Mixed-effects, mixed distribution model

[turrell]

Dear George,

I have collected three waves of data (2007, 2009, and 2011) on the number of minutes that people report walking for transport in the previous 7 days. The sample-size at each wave comprised 200 neighbourhoods and 11,000, 7900, and 6900 respondents in 2007, 2009 and 2011 respectively. Hence I have a three-level repeated measures dataset (neighbourhoods, individuals, time).

The variable at each wave has an excessive number of cases with zero-values (~60%), reflecting the fact that most people didn't walk for transport during the survey reference period; the rest of the cases have non-zero values that range from 1-840 minutes. Hence, the non-zero data arise from a continuous distribution, and are not independent counts.

Tooze et al (Statisitical Methods in Medical Research, 2002;11:341-355) propose a model for repeated measures data with clumping at zero, using a mixed effects mixed distribution model with correlated random effects. The model contains components to model the probability of a non-zero value and the mean of non-zero values, allowing for repeated measurments using random effects and allowing for correlation between the two components. They used the MIXCORR macro in SAS PROC NLMIXED.

Can this type of model be estimated using "runmlwin"?

Many thanks in advance. Regards.
Gavin

[GeorgeLeckie]

Hi Gavin,

No I'm afraid you can't fit such models (nor zero inflated count models) in MLwiN and therefore not in runmlwin either.

I think from within Stata, your only options would be gllamm which I think would be rather slow with your data dimnsions, or possibly sabrestata. The latter is fast and can certainly fit some models with excess zeros, but can't include random slopes.

Best wishes

George