Hi, I had conducted CFA for my data using MLR estimator since my data is considered as continuous data. Then, I realized that there are few items that have negative factor loading although they works perfectly well in EFA (all the items lie in a factor). I try to separate the items that have positive and negative loadings and running the CFA again and obviously the factor correlation is negative which is I believe not a good sign. So, how to deal with this situation? Is it caused by different types of estimator? My friend tries to use AMOS which is the default estimator is MLE, but the items are pretty ok and he dis not face this problem (negative factor loading in CFA)
It sounds like you get negative loadings where you expect positive ones. If all loadings on a factor are negative, that just means that the interpretation of the factor is reversed and you should get a reverse sign factor correlation.