct with 5 measured indicators. 2 of the indicators have negative loadings, whereas the other 3 have positive loadings. The descriptive correlations also support this. Is it necessary for me to reverse score the data so that all loadings are in the same direction before generating factor scores? But doesn't this assume that only the "question" may be worded negatively (and not the item scaling)? What if the original directions of the item scaling are what I am interested in (for example, higher values equal greater amounts of the construct even though they are negatively correlated)  in this case, is it OK to construct factor scores with the nonreversed items? Thank you 


Having some negative loadings is not at all problematic  it's just like having negative coefficients in regression. 

Back to top 