Anonymous posted on Wednesday, April 25, 2001 - 10:15 am
Hi. I have a latent quadratic growth curve mixture model with four latent classes over ten time points. In assessing my model, I mapped the fitted estimates to the raw data and for 3 of the groups the fitted and raw data are "matched," but for one group (11% of the sample n = 215) the last three time points do not map onto the raw data. The estimated end point is quite a bit lower than the raw data. What might contribute to this difference for this one class? Thanks in advance.
The reason that I asked about TYPE=MISSING and TECH7 is that for TYPE=MISSING, mapping onto the TECH7 values may not match because the TECH7 values come from the listwise deleted sample not the entire sample.
If you are using sample values from SPSS then I assume that you are putting people into their most likely class, whereas the model is estimated with individuals in each class fractionally. It may be that people are not as definitively classified in your last class. Also, I'm not sure how you are handling the missing values in SPSS, i.e., whether the SPSS sample is the listwise deleted sample or the entire sample.
Hi, From my understanding, mixture models are estimated from the mixture indicators distributions and not simply from their correlations/covariances. So, this logically means that mixture models should always be estimated from raw parameters and not from correlations or covariances matrices. Am I right ? Could one estimate mixture model from a correlation or covariance matrice ? This may sound like a stupid question. Sorry for this, but thank you for your time !