|
|
Establishing model-fit with count ind... |
|
Message/Author |
|
Paul Agius posted on Thursday, April 16, 2020 - 5:17 pm
|
|
|
I'm writing a grant proposal where in the mixture model (LCA) measured indicator data will be counts - link functions are likely to be poisson, negative binomial etc. I'm likely testing 2-5 classes with between 10-25 indicators (potentially more). Are there model-fit strategies for count data given chi-square and LR chi-square would to be likely problematic given the large degrees-of-freedom implied here. Are there strategies others have employed with similar data - e.g. some form of grouped, model-based, probabilities compared to observed?? If so, any guidance on how one might go about undertaking this in the Mplus framwework would be appreciated Thanks in advance Paul |
|
|
I think a good way to choose between models is to use BIC. Mplus also provides fit information for univariate and bivariate cases using TECH10. Have a look at how we did it in a longitudinal counts context in Section 6.5.1 of this paper which on our website under Papers, Growth Mixture Modeling: Muthén, B. & Asparouhov, T. (2009). Growth mixture modeling: Analysis with non-Gaussian random effects. In Fitzmaurice, G., Davidian, M., Verbeke, G. & Molenberghs, G. (eds.), Longitudinal Data Analysis, pp. 143-165. Boca Raton: Chapman & Hall/CRC Press. download paper contact first author show abstract |
|
Paul Agius posted on Sunday, April 19, 2020 - 6:37 pm
|
|
|
Thank you for your prompt and helpful reply, Dr. Muthen. After playing around with my pilot data, am I correct in saying the univariate and bivariate residual tests in Mplus are limited to models with count data with relatively small maximum counts (say 10 or less?)? I have indicators which are scored/measured with much higher values and variance? Also, are LMR and BLRT tests of comparative fit as problematic as the general chi-square measure of fit for the data, in the context of count indicators with relatively large values? Thanks again for your advice. Cheers Paul |
|
|
Q1: Right Q2: I wouldn't use those tests but instead BIC only. |
|
Paul Agius posted on Monday, April 20, 2020 - 7:29 pm
|
|
|
Thank you. |
|
Back to top |
|
|