www.cmm.bristol.ac.uk/forum

Hi - yes, the estimated (variance and covariance of the) random effects of a logit model will differ to those from a probit model due to the different link functions used. You can try this yourself, e.g. by fitting a simpler model, such as a univariate response model, and switching between logit and ...

Hi - you may find the document "Imputation for Multilevel Models with Missing Data Using Stat-JR" helpful: https://www.bristol.ac.uk/cmm/media/software/statjr/downloads/manuals/missing_data/v3.0/missing-data-with-statjr.pdf

For a 3-level model, you will need to use the Stat-JR template ...

Hi - you can assess the homoscedasticity assumption of the level-2 random effect(s) by plotting the predicted random effects, at that level, against the covariates or fixed part prediction (such plots are commonly-used to visually inspect this assumption).

Since your question concerned using MLwiN ...

Hi - the default for MLwiN (when using (R)IGLS) is to reset negative variance estimates, and any associated covariances, to zero, so it is possible that you are getting negative variance estimates during estimation for some reason, and they are being reset in this manner.

If you wished to explore ...

Hi - sorry I didn't have the opportunity to reply to your earlier post - if you just have missing data in your (univariate) response in your model of interest, then imputing values via multiple imputation will not contribute any further information to your analysis - e.g. the introductory section of ...

Hi - can I just check: did you install the C++ compiler MinGW? Instructions can be found here: http://www.bristol.ac.uk/cmm/software/s ... stallation

Best wishes,

Richard

Hi Mansoor - you might be interested in new functionality which has recently been released to support the analyses of incomplete datasets in Stat-JR (distributed with MLwiN); see http://www.bristol.ac.uk/cmm/news/2016/85.html . This new functionality includes a template (NLevelImpute) which performs ...

Note that one-sided Bayesian p-values are another option which can be helpful; these can be toggled on/off in R2MLwiN via z.ratio. There is an example in Section 6 of the R2MLwiN article ( http://dx.doi.org/10.18637/jss.v072.i10 ) recently published in the Journal of Statistical Software; as the ...

Hi - the error you are getting is from MLwiN, and it suggests it is having difficulties estimating the model you have specified (given the data). Since you have a large number of fixed effects (as each of the factors will consist of a number of levels, and so a relatively large number of dummy ...

Hi - thanks for your interest in the software. The forum message Chris linked to illustrates how to call Stat-JR from R and bring the results back. It's not actually necessary to write a Stat-JR template in order to do that (one can use the templates provided in the Stat-JR package), although it ...