Statistics don't lie, but Staticians do.
Three things to look for when checking polls acuracy:
- Number in sample size (Bigger is better)
- Accuracy usually expressed in + or - a x%
- Accuracy is nothing without the confidence level (almost never shown on TV)
So when you are looking at a poll with a sample size of 1,000 +or- 3% the confidence level is probably low, like around 65% which would mean its 45% likely to be wrong. And that's if the pollsters are honest in filtering the sample.