Sampling Problems in the Columbus Dispatch OH Poll

The Columbus Dispatch just came out with a new poll in OH on the presidential race showing Obama leading Romney by 50% to 48%. Being the number geek I am, I decided to check through the crosstabs to get a better understanding of what was happening. I came across something very interesting where they reported on how the vote has shifted since 2008. They show 5% of McCain’s voters shifting to Obama and 11% of Obama’s 2008 voters shifting to Romney. They also reported on third party votes and those who did not vote in 2008 but are planning on voting in 2012.

When checking their 2008 numbers I came across a discrepancy. The actual vote totals in 2008 in OH were 51.4% for Obama, 46.8% for McCain and 1.8% for other candidates. However, ignoring the 2008 nonvoters, the Dispatch poll has Obama at 52.9% in 2008, McCain at 44.8%, and other candidates at 2.2%. That is an undercount of McCain votes by 2.0%, an overestimate of Obama votes by 1.6%, and an overestimate of other votes by 0.4%.

If you readjust the numbers according to the real 2008 votes, it flips the final results of the poll. (I am including their polling of 2008 nonvoters in the final total.) Assuming party shifts are correct, correcting for the 2008 number will give Romney a lead of 49.5% to Obama’s 48.6%.

Please note that this is NOT unskewing the polls for party ID. I am readjusting the percents according to known real numbers. We know what the real vote was in 2008, so if current polling doesn’t match those numbers then it is a verifiable, correctable, systematic sampling error.

Finally, another thing smells fishy about the crosstabs. They indicate that people who did not vote in 2008 but plan to vote in 2012 are breaking for Obama by 57% to 36% with 7% going to other candidates. But we know GOP turnout was depressed in 2008. In other words, there were a lot of republicans who did not vote for president in 2008, but enthusiasm for Romney is greater this time around than for McCain. While there are a lot of first time young voters attracted to Obama to account for his portion of the 2008 nonvoters, a split of 57%/36% seems way to big. Unfortunately, there is no way to objectively double check those numbers. (I did not correct for this in my analysis above.)

I have not looked through other poll crosstabs to check their 2008 vote split. I don’t know if I will have time.

Bottom line is that, in this one poll, Ohio is a lot closer than some media are reporting. Correcting in just this polls flips the race by about 4%. And, of course, it will all come down to turnout. No matter what the polls say now, we need to get our people to vote.