I have a set of crime count data where it appears that the data take on a negative binomial distribution. I have had some success converting the dependent variable (a crime count) into a rate and then log-transforming it and using a log-log Spatial Durbin model or Spatial Lag and Spatial Error models.
Conversely, I have run a negative binomial regression on the count data (with a population offset) with no consideration for the spatial autocorrelation.
I am wondering if anyone has found a reliable way to apply spatial autoregressive techniques to GLM's (NB, Poisson). If anyone has been successful with this in Stata, I would be particularly curious to learn how.
If I were to run -nbreg- but applying my weights matrix and lagging the dependent variable, am I able to directly interpret the coefficient on that term as I would the rho term in a spatial lag regression (of course with considering how the NB coefficients differ from ones generated by an OLS)?
أكثر...
Conversely, I have run a negative binomial regression on the count data (with a population offset) with no consideration for the spatial autocorrelation.
I am wondering if anyone has found a reliable way to apply spatial autoregressive techniques to GLM's (NB, Poisson). If anyone has been successful with this in Stata, I would be particularly curious to learn how.
If I were to run -nbreg- but applying my weights matrix and lagging the dependent variable, am I able to directly interpret the coefficient on that term as I would the rho term in a spatial lag regression (of course with considering how the NB coefficients differ from ones generated by an OLS)?
أكثر...