I am working on a project where I am looking at Egyptian child health outcomes using the 2014 DHS. I am interested in looking at within family variation, and so I want to create a "siblings only sample". This would allow me to have "mother fixed effects."

Could someone give me some advise on how to do so?

]]>

Are you looking at outcomes that are only available for children under age 5? If so, you can only use children (and siblings) who are in the KR file. I just looked at that survey. 47% of women had no births in the past 5 years. 35% had 1, 16% had 2, and only 2% had more than 2.

You are probably thinking of random effects for mothers, rather than fixed effects. For fixed effects you would have a coefficient for each mother--thousands of coefficients. Random effects would make more sense but would be harder to calculate because the ratio of children to mothers is small (for statistical purposes).

If a woman had only one child in the past five years, that child would have no siblings. Do you drop children with no siblings?

Say a woman has had 3 children in the past 5 years. Call them A, B, C. Do you want to construct pairs AB, AC, BC? Note that each child will appear in two pairs. Is that what you want?

]]>