QC125 is a new website that looks at elections and similar stats in Quebec. I highly recommend reading their latest article, which is in French. I thought I would summarize it in English for those of you, who like me, are unilingual.
PJF, who runs the site, ran a little simulation. He generated 5,000,000 votes (a roughly quebec-sized electorate) and then randomly picked 1000 of these voters to be "polled"
The results of the polls are here
None of the parties actually moved up, or down, during this period. Red was assigned 32%, Blue 28%, Green 24%, and Orange 12%.
This is why we poll average. This is also why those of us who use polls to analyze politics, go out of our way to ignore single polls. A single poll showing Red and Blue tied means nothing, and this shows exactly why, just as a single poll showing a lead of 8 points for Red means nothing, nor a poll showing Green ahead of Blue.
Poll averaging is critically important as it gets rid of the "margin of error" each poll comes with naturally. With a margin of error of 3% (as in the above) a poll showing, for example, 43% and 43% is identical to one showing 49% and 37%. Poll averaging helps us weed out this margin of error and figure out what is really going on.
Next time someone tries to show you an individual poll and suggest one candidate or another is therefore winning, show them these two posts, and remind them that error in polling is part of the process and not a mistake.