st: Correcting for autocorrelation with negative binomal regressionin Stata

[Chandy C John <chandy.john@case.edu>]

Title: Correcting for autocorrelation with negative binomal regression in Stata

Hello to the Stata group,

I am using Stata to analyze how malaria incidence relates to rainfall and temperature in an area with 7 villages. I have data for all 7 villages for each of the variables, but want to look at the overall area. Because there is overdispersion with Poisson regresssion, I am using negative binomial regresssion. 
I have 3 questions:

1) Is there a way to correct for autocorrelation between incidence months (which does occur, as these areas have outbreaks and periods of long low transmission) within Stata? Will the “robust” command do this for me? Right now I write my command line as:

nbreg cases rainfall mintemp maxtemp, ir exp(persons) robust cluster (villnum)

Will this give me values corrected for autocorrelation?

2) How do I compare models, e.g. models with different lags of rainfall and temperature, to decide which model is best? Can I use the log pseudo-likelihood as I would a log likelihood for comparison of models to see if they differ significantly?

3) Is this the best descriptive way to do this? I have not done time series analysis, which is another option, but to explain the results from time series analysis in a paper that aims to set up a predictive model seems much more difficult to me, since it does not give something like an IRR. However, my knowledge of time series analysis is very limited, and my biostats colleague has also had limited experience with time series analysis.

Any help greatly appreciated!

Chandy John
Case Western Reserve University, Cleveland, Ohio