Reasons for markedly different results using MLR vs. WLSMV in CFA?

[Tina]

A scale I examined has a 6-point likert scale and detailed anchor points. In addition, the items are distributed skew. I did a CFA with MLR as well as with WLSMV. With WLSMV as estimator, the scale performs much better in terms of fit indices (RMSEA, CFI) than with MLR.
What could be possible reasons for this? Is this solely due to the method of calculation? Or does that also say anything about my data?

[Mauricio Garnier-Villarreal]

Tina

The main reason for this is because MLR still assumes that the items are measured as continuous. So, the assumption of continuous metric is a misspecification of the model

[Tina]

Thank you very much!