A large grain of salt

Polls are simply not very good outside of Republican on Democrat, general elections. We've seen this demonstrated again and again this primary season. At least now that we're down to essentially two-person races (an admittedly less secure conclusion for the Republican race), they have at least one source of error removed, and should do somehwat better. But they still have the problem of projecting likely voters. This is much easier for general elections, because essentially the same people end up voting in those elections cycle after cycle. Primaries are much, much more variable in their turnout, largely because that turnout is usually considerably lower than for general elections. And it's not just the quantitative turnout, the % who vote, it's the qualitative turnout, the demographic breakdown of who ends up voting, that varies a lot from primary to primary, despite being fairly predictable for general elections.

Bottom line: the polls for these primary elections simply can't be expected to be accurate even within 10-15% swings. Polling is a science, not magic. If the pollsters don't have a good handle on who's actually going to vote, they can't get an accurate prediction by sampling a defined pool of likely voters. They have a good handle on who's going to vote on 4 November, 2008, but don't have much to go on for any race until then. So their final polls for the general Election Day will probably be very close to the mark; but their predictions for Mega Tuesday are nearly worthless.

Posted at February 3, 2008 12:18 PM in response to Chart: Totality Of Polls In Feb. 5th States Shows Race Is Tightening