Of course.
You use past data to make a model. The model, by definition of a model, fits the past data.
THAT is all the CO people are saying when they say their model predicts every election since 1980. It’s their way of saying “our model fits all the data since 1980.” That’s all. Nothing more to it.
Yes, and if Bible Code II had been published in 1997 (the publication date of the original Bible Code) instead of 2002, it would have predicted the 9-11 attacks. A bit of contemplation of why the first book failed to reveal this rather significant event despite its “analysis” of the very same Bible that was further “decoded” in the sequel may shed light on the issue.
And most election prediction models don’t do that. They pick certain variables.
Good idea.
I didn’t say this was a good model. I hope it isn’t!
That is an incorrect use of the word “predict.” It is not in line with what “predict” means in statistics.
If they’re saying that without also saying “and that is why our model ain’t worth doodly until and unless it correctly predicts some new elections”, they are being either incompetent or dishonest.
The day after the lottery, I’m going to predict the winning numbers. Decent model? Hell, the chances of predicting the winning number is one in hundreds of millions. It’s an excellent model!
EXACTLY!
And everyone understands that. You can’t predict the past.
They should have said “it would have predicted” or something like that.
I can see that they used the word in a non-statistical, non-precise way. They used it for the layman. Don’t get too wrapped up in the precise meaning.
Just for reference, here is Silver on models from earlier in this election cycle (prior to this latest model). http://fivethirtyeight.blogs.nytimes.com/2012/08/07/models-models-everywhere/
If you can find a cause-and-effect relationship between an independent variable and lottery numbers that is strong and lasts over a long period, let me know what it is. Elections are not lotteries. They are not random.
I predict that this winter will be colder than summer was. Do you want to bet against that prediction?
Or that *also *goes without saying.
Lance, to be frank, you’re backpedaling. You started out saying that the past “predictions” indicate it’s a strong model. You’re now finding ways to back off from that assertion.
It’s okay, man, if you made a mistake at first. 
We all do it!
Sure, it’s trivial to create a model that accurately predicts every single lottery that has ever been conducted with 100% accuracy. You’d have to be an idiot to claim that means anything for a future lottery though.
Okay, let’s be fair here though. The CO model is unproven; it’s only a hypothesis. But that doesn’t make it junk science, it just makes it a hypothesis.
If in fact the CO model correctly predicts the 2012 election, then it moves from “Hypothesis” to “a hypothesis that’s a little bit tested and came out okay.” Then you adjust it a little and see what happens in 2016, 2020, and so on. Modelling based on past data is perfectly good science; it’s just not yet proven that their model works.
I mean, I always default to Nate Silver’s methods. But prior to the 2008 election, his model wasn’t any more proven than the Colorado model. The 2008 election results then made Silver’s model hold up; the evidence from that election suggests Silver’s method is sensationally accurate. But hey, maybe what worked then will not work as well now. If it remains accurate in 2012 his model will be unassailably the best in the business. But it had to start at just a model based on past elections, just like the CO model.
I’m very curious to see how Silver’s stuff works in a remarkably sparser polling environment. Some days have nothing but tracking polls - we are WAY behind the pace of 4 years ago. I don’t know why, exactly, but it can’t be helping the accuracy of the model. Maybe the pace will pick up after the conventions.
No, I’ve simply finally figured out what the confusion was all about. Glad you understand now.
With an independent variable that predicts the lottery numbers? Go ahead and create such a model.
“For the last eight presidential elections, this model has correctly predicted the winner,” said Berry
They are literally saying that they predicted the last 8 presidential elections, which is false.
From their press release:
[QUOTE=Deliberately misleading press release]
Their model correctly predicted all elections since 1980, including two years when independent candidates ran strongly, 1980 and 1992. It also correctly predicted the outcome in 2000, when Al Gore received the most popular vote but George W. Bush won the election.
[/QUOTE]
You still don’t understand that it would be trivially easy to do this. That’s why fitting variables to past data doesn’t imply you have any predictive power and why the folks at Colorado are being extremely dishonest when they claim they predicted the past 8 elections.
No, they are saying that their model would have predicted the last 8 elections, had it existed then, which of course it couldn’t have.
You are simply reading way too much into their imprecise language.
Do you really think they beleive that someone took their model back into a time machine and tested it? Of course not. It’s simply their way of saying their model fits all 8 elections.
Press releases are written for the press, to write stories for the general public. They are not scientific papers.
Get over it.
Sigh.
Sure it would be. It’s not what they did though. I trust that they are not idiots who created an idiot model. It may not be perfect, or work in 2012, but they are not idiots. You cannot judge their model from a press release.
Who the party blames depends on part on how they do in the Senate races. If they can get 51, they’ll probably take a good, hard look at which Senators won and where, to see what they need to do to win in 2016.
Then again, the current Republican party thinking might be, “It doesn’t matter who we run in 2016; we’ll win, as the Democrats will nominate Hillary Clinton thanks to their Superdelegate process, and they’ll learn the hard way that the country just isn’t ready to elect a female President.”
Also, if the thinking is that they’re in too deep of a hole to throw money at it (and I for one believe that lots of last-second money, and not Diebold (or whatever they’re called now) voting machines, are what turned Ohio to Bush at the last minute), they might try channeling money from Romney ads to Senate ads in close states, which might be the difference in one or two states going to Obama. Remember, three Supreme Court justices will be at least 80 years old by 2016.
Now you’re not understanding the concept, and you’re being misled by the wording of the press release.
The model could not have predicted the last eight elections because the model is built on the results of those elections.
The model is built on previous elections; it does not, and could not have, predicted them. The only election the model is predicting is the 2012 election. We will see in November if the model works. If it does it will be a remarkable achievement, inasmuch as some of their predictions are hard to believe - they predict Minnesota will go for Romney, where Obama has a statistically significant lead, has never trailed at any point, and which has not gone Republican since 1972. Shit, Minnesota even voted for Walter Mondale, and Mondale lost so badly he was mathematically eliminated from being an American. That is one hell of a prediction and if they nail it I’ll be pretty impressed.
What Professor Berry says is, in fact, flatly wrong, if in fact he is being quoted accurately. Having said that, I do not trust a press release to get this stuff right.