Latest Presidential Polls: Mitt Romney is Leading in Ohio, Oversampling Explained
When three statewide polls get released in one day showing your candidate ahead by 4 points, 3 points, and a tie ... you must begin to feel pretty good about your party's chances. In this case, the state is Ohio and the candidate is Obama. When we dig into the raw data, it tells a different story ... a story of why Romney is really leading in the state I promise this article won't get too wonkish, and you will be a better-informed voter after reading it.
Let's begin with Public Policy Polling (PPP) which showed on October 28 that Obama was leading 51% to 47% over Romney.
A closer look at the data shows that Democrats were oversampled by a whopping 8% (D+8). Women were also oversampled by the same 8% (W+8). Crosstabs within the poll show 88% of Democrats choosing Obama and 55% of women. You can see how this skews the results when you oversample by such a large amount.
Now we'll take a look at the Gravis marketing results released October 28 that has Obama leading Romney by 50% to 49%. Not bad... a lead is a lead, right? Not so fast.
Gravis Marketing also oversampled Democrats by 8% (D+8). They also oversampled women by 6% (W+6). Crosstabs within the poll indicate 83% of Democrats and 53% of women selecting Obama. When you take this raw data into account that they oversampled, but not to as great of an extent, this Obama +1 compared to an Obama +4 seems very reasonable.
The third Ohio poll released on October 28 is from the Cincinnati Enquirer. They showed things all tied at Obama 49% and Romney 49%
This poll oversampled Democrats by just 3% (D+3). When it comes to gender-bias, to their credit they stated: "The data were weighted to correct for potential sampling bias on gender and region of residence for respondents." Now that is how to properly correct for oversampling.
To summarize all three polls:
PPP Obama +4: D+8; W+8 Gravis Obama +1: D+8; W+6 Cincinnati Obama-Romney Tied: D+3; W+2*
*Women are 51% of the Ohio registered voters, thus a W+2 assumed for a gender-corrected result.
If we go back to the 2010 elections, we see that the registered voters in Ohio broke down slightly in the favor of Republicans or an R+1.
Even if the demographics have changed from an R+1 and were corrected to anywhere up to a D+2, it is apparent that all three polls would show a Romney lead. I promised you that you wouldn't need to use your calculus to know what to look for in the raw data of a poll to determine the relevance.
This isn't meant to prove to a conspiracy theorist that the pollsters are in the tank for Obama. And it isn't outside the realm of chance to have more female Democrats answer a phone during the time the polls were executed. What makes a poll relevant is whether or not the poll understands and takes into account that these statistical anomalies take place — and what it does to correct them. You can see that we come up with an Obama-Romney dead-heat just by a simple gender-correction in one poll — even though Democrats as a whole were probably slightly oversampled.