|From||Chandy C John <email@example.com>|
|Subject||st: RE: RE: Poisson estimators for space time series in Stata|
|Date||Tue, 18 May 2004 08:29:41 -0400|
Dear JP and Scott,
Thanks very much for your input. JP, you are correct that since the data is spatial, I need to use a method that allows for cross-sectional correlation. Right now I am using the villages within the area as the cross-sectional areas, and occurrence of malaria in them is highly correlated. So it appears that xtgls would be the code to use, because with xtgee independence of observations across villages is assumed.
If I express the malaria cases as incidence/area (rather than cases/area), will this get around the problem of xtgls not being usable with count data?
The other choice is to use ARIMA, but as best I can tell, this would allow me to only use the aggregate data for the whole area (there’s no “cluster” command, which would allow me to use the data for individual villages), and I’m not sure of the minimum number of observations that would be considered appropriate if one wants to use ARIMA.
Any further input greatly appreciated!
If my understand is correct xtgee does not allow for cross-sectional
correlation, which might be a relevant issue if you are using spatial data.
In order to take that into account you would have to use either xtgls (if
N<T) or Timothy Conley's code, presented in his paper "GMM estimation with
cross sectional dependence" Journal of Econometrics (92), 1999, and
available for download from his website (try to "Google" his name). The
problem with both codes is they are not for count data.
My guess is that Stata does not have a straight forward solution for your
All the best,
[ mailto:firstname.lastname@example.org ] On Behalf Of Scott Merryman
Sent: Friday, May 14, 2004 5:06 PM
Subject: st: RE: Poisson estimators for space time series in Stata
How about -xtgee-?
xtgee var1 var2, link(log) family(poisson) i(id) corr(ar1) t(time)
[ mailto:email@example.com ] On Behalf Of Chandy C John
Sent: Thursday, May 13, 2004 9:24 AM
Subject: st: Poisson estimators for space time series in Stata
Are there any ado files or other programs that will allow Poisson estimators
for space time series in Stata? I have some rain and temperature data that I
wish to compare to malaria incidence over time, and with regular Poisson or
negative binomial regression, I face the problem of autocorrelation between
months of malaria incidence. I have done time series analysis with the ARIMA
command, but I am not sure if it is entirely appropriate for the relatively
small dataset I have (rain, temp, malaria cases over 36 months).
I have found a program within R that will do Poisson estimators for time
series (PESTS), but I'm not familiar enough with R to know how to input my
data. Does anyone know of any Stata programs that will incorporate moving
averages and autoregression into Poisson estimations?