from that particular article: the results of the student vote have mirrored the outcome of the general election all but twice: In 1948, kids voted for Thomas E. Dewey over Harry S. Truman. In 1960, more students voted for Richard M. Nixon than for John F. Kennedy.
This is pretty interesting to me. Somehow, the Scholastic Poll managed to mirror the failed phone poll that also predicted Dewey would defeat Truman. Wonder how that came about. Just a coincidence, or something the same in the two polls? But what?
Then there’s the 1960 poll, which predicted Nixon, over JFK, when many pundits said was won by JFK’s good looks on the televised interview – so what’s that about, kids weren’t affected by JFK youthful good looks, and they liked scary unshaven Nixon? That seems counter-intuitive to me.
Anyway, it might just simply be coincidence, but lots of people in the Fark.com discussion page are giving lots of credence to this poll of school kids. The good track record seems to mirror national trends, the guess is, the kids are hearing what their parents are really saying in their own homes, whereas people are a bit biased, or self-editing at least, when talking to pollsters. But if anyone has a guess as to the two cases above, I’d like to hear them.
Like I said, I don’t really care why it’s right (cute statistical breakdown, 'tho.) I wanna know why it’s, in the first case, as wrong as a famous case, and in the second case, wrong for the wrong reason, or not right for the right reason, or something.
I think the point is it’s not a particularly accurate predictor of elections at all, and further, there’s absolutely no reason to believe there’s any meaningful reason it failed in either of the two cases. Would it make sense to pick any of the elections it proved correct and speculate why it was right? It’s easy to pick through past data for examples that appears to show a significant ability to predict future elections, but that’s not a reason to believe per se that there’s any actual predictive power or that there’s an interesting substantive reason behind its successes and failures.
If it were a 50-50 shot of calling the election, the probability that it would get 15 out of 17 elections right is about 0.001. That suggests that it’s not a tossup, and that the poll may have some predictive power. I’ll leave explaining it to someone more qualified.
I’d think there are two primary causes for the predictive power (if any) of this poll. One, kids watch TV and TV has political ads. The candidate with the more ads is more likely to win the election, and also more likely to get the kids vote in this poll. Second, kids here what their parents say. If they say good things about one candidate and bad things about another the kid is more likely to vote for the “good” candidate. Certainly not as reliable as just polling the parents directly, but probably a decent proxy.
Under the null hypothesis, the outcome we’ve seen is so extremely unlikely that any reasonable person would accept that there’s something else going on.
Have enough people toss coins and one of them will get it right by chance. It doesn’t mean there’s anything special about them or their coin.
But a poll conducted in October is hardly a 50-50 shot like a coin toss. By this stage of the game, dirty one of the candidates may well have established a clear and obvious lead. Dirty laundry has been aired and bumps from conventions, debates, and vice-presidential announcements are (mostly) behind.
My guess is that the kids were not allowed to stay up and watch the debate, so they did not get influenced by the look of Nixon on the TV the same way their parents were.
I’d suspect it might in some ways be better than polling the parents - Daddy might not say he won’t vote for a n***** when Gallup calls, but he’ll say it at the dinner table.
>So what are the odds of the Washington Redskins predicting the outcome with 17 outta 18 correct? I’m for coincidence myself.
If you want to evaluate the predictive ability of one specific method, you might reasonably ask whether the odds of its having randomly produced the record it did in fact produce are below, say, 5%. You’d say a method passing that test was correct, and that your confidence level was 95%.
However, if you go hunting around through statistics you can find on dozens and dozens of methods, and you find several that pass the 95% test above, you can’t claim a confidence level of 95% in concluding the ones you found were correct. That is, if they were all random to begin with, you should expect about 1/20 of them to pass that test, and the fact that something like 1/20 of them did pass would not mean there were any significant methods.
IANA statistician and would appreciate one who could correct, clarify, confirm, provide names for, or otherwise improve this statement.
Assuming for the sake of argument that the Scholastic poll really is 50-50, the question then becomes how many purported election predictors there are. If there are over 1000 bogus predictors, then you’d expect that one of those bogus predictors would give you record comparable to 15 out of 17.
So what are the odds of the Washington Redskins predicting the outcome with 17 outta 18 correct? I’m for coincidence myself.