If I could have, I would have compared stably-coupled LMs and GFs (gay fathers) with everyone else. As I noted earlier, the longer the household stay of the two-LMs, the better the kids’ outcomes appear at face value, but a meaningful statistical comparison of those few is just not possible. Some feel it was unethical to have moved forward once I realized this. That’s extreme, a standard to which research prior to this study has not been held. I thought the information we could learn deserved a hearing.
The problem appears to be that even out of a starting sample of 15,000 or so, he could only find 163 children who were aware of their mothers having any kind of lesbian relationship. All those people were included as LM (lesbian mother) data points, regardless of whether they met the conditions for any other classification. About which there are two things to say:
1. My training in statistics and research design, such as it was for a theorist who nevertheless took the empirical side of political science seriously, is that statistical comparisons are only valid when they compare likes that are as closely related as possible. But the category of LMs already has a great deal of internal variation.
2. If there's not enough data, there's not enough data. Regnerus points out, usefully, that breaking up the category of LMs any further would lead to sample sizes too small to draw any valid conclusions. The point is important, and features in the best criticism of the study: it has not been widely acceptable for two women to raise a child together for long enough for us to draw any conclusions at all about whether they are better or worse than other parenting combinations. The data isn't there, and attempting to draw conclusions from what little data we have is senseless at best, dangerous at worst, and not unlike determining one's favorite ballplayer is going to have an MVP year because he's hitting .400 through the first 20 games of the season.