Lee’s Summit, Septermber 26, 2012 – The latest Quinnipiac University/New York Times poll shows that Romney is behind President Obama in Ohio by 10 points, and was ahead by similar margins in Florida and Massachusetts according to Reuters. Jan Crawford in a piece done for CBS this Morning today said “It is getting tough here for Mitt Romney. Our latest poll now shows the president is up by 10 points. The poll shows the president also has a significant lead with women voters in all three of those swing states that we surveyed. But here in Ohio the president is up by 25 points with women voters. Now, today Romney will be focused on the economy. That’s an issue that’s been the cornerstone of his campaign, but there’s trouble in our poll there for Romney, too! In our poll, for the first time, the president has taken a lead on that key issue in all three states.”
Most of us, after hearing this would say: Don’t even bother voting on November 6th – It’s over for the Republicans. Perhaps that’s the point behind some of these polls.
Do you understand the science of polling political questions like this? No? Well, I had no clue either so I did some digging and here’s what I’ve found.
The size of the sample matters in terms of statistical confidence (how well your sample can predict the results of the entire voting population). They always tell us the “sample error” and it’s usually two or three percent give or take a little.
However, there is something else that matters too. Yes, obviously whether or not you’re a likely voter matters – they define that by your voting history and hope it projects to your future actions. But more importantly there is one more factor.
MAKE UP OF THE SAMPLE:
The science behind picking who will be polled and who won’t is very precise. Finding the numbers, that’s harder. You have to have the right ration of men to women, married, single, divorced, parents and non-parents. All of these factors have to represent the population.
One more thing has to be taken into account, and this is the key to some of these polls and their results. You have to have the proper proportion of Republicans, Democrats, Independents and others. If you don’t then you have a built in error.
If I poll all Republicans, Romney wins by a landslide. If I poll only Democrats, Obama wins by a landslide. It also applies to the subset of women that are polled. If I poll mostly Democrat women then the President scores highly with women, but if I question mostly Republican women the President does poorly. So, how do they pick the proper percentage of each?
Pollsters go back one, two or three election cycles and try to discern from historical data the percentages they expect. If you look at the 2010 midterm only, then you’d have to conclude that many more Republicans than Democrats are going to show up at the voting booth. But, if you look at the 2008 turnout, you’ll have to give the advantage to the Democrats – because in 2008 they had their machinery out in force and gathered the most votes; as a percent of Politically Affiliated voters.
BACK TO THE POLLS
The Florida Poll has Obama up significantly in the Quinnipiac University/New York Times poll. The pollsters at Quinnipiac oversampled Democrats by 7% giving the poll a Democratic lean based on their perception that the 2008 numbers will repeat themselves: Ignoring 2010 as a statistical anomaly unworthy of consideration. They expect the turnout in Florida to be 43% Democrats, and 36% Republicans the rest being Independents or Others.
The Pennsylvania Poll oversamples Democrats based on the 2008 results as well, and has Democrats at 48% and Republicans at 37%.
In the Ohio Poll Quinnipiac did not ask, or is not showing, the party affiliation of those they sampled; but if I had to guess based on what they do in Florida and Pennsylvania they are oversampling Democrats as well to match the turnout in 2008.
If you look at the Professional Pollsters and rank them based on their accuracy predicting the 2008 presidential race – the study done by Costas Panagopoulos, Ph.D., of the Department of Political Science at Fordham University, here’s what you get:
- Harris Interactive
- GWU Lake/Tarrance
- Diageo / Hotline
To find the Quinnipiac you have to get towards the bottom of the list near 18th place (shown as 15th because of the ties – click here). At the bottom of the list are CBS, Gallup, Reuters / C-Span / Zogby, CBS / Times and bringing up the rear is Newsweek.
Forbes has a great article that helps me understand all of this, the title of it is Romney and Obama tied in Rasmussen Poll: Most Accurate Pollster in 2008 and 2004 (click here). It clearly covers in detail what I’ve said here and warns of the bias of some of the pollsters:
Dr. Panagopoulos also made an interesting observation from the 23 polling organizations in his report.
- Only 4 polling firms appear to have overestimated support of the Republican candidate (Rasmussen/Pew, GWU, Diageo/Hotline)
- While 17 overestimated the strength of the Democrat candidate (YouGov, Harris, CNN, Ipsos, DailyKos.com, Democracy Corps, FOX, Economist, IBD, NBC, ABC, Marist College, CBS, Gallup, Reuters, CBS/Times, Newsweek.)
Isn’t that interesting, that seems a bit skewed, doesn’t it?
So the question for me was; what is Rasmussen saying? Watch the video and find out. Here’s a hint – they are not that far apart.
The Lee’s Summit Conservative