When it says after a poll, "margin of error +/- 3%" it means that there's a 95% likelihood that the actual proportion of the total population that would answer the survey in the way indicated is within three percent of the number given. In fiver percent of cases the real number may be even more different.
Now, what most people don't realise is that this margin applies to each of the percents. In other words, if the poll has the Conservatives on 40% and Labour on 30%, then it would still be within the given ‘margin of error’ for the Conservatives to be on 43% or 37%, and for Labour to be on 27% or 33%.
That means when the lead is given as 10%, the real lead can be between four percent and 16% and still be within the ‘margin of error’. Now of course it’s more likely to be in the number stated and not the extreme of the margins, but no pollster quoting a three percent margin of error (which is what you get with a poll of 1000) can consider, say, a five percent or an eight percent as significantly different. When you see the polls bouncing around, it doesn’t matter there’s been a big shift in actual opinion.
The true value of daily polling is that when you look at a whole sequence of polls you can be confident that the trend-line is real. With a weekly or monthly poll, it takes much longer to know whether a particular three-point jump means anything at all. Whichever time series you're looking at, make sure you watch for the pattern, don’t over-interpret small movements.
