## Friday, 24 May 2013

### Simpson gets married (and divorced?)

While I was waiting for my coffee this morning, I flipped through the newspapers on one of the tables in my local coffee place when my eye got caught by this article in The Times (I think to see the full article you need a subscription $-$ which I don't have on principle grounds. But you can access the actual report even if you don't subscribe to Murdoch's flagship newspaper).

The news is that "Most children from broken homes had unwed parents" and the evidence comes from a recent report published by The Marriage Foundation

The report, based on data from Understanding Society (US), the ESRC longitudinal survey of British households, says that:
1) 45% of young teenagers (13-15 years old) are not living with both parents;
2) Half of all family breakdown takes place during the first two years;
3) Amongst parents who remain intact, 93% are married.

Now: I can see that there might be something going on and that it may be more likely that married couple remain intact longer (although I think one should bare in mind that, as reported here, "Based on marriage, divorce and mortality statistics for 2010, it is estimated that the percentage of marriages ending in divorce [in the UK] is 42%").

But surely this average 93% figure will possibly suffer strongly from Simpson's paradox, ie it is highly sensitive to conditioning on some other covariates (which are not considered here $-$ NB: that's not to say that US doesn't account for them; just that the report didn't bother considering them).

For example, kids who happen to become pregnant when they are 16 and are not in a stable relationship may well be at a high risk of breaking up within two years of the child being born. And probably, if they do get married as a result of the unwanted pregnancy, they may contribute highly to the proportion of divorces. But this doesn't necessarily apply to thirty-something who have been in a stable relationship for a few years before deciding to have kids, while for some reason not getting married! Because these various categories are likely to not be uniformly distributed in the underlying population, one should weight for their relative frequency when giving a population summary.

Another interesting passage from the report says: "... provides further evidence that the trend away from marriage is the driving force behind family breakdown. Out of 47% of children born to unmarried parents today, the model predicts that just 11% will reach their 16th birthday with both parents intact and unmarried. The rest will either marry or split up".

So, it seems to me, the evidence provided is rather quite weak: what if all of the rest actually got married? Should we then blame marriage for breaking up those families? (Of course, that's just an extreme case, which is unlikely to happen. But the report conveniently fails to give the distribution of "the rest" and, it seems to me, makes some unwarranted inference from the data).