Modified Poisson Regression using GLM: missing categories

[Olami Mario]

I want to know the effect of selected background characteristics on intimate partner violence among women in 35 low and middle-income countries using demographic and health survey (DHS) data. To account for fixed effects of the surveys, I generated variable - survey, assigned a code to each survey and included i.survey in the model. The relative risks of IPV were derived using modified Poisson regression which was estimated via a glm command.


my code is: glm sexvio i.age i.literacy i.occupa i.survey, link(log) family(poisson) eform


**this code worked well until I introduced the svy command- see below


svy: glm sexvio i.age i.literacy i.occupa i.survey, link(log) family(poisson) eform


aside from the reference category, the output missed out information on three surveys out of 35. there was no information on why they were missed.


Pls, how can I solve this problem?

[Carlo Lazzaro]

Olami:
are you sure that the three omitted surveys have no missing values in any variable?

[Olami Mario]

Hi Carlo,

Thank you. I just checked through each of the missing surveys, there are no missing values in any of the variables.

[Olami Mario]

Is this correct?

I once read svy don't work with poisson, hence, using weight with robust can be used to obtain the irr.

Thus, I used the code below, the model gave estimates for all variables. There is very little difference in the derived estimates when compared with those I earlier obtained using svy:glm.

However, asking experts on this platform, please is the code below correct:

glm sexvio i.age i.literacy i.occupa i.survey, link(log) family(poisson) eform robust, [iw=dwt]

[Carlo Lazzaro]

Olami:
not quite.
-svy- allows -poisson- regression model, as you can read in -help svy_estimation-.
As an aside, I wopuld check for the usual nuisance of -poisson-: extradispersion (that, in almost the majority of cases means overdispersion; see -nbreg-, that supports -svy- as well).

[Olami Mario]

Thank you

[Olami Mario]

I used the Stata code - edit if !e(sample) posted by Nick Cox in the thread below, and found one of the variables used to derive svyset had lots of missing values. I have made corrections, my model works well now. Thank you all. This platform helps a lot

https://www.statalist.org/forums/for...gression-model

[Nusrat Jahan Sathi]

glm y i.x1 i.x2 i.x3 i., link(log) family(poisson) eform

Using the above command, I have found IRR. My data is from a cross-sectional survey. But I need to estimate PR (Prevalence Ratio) from Poisson regression. Are IRR and PR the same?

[Carlo Lazzaro]

Nusrat:
Clyde Schechter has already provided you with an excellent reply at https://www.statalist.org/forums/for...error-variance.
Hence, it's up to you to discover (and, if still interested, share) whether your cross-sectional data are the total (old and new) cases (prevalence) or the new ones (incidence) only of a given whatever.

[Nusrat Jahan Sathi]

Thank you very much, Carlo Lazzaro.