As the 2008 campaigns have come to a close, I thought it might be good to look at the former polls to see how accurate they were as well as how accurate my favorites polling site, Pollster.com, is as a whole.

Below is a picture of the Presidential and Senatorial races right before the elections (I decided not to attempt the Congressional races due to fewer polls and many more races). First off, lets look at Pollster.com to see whether its a good site to use in the future:

According to the averages Pollster conducted in the Presidential campaign, it was correct on every state except Indiana and Missouri, expecting Obama to win Missouri and McCain to win Indiana (despite the exact opposite being the case, although they were within the margin-of-error (MOE) on Pollster). This however doesn’t account for the precise numbers the candidates won by. As you can see: MT, ND, MO, IN, OH, NC, FL, and GA are all in yellow, signifying that they’re within the MOE and thus too close to call for either candidate. In actuality, not all of these states were so close. Most notable is North Dakota, which Pollster gave to McCain for less than 1 point despite him winning by a sizable 8 points. In Arizona, McCain’s homestate, some pundits in the last days thought Obama could score a big upset. Pollster put the race at a meager 5 points, but it actualized into a 9 point win. In Pennsylvania, the state where McCain made his last stand, Pollster put Obama ahead by 7 points, but on election day that lead grew to 11. Again in Nevada, Obama was only 7 points ahead but won by 12. Pollster put Obama 9 ahead in New Mexico, but he won by a whopping 15! Lastly in Iowa Obama was expected to win by 13 points, but won by only 9.

This however, is the end of Pollsters flaws. In spite of the within the MOE IN and MO, every state, especially those closest and most important like FL, NC, VA, GA, CO, were so close its almost scary! And their analysis of the dark red/blue states were also incredibly accurate, predicting to a tee just how much each candidate would win them by; for instance, McCain was predicted to win the trusted Republican state of Texas by 13, and won by 12. For states where much less polling is done, Pollster was continually right on the money.

As for the specific pollers, here are some of interesting examples: Times/Bloomberg expected a win for Obama in Florida by 7 points, however the win was by less than 3. CNN/Time expected a win for Obama in North Carolina by 6 points, but the win was by less than 1. Reuters/Zogby expected a win by McCain in Indiana by 6 and than in another poll by 5, but Obama won by about 1 point. Zogby (alone) expected the same win by 11 points. DailyKos.com/Research2000 predicted a win for McCain in North Dakota by only 1 point (an earlier poll had them tied), but McCain went on to win by 7 points. Strategic Vision put McCain ahead in Ohio by 2 points, but Obama instead won by 4. Rasmussen expected a win by Obama in Nevada by 4 points, which turned out to be 12 points. Lastly, a Reuters/CSpan/Zogby poll expected a win by Obama by 13 points infamous win by Clinton in New Hampshire.

Now although I have mentioned the failings of quite a few of the greatest pollers around, this is only to say that not all (or any) of them are perfect. Everyone gets it wrong now and then, the trick is to see a pattern of those who usually are right or wrong. To prove such a pattern would make an already long post even longer, but in my experience, and corroborated by the polls for the 2008 election, some of the ones to be wary of include YouGov/Polimetrix, ARG, CNN/Time, Quinnipiac, SurveyUSA, and Diageo/Hotline. More importantly, the polls one ought to pay attention to include Gallup, Rasmussen, FOX, Reuters, and Zogby. The other ones you ought to watch with discretion and do research on them and their history of polling before trusting them too much. In the end though, and in my humble opinion, Pollster correctly uses the averages of all polls to create what seems to be a very accurate conclusion overall.

As for the Senate elections, probably considering the fewer amount of polls, Pollster.com was not as accurate as before. Oregon was expected to go to Democrat Merkeley by about 6 points, but instead it became a nail-bitingly close race going to the Dems by only 3 points. And the race in Mississippi seemed to be a lot closer, just 5 points, but somehow that became 10 points on election day. Finally, Hagan won by 9 points in North Carolina, but Pollster predicted a slim 4 points. Lastly, although within the MOE, Pollster predicted a win by Franken over Coleman in Minnesota by 2 points, but it looks like Coleman is likely to win unless something unusual is found during the recount. So although not as impressive a record as previously during the Presidential election, the results are still close and they were only technically wrong on the Minnesota race (although that and the Alaska race are still to-be-determined).

In conclusion, I still believe that Pollster.com is one of the best and most accurate polling sites online. Obviously, due to the means in which they do polling averages, its helpful to have multiple polls. Still, in the end, I am happily impressed with their ability to call the 2008 elections and will continue to use their site in the future. Another interesting fact is that Intrade.com, a site where people bet on the elections and more, is widely known to be incredibly accurate and had the exact same results as Pollster on election day. For any readers who might have their own take or have studied polling data and/or pollsters, please feel free to comment on what you think that way we can all be ready come 2010!