Assuming a completely random sampling method, a sample of about 1000 people is all that is necessary to determine statistics within a margin of error of +/- 3% 19 times out of 20 for a population the size of Toronto.
Look up "margin of error", "confidence interval", Google "confidence interval calculator" etc. Pollsters are not amateurs. They even try to account for sampling error (which varying levels of success...). But, the sample size is not the issue here.
That's not really an endorsement for Ford considering that if you look at the actual polls, his numbers are quite dismal - he is the incumbent and a sizeable majority would not consider voting for him. There's also the issue that there really aren't that many alternatives to Ford being pitched right now, personally I'm not really a huge fan of anyone in the mayor's race at the moment.
But the math works out.
Edit: euler, this is not specifically directed at you, but at all those who think these polls matter.
First, it doesn't take much to design a poll with a predetermined outcome.
Second, you are correct that pollsters are not amateurs. They are professionals whose profit-margins are razor thin, but also whose advertising relies heavily on political polls. This tension creates a double-bind situation wherein pollsters
must conduct these polls at no charge, but they also
must do so at little-to-no cost. Invariably, the end-result is compromised by the downward cost pressure.
Third, sample sizes, adequate statistically or otherwise, mean nothing. Polls flatten out complexity, nuance, and any measure of contingency. Then, in the media, these already decontextualized views become
even further removed from any subtlety or thoughtfulness as poll results are reduced to an 'approval rating' that fits tidily not only on a news ticker, but also within any given media narrative. Ultimately, the end-product of these polls is a meaningless number - one that hardly resembles the opinions of individuals in the original sample.
Fourth, and finally, these polls make no sense outside of the two months or so prior to election day. This isn't federal or provincial politics, wherein you know more or less in advance what your voting options might be. What if Mark Carney wakes up on Friday, ditches the Bank of England, and decides he wants to run for Mayor of Toronto?
In short,
these polls are fucking useless.