Readers of this spot are by now familiar with the polling maxim: Check the Party Mix, and no, we don't mean this. The Washington Post has another poll out, and while the news is better for the President, the results are still skewed by the party mix. Consistently, on the most important questions, the Post has a mix with about 5% more Democrats than Republicans.

I'm not going to go into each specific question here, but it's very important to document the analytical methodology I'm using. I use Microsoft Excel's Solver function. Now, Solver works by starting with a set of suggested numbers, and converging on a solution. If there are multiple solutions, Solver will pick the solution closest to the starting point. This is important, because there are an infinite number of solutions to the party mix for any questionnaire answer. I chose to start with a 33-33-34 D-R-I mix. In every case, Solver found that the total percentage for both answers (Kerry/Bush or Good/Bad) needed about 5% more Democrats. Solving for both answers was necessary due to rounding error - the Post only published results to the nearest percent.

Seized by self-doubt, I wanted to make sure that there wasn't some mix like 30-30-40 that would get the Post's results. There wasn't. In one case, a 20-20-60 mix worked, and in another, there was no solution with equal party proportions that gave the aggregate result. I ran beginning guesses from 50-50-0 to 0-0-100, and never did a reasonable mix with equal numbers of Democrats and Republicans work. Finally, it's worth noting that in every case, the Democrats needed to outnumber the Republicans to get the Post's results. If the errors were purely rounding error, or random, one would have expected the Republicans to outnumber the Democrats just as often.

Look, 5% is nothing like the L.A. Times's double-digit oversample. But it's consistent, it skews the results, and after having it pointed out multiple times, it's negligent. Partisans always take heart from favorable polls, and are disheartened by unfavorable ones. A 50-50 country needs to be represented by a 50-50 sample. Anything else, anything this consistent, isn't reporting, it's campaigning.