Ruth Institute Blog

Douglas W. Allen, an economist at Simon Fraser University, in Vancouver, has just published a highly controversial study in the journal Review of Economics of the Household. It breaks with the conventional wisdom that there is no difference between parenting by a mother and a father and parenting by a same-sex couple.

MercatorNet interviewed Professor Allen about his findings.

* * * * * *

MercatorNet: What has your research found about educational outcomes for children of same-sex couples versus children of opposite sex couples?

Doug Allen: There have been about 60 studies over the past 15 years or so that have asked “do child outcomes differ when the child is raised in a same-sex household.” Almost all of this literature has the following characteristics: the samples are tiny and biased, the outcome measures are subjective and difficult to replicate, and the finding is always one of “no difference.”

Despite the limited scientific validity of these studies, they all end with sweeping policy recommendations. It really is not a scientific literature, but rather a political literature targeted at judges, lawyers, and politicians.

Then came a paper by Michael Rosenfeld, published in Demography 2010. This paper had a large random sample and looked at normal progression though schools in the US. It was, in my opinion, the first solid piece of statistical work done on the question, and he confirmed the “no difference” finding. Later, Joe Price, Catherine Pakaluk, and myself replicated his study and found two problems.

First, he didn’t find “no difference”. What he found was a lot of noise, and so he was unable to statistically distinguish children in same-sex households from children in any other type of household – including ones we know are not good for children.

Second, the lack of precision in his estimates came from a decision he made to throw out children from the sample who had not lived in the same location for five years. This turned out to be heavily correlated with same-sex households. Hence, he inadvertently threw away most of the same-sex households from the sample. Without that information, he did not have the statistical power to distinguish between family types.

So, the three of us restored the sample and used the statistical technique of controlling for household stability. What we found was that children of same-sex households were about 35 percent more likely to fail a grade.

While this was going on, I was using the Canada census to look at some other questions. I noticed several things about the census that differed from the US. one. First, unlike in the US, the Canada census actually identifies same-sex couples. This solves a big measurement problem with the US census, which could include room mates, family members, and opposite sex couples as same-sex ones.

Second, the Canada census had a nice link between the children and the parents, so I was able to control for the education of the parents and their marital status. Poor performance in school is correlated with marital disruptions of parents, so this is an important control. In many ways then, the Canada census is a much better data set for addressing this question, and I decided to simply redo the Rosenfeld study using this data. (The census does not record progress through school, so I examined high school graduation rates instead).

So, what did I find? First, I simply looked at how any child in a gay or lesbian home did compared to children from married, cohabiting, and single parent homes. Most of the discussion in the paper compares children in same-sex homes to those in opposite sex married homes, but a reader can do all of the comparisons by looking at the tables.

I found that on average, children in same-sex homes were about 65 percent as likely to graduate from high school, compared to similar children in married opposite sex homes. That finding seems very similar to the one we found in the US regarding normal progress. Next, I wondered if the gender composition mattered at all, so I separated out the boys and girls. I was very surprised by the results.

On the boy side, I just found a lot of noise. Some boys do well in same-sex households; some do quite poorly. I cannot statistically determine the effect.

Just looking at the point estimates, boys in lesbian homes are about 76 percent as likely to graduate, in gay homes they are about 60 percent more likely to graduate. But neither of these are statistically significant, meaning they cannot be distinguished from zero.

Girls are another story. First, the estimates are very precise. Second, they are low. A girl in a gay household is only 15 percent as likely to graduate, in a lesbian household about 45 percent as likely. The result found by lumping all of the children together is being driven by this girl effect. This result is very robust, I tried many specifications, sample restrictions, and estimation techniques, but it always remained.

So, my paper no only rejects the “no difference” consensus, it points to a finding — that if upheld by other studies — seems incredibly important.

It’s particularly hard on girls, isn’t it? Why is that?

Keep reading.