"Jim C. Nasby" <jnasby(at)pervasive(dot)com> writes:
> That's because unfortunately PostgreSQL only keeps statistics on
> individual columns. There's no stats kept on multi-column indexes; the
> best the planner can do is use the stats for the first column.
That's not what we do at all: we do look at the stats for each column
and combine them.
The deficiency is that we don't have any stats about cross-column
correlations, and therefore must assume that the columns are independent
(in the statistical sense). This is a poor assumption in the real world.
But it's a lot subtler than "we only consider the first column".
regards, tom lane