Stanford stumbles on election prediction.
Stanford University has made a classical error by predicting a Bush victory with an “84% probability.” (*)
There is only going to be one election. There is no probability or odds that can apply.
What would have been proper for Stanford University to have said is this: “Out of 100 imaginary elections, Bush would beat Kerry about 84 times. Neverthless, in the election that will be held tomorrow, we just don’t know who will win.”
That would have made a less exciting headline, however.
It’s too bad that the nation’s elite—at Stanford—fail to understand basic statistical analysis.
Stanford’s “methodology” section (†) is laughable. It ignores the most basic point. One cannot apply probability to single future events.
Stanford’s nonsensical “prediction” also shows the general poverty of social science as it exists in the year 2004.
November 22nd, 2004 at 14:43
Actually, one can apply probability to a single event. An 84% Bush victory means it is more likely that Bush will win. Perhaps you missed this sentence: “Surely, this “final answer” is but a probabilistic one, and should not be interpreted as us calling the winner.” Stanford did not claim to predict a Bush victory in any definite sense. However, if you offered to pay them $20 if Bush won provided that they’d pay you $20 if Kerry won, they would accept that bet. To illustrate the role of probability in a single event, suppose I offer you a bet. I’ll roll one 100 sided die, and if it is a one, I’ll pay you 20 dollars. Otherwise, you pay me $20. You claim that probability has no part to play in your decision. Clearly, however, it should. You would be silly to take the bet, even though it is a single event. Rest assured the people at Stanford have their heads on straight. I do not see the connection between Stanford’s analysis and the poverty of social science. Perhaps you also overlooked the fact that the analysis was done in the engineering department. In the future, you may find it safer to limit derogatory comments to subjects you have some basic knowledge about.
November 23rd, 2004 at 00:06
The phrase “more likely that Bush will win in the single event” is nonsensical.
This is beyond a question of basic knowledge of statistics. My challenge drives at the underlying epistemology that supports Stanford’s reasoning.
I would add that this being Stanford Engineering’s attempt at election prediction (social science) only proves my point more.
My comment is clear, and thus I feel no need to explain at greater length my position.
November 23rd, 2004 at 20:58
Did you miss the part about bets? The statement makes sense because you are better off betting on Bush. This reflects the fact that he is more likely to win.
To further illustrate the concept of probability relating to a single event, have you ever watched poker? You claim that if we play just one hand, we cannot say anything about the outcome. Yet, I would far rather bet that someone will get a pair than someone will get a royal flush. We are not predicting the hand. We do not know what it will be. But we can say how likely it is that a particular outcome will occur.
If you truly are not swayed, I would like to make some bets with you. I offer you a single event: I pick a number between -1,000,000,000,000 and positive 1,000,000,000,000. If you correctly guess the number I choose, I will pay you $1,000,000 dollars. On the other hand, if you lose, you will pay me $1,000,000. If you decline this bet, think about why.
In fact, if you believe you cannot say anything about the outcome of the bet, you have already assigned a 50/50 probability to it, reflecting a state of no information.
I hope you can see that the underlying epistemology is sound. You don’t seem particularly interested in understanding my point of view, but if you do I will have far more confidence in the rest of your comments. You were talking about probability requiring averages. Probability in a single event can be thought of, as you said, as stating that if you could run the election an infinite number of times 84% of the them would show a Bush victory. That is correct. And that is why the probability of Bush winning does convey some information, even though it is a single event.
You said
Stanford University has made a classical error by predicting a Bush victory with an “84% probability.”
What would have been proper for Stanford University to have said is this: “Out of 100 imaginary elections, Bush would beat Kerry about 84 times. Neverthless, in the election that will be held tomorrow, we just don’t know who will win.”
Stanford did just that. “84% probability” means, literally, if you ran the election an infinite number of times, 84% would go to Bush. Stanford clearly said they did not know who would win. They did, however, feel Bush had a better shot.
June 12th, 2005 at 10:04
“84% probability” means, literally, if you ran the election an infinite number of times, 84% would go to Bush.
Thank you for pointing out the crux of the problem.
It is nonsensical to refer to running the election an infinite number of times, or even more than once. Hence, it is nonsensical to say that out of that series, 84% (or some other percentage) of the outcomes would be for Bush. Hence, Stanford made a nonsensical statement.