]]>

1. If you are interested in inequality, it is not at all obvious you'd want to drop any non-stunted/wasted children. Inequality is usually about the spread of the entire distribution, not the right tail.

2. It is not at all clear why you'd be worried about the R^2 of the regression... what is your model trying to do that makes the R^2 matter in a way that the standard errors on your variables of interest aren't the more important measure of precision? I just mean - why do you care about the R^2?

3. You can't really use a log model with negative values, so it is hard for me to understand how you tried that and it didn't work.

4. If you gave me a little bit more context I might be able to offer more concrete suggestions, but in general I think this question is too broad of a statistics question to be appropriate for the DHS staff on the forum. It sounds more like you need a statistical consultant, and while there are some commenters here like me who can sometimes help, it is really outside the bounds of what this forum is designed to do.]]>