Wen-Hsu Lin posted on Monday, October 26, 2015 - 12:48 am
In the SAMPLE STATISTICS section, should I expect the ESTIMATED SAMPLE STATISTICS to be identical to SPSS output? I am asking because I saw very strange outcome in the ESTIMATED SAMPLE STATISTICS. For example, I declare date as categorical and it showed number of individuals in each groups and the proportion, which is identical to the SPSS output. However, the estimated threshold was negative (-12.486). When I checked another continuous variable, pbad, the estimated mean from Mplus was 9.928 but the same number was 12.265. It should be identical, isnt it? Thank you
SPSS provides sample statistics for the number of observations that have no non-missing data for a variable. The sample size can vary from one variable to another. Mplus as the default uses all available information. It is also possible that you are reading your data incorrectly. Check your sample size and be sure there are no blanks in your data set. The sample statistics between SPSS and Mplus will agree if the same set of observations are involved.
Wen-Hsu Lin posted on Monday, October 26, 2015 - 6:32 pm
Continue on the same topic. I ran a null model just to check if I read my data correctly. The mean and frequency distribution were almost identical. This provided me the confidence that I did not read my data incorrectly. However, when I ran my final model (type=mlr, integration=montecarlo), the categorical data still had very similar distribution but the mean for the continuous variable was very different. Even when I ran with ML, the mean of the continuous variables were very different. For example, in the null model the mean for the communication were 2.846(n=2690)and 2.849(n=2582) from SPSS and Mplus, respectively. However, the mean became 6.623(n=2582) at the sample statistic section in the Mplus. Is this because model specification?