Monday, September 20, 2010

Going Meta, What a Meta-analysis tells you

I have not been updating the blog much lately. I have been hard at work on a project funded by the British Government's Department For International Development.

The project is a meta-analysis of how giving poor families money influences their children's height. A meta-analysis means I'm examining a bunch of different studies (a couple of dozen is what we found after weeks of searching databases). Why children's height, because it's a good indicator of health?

I'm not ready to be cited yet, but I have learned there really aren't that many studies of how cash influences children's health (a couple dozen). Cash programs (and particularly those with conditional health check ups and schooling) seem to be extremely popular with governments and aid agencies.

I hope our study sheds a little light on what might influence a program's effect on children's height (and health). Since most of the support for these cash payments is based off of results on a handful of programs continuing to analyze differences between countries is important.

A study in Brazil (Morris et al. 2004) where a cash payment actually decreased child height, gives us a little pause. The authors suggest that families feared they might lose the transfer if children were too healthy (even thought this wasn't the case). This shows even if a type of program would work well in all places, you might not get a well functioning program everywhere.
Bookmark and Share


Paul Bailey and Louise Schoggen said...

What does, "beneficiary children were 0.13 Z-scores lighter" mean? Also, I'd think it would be more like the English milk study, but I just read the abstract.

Mike said...

paul&louise, the quoted text means that the kids who received the cash gained, on average, less weight than did those who did not receive the cash. a z-score is a statistical measure, you can find more here:

As to the topic of the post, .13 seems to be a relatively weak correlation, which brings me to my next point: is this a (weak) correlation = causation argument? i'm not saying cash transfers are awesome or anything, just that the article linked here seems to make some big leaps in my mind

Seth Gitter said...

Mike and Paul,

Thanks for the comments, should have answered Paul's earlier.

First children are weighed then given a z-score based on a standard normal distribution. (Paul is an econ grad student, if you aren't z-score = 0 is an average child, but falling .13 z-scores doesn't mean falling 13%, but we can think of it as a slight decrease.)

Now to answer Mike's question about causation and correlation. Brazil's program was tested with a randomized experiment. So half of the kids got money from the CCT and half didn't. The hope is if the children getting the money and getting nothing are the same, causation = correlation.

I have seen another recent study that suggested problems with the study I cite. I should dig that up.