Predicting the winner of the US Presidential Election.
September 2, 2008 4:24 PM Subscribe
How can I accurately predict the overall winner of the 2008 US Presidential Election. I would also like to know the best methods of accurately predicting the winner of the electoral votes of each individual state.
I am familiar with the Iowa Electronic Markets as a fairly good indicator of predicting the national election, but what are some other good indicators or predictors of which candidate will win each individual state. These may be past voting trends, current registered voters by party, and others. Please lead me to these resources.
Thanks.
I am familiar with the Iowa Electronic Markets as a fairly good indicator of predicting the national election, but what are some other good indicators or predictors of which candidate will win each individual state. These may be past voting trends, current registered voters by party, and others. Please lead me to these resources.
Thanks.
Best answer: fivethirtyeight is a website that compiles polls (resources on the left hand side has more links) that may be helpful in researching good indicators.
posted by ejaned8 at 4:37 PM on September 2, 2008 [1 favorite]
posted by ejaned8 at 4:37 PM on September 2, 2008 [1 favorite]
Christopher Wlezien (formerly of the U of Houston, Go Cougars!) has done some pretty strong work in election forecasting; you may want to look up some of his work surrounding the 2000 and 2004 elections.
posted by Doofus Magoo at 4:38 PM on September 2, 2008
posted by Doofus Magoo at 4:38 PM on September 2, 2008
fivethirtyeight.com uses a pretty involved methodology to predict election outcome from polling data. They did well in predicting the Democratic primaries this year.
posted by mr_roboto at 4:40 PM on September 2, 2008
posted by mr_roboto at 4:40 PM on September 2, 2008
A little off-the-wall, but I've read several times about a grade-school (first graders, I believe) "election" that has accurately predicted the outcome of the presidential race. The last I read of it, they had only been wrong once out of a bunch of elections. Google isn't helping, but hopefully this jogs someone else's memory.
The concept itself is a little on the ridiculous side, but then again they're voting for who their parents are going for.
posted by Xere at 4:42 PM on September 2, 2008
The concept itself is a little on the ridiculous side, but then again they're voting for who their parents are going for.
posted by Xere at 4:42 PM on September 2, 2008
Electoral-vote.com is pretty neat for reading and understanding this sort of thing.
posted by kdar at 4:48 PM on September 2, 2008
posted by kdar at 4:48 PM on September 2, 2008
Option one: look at some of the various prediction pages (538, pollster, electoral-vote, Iowa markets, etc) and make a guess,
Option two has several steps:
Step 1: Become deeply familiar with the literatures on American voting behavior, turnout, and campaigning. You can probably get away with only reading everything written after 1980; you'll get a good sense of the stuff before from that.
Also Step 1: At the same time, anytime you find a book or article using a statistical method you're not familiar with, or using a correction for a problem you're not familiar with, stop and read all of the citations they point to for that method or correction. Then look up current information on that method or correction, especially for citations in older books (ie Mo Fiorina's treatment of logit/probit in Retrospective Voting in American National Elections, model some appropriate data using that technique, and interpret the results.
Step 2: Become deeply familiar with the literature on forecasting presidential elections, and with the statistical techniques underlying their models.
Step 3: Congratulations! You are now an expert in this. Also, it's ten years later. Anyway, build a model that seems good to you.
Step 4: Test your model with retrodictions: try to predict known outcomes with only data that would have been available at the time.
Step 5: Iterate between steps 3 and 4 until the model works to your satisfaction.
posted by ROU_Xenophobe at 4:52 PM on September 2, 2008 [1 favorite]
Option two has several steps:
Step 1: Become deeply familiar with the literatures on American voting behavior, turnout, and campaigning. You can probably get away with only reading everything written after 1980; you'll get a good sense of the stuff before from that.
Also Step 1: At the same time, anytime you find a book or article using a statistical method you're not familiar with, or using a correction for a problem you're not familiar with, stop and read all of the citations they point to for that method or correction. Then look up current information on that method or correction, especially for citations in older books (ie Mo Fiorina's treatment of logit/probit in Retrospective Voting in American National Elections, model some appropriate data using that technique, and interpret the results.
Step 2: Become deeply familiar with the literature on forecasting presidential elections, and with the statistical techniques underlying their models.
Step 3: Congratulations! You are now an expert in this. Also, it's ten years later. Anyway, build a model that seems good to you.
Step 4: Test your model with retrodictions: try to predict known outcomes with only data that would have been available at the time.
Step 5: Iterate between steps 3 and 4 until the model works to your satisfaction.
posted by ROU_Xenophobe at 4:52 PM on September 2, 2008 [1 favorite]
Depending on how much credence you give admissions of vote-tampering software and the like, you might want to take into account which states have outlawed black-box electronic voting systems since 2004, and which are still using them.
posted by -harlequin- at 4:52 PM on September 2, 2008 [1 favorite]
posted by -harlequin- at 4:52 PM on September 2, 2008 [1 favorite]
Just look at where people put their money. Polls are flawed more often than not.
posted by Autarky at 4:58 PM on September 2, 2008 [1 favorite]
posted by Autarky at 4:58 PM on September 2, 2008 [1 favorite]
Fivethirtyeight (suggested & linked above) is the answer.
Predictions for all 50 states and the whole election (with a percentage likelihood for each) + polling data from lots of sources for all 50 states + daily blog posts with analysis of polling and prediction methodology. Their methodology seems very sophisticated -- taking a variety of trends and factors into account.
I really doubt you're going to find anything better than that.
posted by Jaltcoh at 5:00 PM on September 2, 2008
Predictions for all 50 states and the whole election (with a percentage likelihood for each) + polling data from lots of sources for all 50 states + daily blog posts with analysis of polling and prediction methodology. Their methodology seems very sophisticated -- taking a variety of trends and factors into account.
I really doubt you're going to find anything better than that.
posted by Jaltcoh at 5:00 PM on September 2, 2008
The reason people do polling is because when it's done correctly, it's reasonably accurate. The easiest way to get a feel for it is to read polls.
posted by Pants! at 5:13 PM on September 2, 2008
posted by Pants! at 5:13 PM on September 2, 2008
Polling is reasonably accurate, yes, but with the following caveats:
1- It only measures opinions at that exact moment. By the time they are released, news cycles may have changed someone's opinion.
2- It depends heavily on the questions being asked, their order and their tone.
3- There is a difference between what people say to the questioner and what they actually do in the voting booth.
4- Was the poll actually done with statistical accuracy, or is there any self-selection or bias in the sample?
5- And it depends on the biases and "conclusions" of the people compiling the data, and the subsequent "journalism" on the data. They can ask any number of questions worded in any variety of biased ways- "would you vote for a former POW" gives you 80%. "would you vote for a cancer victim" might give you 40%. Either way, the next night the poll goes on FOX as "80% will vote for McCain!!" and Dan Rather goes on and says "60% won't vote for McCain!!"
It's impossible to predict- you could poll everyone in the country on Nov. 3, and then it rains on Nov. 4. The election becomes completely different from the poll.
posted by gjc at 6:11 PM on September 2, 2008
1- It only measures opinions at that exact moment. By the time they are released, news cycles may have changed someone's opinion.
2- It depends heavily on the questions being asked, their order and their tone.
3- There is a difference between what people say to the questioner and what they actually do in the voting booth.
4- Was the poll actually done with statistical accuracy, or is there any self-selection or bias in the sample?
5- And it depends on the biases and "conclusions" of the people compiling the data, and the subsequent "journalism" on the data. They can ask any number of questions worded in any variety of biased ways- "would you vote for a former POW" gives you 80%. "would you vote for a cancer victim" might give you 40%. Either way, the next night the poll goes on FOX as "80% will vote for McCain!!" and Dan Rather goes on and says "60% won't vote for McCain!!"
It's impossible to predict- you could poll everyone in the country on Nov. 3, and then it rains on Nov. 4. The election becomes completely different from the poll.
posted by gjc at 6:11 PM on September 2, 2008
A couple of other sites that track this that you might want to look at:
Princeton Election Consortium
Election Projection(leans right)
posted by procrastination at 6:33 PM on September 2, 2008
Princeton Election Consortium
Election Projection(leans right)
posted by procrastination at 6:33 PM on September 2, 2008
Fivethirtyeight is run by Nate Silver, who supports Obama. I don't know of any bias in his simulations, but it would be nice if a McCain supporter were independently doing the same type of analysis.
posted by lukemeister at 6:49 PM on September 2, 2008
posted by lukemeister at 6:49 PM on September 2, 2008
RealClearPolitics compiles polls and has no definite bias AFAIK. P.S. Polls are flawed yada yada yada.
posted by ALongDecember at 9:05 PM on September 2, 2008
posted by ALongDecember at 9:05 PM on September 2, 2008
Electoral scoreboard developed by two astrophysicists
posted by lukemeister at 9:35 PM on September 2, 2008
posted by lukemeister at 9:35 PM on September 2, 2008
A little off-the-wall, but I've read several times about a grade-school (first graders, I believe) "election" that has accurately predicted the outcome of the presidential race. The last I read of it, they had only been wrong once out of a bunch of elections.
If you just make a random guess you've got a 50% chance of being right, in a close election (like the ones the US has).
So the chance of correctly predicting the presidential election five times in a row if you were guessing randomly is 1/(0.5^5) or one in 32.
If you phoned up 100 grade schools which ran pretend elections, and got the results for the last five elections, there's a 1-((1-(1/32))^100)=96% chance at least one school would report correctly predicting the last five elections. If you phoned around 200 schools that rises to 99.8%
My point being, the kids were probably right by chance as much as by real predictive ability.
posted by Mike1024 at 6:03 AM on September 3, 2008
If you just make a random guess you've got a 50% chance of being right, in a close election (like the ones the US has).
So the chance of correctly predicting the presidential election five times in a row if you were guessing randomly is 1/(0.5^5) or one in 32.
If you phoned up 100 grade schools which ran pretend elections, and got the results for the last five elections, there's a 1-((1-(1/32))^100)=96% chance at least one school would report correctly predicting the last five elections. If you phoned around 200 schools that rises to 99.8%
My point being, the kids were probably right by chance as much as by real predictive ability.
posted by Mike1024 at 6:03 AM on September 3, 2008
... a grade-school (first graders, I believe) "election" that has accurately predicted the outcome of the presidential race.
You may be thinking of Scholastic Magazine's poll (scroll down), which has successfully predicted all the presidential elections since 1940 except for Dewey/Truman (1948) and Nixon/Kennedy (1960).
posted by clerestory at 6:11 AM on September 3, 2008
You may be thinking of Scholastic Magazine's poll (scroll down), which has successfully predicted all the presidential elections since 1940 except for Dewey/Truman (1948) and Nixon/Kennedy (1960).
posted by clerestory at 6:11 AM on September 3, 2008
fivethirtyeight's goal is to predict the results on Election Day. The other sites show the Electoral Vote based on current polls, so they're not predictions.
posted by kirkaracha at 9:08 AM on September 3, 2008
posted by kirkaracha at 9:08 AM on September 3, 2008
In theory, prediction markets (like Intrade) are the most accurate source of information. If you find a more reliable source of information that says the market is undervaluing a particular candidate, you could make lots of money by buying shares for that candidate. The efficient markets hypothesis says you can't do this.
In the real world, there are some caveats to this, but I generally view prediction markets to be much more credible than any polling website (even one with analysis and statistical inference).
posted by lunchbox at 9:10 AM on September 3, 2008
In the real world, there are some caveats to this, but I generally view prediction markets to be much more credible than any polling website (even one with analysis and statistical inference).
posted by lunchbox at 9:10 AM on September 3, 2008
« Older Is it Hardi or Hardly Planked? | Trying to treat my cat right but it's coming out... Newer »
This thread is closed to new comments.
posted by rancidchickn at 4:29 PM on September 2, 2008