What works to reduce child marriage?

Child marriage, despite being against the law in a lot of places, is still fairly common.   One estimate, cited in a new paper by Nina Buchmann, Erica Field, Rachel Glennerster, Shahana Nazneen, Svetlana Pimkina, and Iman Sen, figures there will be 142 million new child brides between 2011-2020.  
 
So what might a policymaker do?  One option would be to empower girls.  Another would be to give the household (specifically the girls’ parents) a transfer conditional on waiting until they’re older.   Buchmann and co. work with Save the Children in Bangladesh (the country with the second highest child marriage rate in the world) to try these out.  
 
Working within the infrastructure of an existing Save the Children food security program, this program was rolled out (randomly) to three different groups across six sub-districts in south central Bangladesh.  One group received an empowerment program.   This program was implemented by the Bangladesh Development Society and included a safe space for girls to meet and socialize as well as a curriculum around basic literacy, numeracy, life skills, nutrition, and reproductive health delivered by peer educators.   The second group received a transfer of cooking oil (with a value of $16 per year for 2 years) conditional on their daughters’ not getting married until 18 (the legal age of consent).   Only girls could pick up the transfer.  Finally, a third group got both the transfer and the empowerment program.  
 
Buchmann and co. collect a serious amount of data to look at the impacts of this program.   Their overall sample is 15,739 girls, with double shares in both the empowerment and control group.   The program ran from the end of 2007 to August 2010, and they go back about 4 and a half years later to interview folks. 
 
What do they find?   The conditional transfer reduced child marriage (i.e. under age 18) by 23 percent (or 6.3 percentage points).   The empowerment training had no significant impact.   And the combination of training and the transfer had no impact beyond the transfer.   Looking at the distribution, the program seems to have shifted the pre-18 marriages to the years just after 18, with treatment and control converging around 22.  
 
The transfer also seems to have delayed age of first birth.    Teenage childbearing for girls in these households drops by 13 percent (2.9 ppts).  There is no significant impact of the empowerment program – either alone or when combined with the transfer.  
 
Finally, Buchmann and co. look at schooling outcomes.   While the program was open to both in and out of school girls, Buchmann and co. focus on the girls in school at the start of the program since, as they observe, once girls drop out, it’s unlikely they will go back.   The transfer results in a 12 percent (3.3 ppt) increase in the likelihood that girls were in school at age 22-25 (significant at 10 percent).   Again, there doesn’t seem to be much impact from the empowerment program.   The one exception is when they focus on the girls who were 15 at the program start (and hence got the longest exposure).   These girls show a 10 percent (2.9 ppt) increase in the likelihood of being in school (significant at 10 percent).  
 
In addition to the interesting overall results, there are a couple nice methodological points in this paper.   First, before discussing the results, Buchmann and co. tell us explicitly what effect size they were powered for in each arm. 
 
Second, they spend a fair amount of time talking about their data check methods – which included not only backchecks, but follow-up phone interviews for households who showed backcheck errors in a lot of different variables. 
 
Third, they are quite transparent about their attrition/missing data.    They faced a range of issues.   First, the data entry firm lost some hard copies of data and also mis-entered ID numbers.   Second, there was a cyclone that displaced entire communities.    Finally, there is your garden variety difficulty in tracking 4.5 years later.   It’s refreshing to see this level of honesty on the things that often go wrong, as well as the effort Buchmann and co. spent to figure it all out (and, of course, to show us it doesn’t compromise their results). 
 
Fourth, Buchmann and co. not only do the cost effectiveness of their own intervention (i.e. the conditional transfer), they have an extensive appendix where they benchmark it against a range of other interventions that affect marriage and show its good value for money.  
 
All in all, the fact that the transfer works makes a lot of sense.   As Buchmann and co. point out, the later women get married, the higher the dowry the family would have to pay.   It would be interesting to see how the empowerment versus conditional transfers works out in a country where dowries are less of an issue.    But where they are, it sure seems like putting conditional cash on the table is a good option.