What if Poll Bias were reversed? The case for weighting.

The case for Weighting.

As you know, I regularly rag on Public Policy Polling (whom I simultaneously love and hate) for their lack of weighting polls. For instance, the other day I re-weighted their polls to accurately reflect the current voting population in Nevada. PPP’s sample had a magnificent 45% of respondents as Democrats despite 2010 turnout being only 35% Democrats and Democrat-heavy 2008 only 38%. So, it’s little shock that Obama performed well in a poll that has more Democrat respondents than have ever turned out to vote.

And every PPP poll is like this. While I trust that their sample is random, and I very much like the fact that they release all cross-tab data, it’s still unfortunate that their public releases fail to mention this systematic bias if they aren’t going to account for it. Now, as a Democrat and Daily/Kos polling firm, I understand there is no need in their opinion to publish opinion that obviously favors the side they agree with. But, do they really think that the Democrat prescence in Nevada has increased by between 7-10% in the last two years, or over the last couple of months?

What would poll results look like if the uncorrected poll bias that is published by PPP were reversed and favored Republicans by the same margins instead of Democrats? Let’s look.

In PPP data, they released a poll that had a respondent make-up of 45% Democrat (!!!), 35% Republican, and 20% Independent. In 2010, turnout was 35% Democrat, 33% Republican, and 32% Independent. While the Republican total was fairly accurate, about 40% of the Independent polled were replaced by Democrats.

The results of such a bias are not unexpected and PPP published that Obama would beat Sarah Palin in Nevada by 50-39%.

If the equal bias were reversed and this time in the Republicans favor, the new make-up of those polled would be 43% Republican, 37% Democrat, and 20% Independent.

By flipping the  bias from Democrats to Republicans the results would be quite drastic. The old Result with Bias at Dem +10 gave Obama a 50-39% victory. But an equal “random sample” of bias in the Republicans favor would give Palin a 45-44% victory. If you extrapolate the “unknown data” and assume Palin and Obama win the unknowns at the same respective rates that they are, the new result is Palin 52.5% and Obama 47.5%

Now, let me ask you: Would any polling firm let that sample stand without acknowledging it? Doubtful, but PPP does it every day, and thus a 10-point Democrat-sample bias gives Obama an 11 point victory and a 10 point Republican-sample bias would give Palin a 5 point victory in Nevada. I will continue to re-analyze the polling data offered by PPP to eliminate the bias and give a more accurate reflection of the races at hand.