I'd say that the taxonomy issues are more complicated than the granularity
ones. In various parts of the United States, there are:
* church marriages
* civil marriages
* common-law marriages
* deemed or putative marriages
* civil unions
* domestic partnerships.
California and Massachussets allow same-sex civil marriages. California also
allows domestic partnerships to both same-sex and opposite-sex couples, I
believe, and so does Oregon, but Oregon does not allow same-sex marriage.
At least one church (the United Church of Christ) allows same-sex church
marriages.
So, as I posted on Dennis' blog yesterday, unless the census takers actually
capture separate data points for all the different legal unions provided for
under state and federal law, producing aggregate statistics is not going to
be feasible.
The granularity issue is that you can only re-code partnerships one at a
time. You can delete data in bulk (which is what the Washington Post was
talking about, converting same-sex couples to unmarried singletons) but you
can't improve the data unless you know the identity of the couple.
Gordon