|
Gallup has been polling presidential elections since 1936. Today, they predicted Obama to win the election 55 to 45. The poll itself said 53 to 43 though. They were just equally allocating undecideds to come up with the 55-45.
How accurate is Gallup's final election poll?
Well, in the history of Gallups polling, their final poll before an election has usually been very close to the true result.
1. The percent of the vote gotten by eventual winner of the election (I am going by popular vote since thats what the poll measures, so for the purposes of this analysis, Gore "won" in 2000) was only off by an average of 2.61%. Obama got 53%, so he can afford the average error to go against him.
2. The vast majority of the polls are very close when it comes to predicting the winner's percent of the popular vote. There are only 4 times where Gallup's final numbers have estimated the winner's percent of the vote more than 3% off the actual results. The worst one came in 1992, when they overestimated Clinton's vote share by 6%. However, this was a complicated polling year, with Perot being a major factor. The next worse one was estimating Truman to get 5% less than he did. They also estimated Roosevelt to get 5% less in 1936 and Eisenhower to get 4% less in 1952. However, polling is lot more scientific than it was half a century ago. Since then, their errors have been very small, besides the complicated 1992 election. Furthermore, with large amounts of early voting making it easier to identify likely voters and who they will vote for (since someone who has voted is obviously a likely voter, and their choice is final), the chances that a final election poll will be off are much lower than half a century ago. Therefore, if it assumed that this year wont be a wildly off polling year because of early voting and developed polling techniques, then the result will probably be within 3% like 14 of the 18 final Gallup polls. Obama has 3% to spare and will still have 50%.
3. The margins of victory they estimate naturally hold more average error than percent of the vote garnered by the winner because it involves a margin of error for both candidates. However, the average error between the actual margin of victory and Gallup's predicted one is 4.72%. The worst example was in 1936 when they underestimated Roosevelts victory by 12%. However, they were actually one of the more accurate polling firms in that election. In 1948, they estimated a 5% Truman loss, and it was a 5% Truman victory, so they were 10% off. However, that last poll was a couple weeks before the election so it cannot be expected to have been that accurate, and Truman is widely believed to have had a surge in support over the last couple weeks. That leaves the other 16 results all being below 10% off. Obama's predicted margin is 10%.
4. No presidential candidate who has garnered over 50% in the last Gallup poll has lost the election. Obama is well above 50%, with his 53%.
Lets be honest. Polling just is never so ridiculously far off. To think that Gallup will have Obama ahead by 10% two days before the election, and then he'll lose is preposterous. No national poll has given McCain the lead or even been a tie since a Battleground poll done from September 21st to September 25th. It's absurd to think that Obama will lose despite all that polling evidence.
|