I am estimating a three-category variable, smoking status in last 12 months (Non-smoker, Daily smoker, Occasional smoker). I have raw weights from the survey, which I would like to include in my model. However, when I specify the multinomial logistic regression model to include the weights, I get this error message: Wrong length weight column (weig). This problem occurs only with multinomial logistic model; binary and linear models fit fine with the weights. I am using both the standard MLwiN version 2.02 and version 2.1 Beta 5.
Thanks,
Walt
Re: Weighting in multinomial logistic regression
Re: Weighting in multinomial logistic regression
MLwiN creates an expanded version of the dataset when you fit a multinomial model, with one row per response category per level 1 unit (look at the data for resp, resp_indicator, and [level 1 name]_long, and see the chapter in the User's Guide, to understand what it's doing). I wonder whether perhaps it doesn't know to expand the weights? You could try expanding them yourself and giving MLwiN the expanded version. You could do this by using the Merge window (Data Manipulation -> Merge(replicate)). Your Merge from ID will be your level 1 ID (whichever level the weights are at) and Onto ID will be [level 1 name]_long.
That said, I wouldn't really recommend fitting a weighted multinomial model in MLwiN. That's because you should really be using MCMC to estimate multinomial models (see the FAQ http://www.cmm.bristol.ac.uk/MLwiN/tech ... entresults), and MLwiN can't use weights with MCMC.
That said, I wouldn't really recommend fitting a weighted multinomial model in MLwiN. That's because you should really be using MCMC to estimate multinomial models (see the FAQ http://www.cmm.bristol.ac.uk/MLwiN/tech ... entresults), and MLwiN can't use weights with MCMC.