So after having read Black Swan, my mind is filled with all these questions. I think I've answered the most recent, but I'd like your feedback.I was thinking about the most recent election in Calgary-Elbow, and the election two elections ago. Two elections ago, Conservatives got around 10000 votes, and Liberals got around 5000. This election, the Conservatives got around 4000 votes, and the Liberals got around 5000.You don't have to be a rocket scientist to look at those numbers and come up with a plausible explanation. "Conservatives didn't vote." Makes a lot of sense. But I was imagining what a statistician would tell me about that result, and I thought it might be close to the hyper-spin that some tween percentage of voters had switched allegiance from the Conservatives to the Liberals.It looks that way in the percentages, but the percentages are obviously not the whole story. Sure, a sample of 9000 is probably as good as a sample of 15000 for a population the size of the voting population of Calgary-Elbow, with some small change the margin of error. The problem, a statistician would point out, is that these are not randomly selected samples. Voters are self-selecting, which means that the voting preferences expressed in a vote are not generalizable across the people who didn't vote. That is, unless you can demonstrate that likelihood to vote is not correlated to voting preference, which seems unlikely.Which made me wonder - why are telephone polls so successful at predicting the result of federal elections nationally, with a sample size as low as 1000? You see, telephone polls have a terrible response rate, much lower than voter turnout, even in a by-election. So the respondents to telephone polls are self-selecting, too. Nor are the self-selections identical. Far too many people refuse to answer the polls for them all to be people who won't vote. That would suggest that the results of a telephone poll would not be generalizable across those individuals who don't answer telephone pollsters' questions. If that was any significant part of the voting population, which I suppose it is, election polls ought to be a lot less accurate than they are.Then it occurred to me. While the likelihood that you will vote is likely correlated to the party you support, your willingness to answer pollsters' questions is likely not. As long as the pollsters correct for whether or not you're going to vote, their results should be generalizable.That's why any number of polling firms can ask a mere 1000 people a few questions, and get answers that are generalizable across the country, while two subsequent "polls" of the people in a constituency will result in unrealistically shifting percentages of support.Which means what? Just further evidence that the election strategy of "identify your supporters, get them to the polls, try not to screw your national/provincial campaign over, and hope the national/provincial campaign doesn't screw you over" is about all you've got in this world. You basically can't control who people vote for at the local level. What you can control is whether or not they vote. The Liberal candidate in Calgary-Elbow won because they followed that strategy, and the Conservative candidate's was screwed over by things over which he had no control.Why we volunteer for this crap, when so few of us will ever have so much as a meager influence over what is done during a campaign or an administration, totally eludes me.Which is, I suppose, another way of saying I know why people don't get involved in politics in this country.