Fixed effects for survey data

[Shayan Khan]

Hi,

I am using a standard DHS-VI survey dataset for Pakistan 2012-13. It is a survey cross-section data with sample weights. My response variable is a binary variable indicating whether a child has ARI or not. On the right hand side I have some covariates. I want to control for heterogeneity across districts but employing the fixed effects model by using districts as my group variable. Now, there are two ways that I am familiar with that I can use to this.

1- svy:logistic y x1 x2 x3 i.dist
2- xtlogit y x1 x2 x3, fe or (when I use weights here it gives error 'weights should be same across all groups)

I want to know which approach would be the right way to do this as I know fixed effects are usually done on panel data. I was told that if I use the 1st approach then I might face 'incremental parameter problem'. Please help.

[Carlo Lazzaro]

Shayan:
welcome to the list.
-I would go with the first of your code, since you're dealing with a survey dataset;
-about the incidental parameter problem, you may want to take a look at: http://www.econometricsbysimulation....blem-with.html

[Shayan Khan]

Thanks Carlo for your response.

Is there any more reasoning that you can provide besides the fact that I am using a survey dataset so 1st approach is preferred?

I recently read in Woolridge's textbook on cross-section and panel data that incidental parameter problem is not an issue with the logit model though it is with the probit model.

[Carlo Lazzaro]

Shayan:
- no, I have no other reason to suggest your first code but te veidence that you're dealing with a survey dataset, in which the same variables might have measured >1 times on different persons rahter than on the same sample of panelid;
- as far as I know, Stata propose conditional fixed effect for -xtlogit- and no fixed effect altogheter for -xtprobit- and wisely so as, in the latter case (please, see -xtprobit entry in Stata .pdf manual)

..there does not exist a sufficient statistic

allowing the fixed effects to be conditioned out of the likelihood...

[Shayan Khan]

Yes, I know we cannot use fixed effects with probit or xtprobit. Thanks Carlo.