search

mkao006 / sws_imputation

Repository for implementing imputation for the Statistical Working System
0 stars 0 forks source link

MLE vs REML #44

Closed mkao006 closed 11 years ago

mkao006 commented 11 years ago

Should MLE or REML used to estimate the model? To test the null and the grouped average model, MLE should be used for likelihood comparison. Need to check whether the variance component is used and whether the sample size is large enough.

mkao006 commented 11 years ago

Given a large sample, the bias of the variance estimation is small for ML. In addition, for model comparison ML is more suitable as exact same fixed effect is required for REML.