One of my (many) pet peeves about political reporting in election years is how the media reports polling data.

A few years ago reporters were trained to report not just the point estimates from a poll (that is, say that Clinton leads Sanders 49% to 47%) but the so-called "margin of error (MOE)" (+-3%) and that if the margin between the candidates is less than the MOE, it is "too close to call".  That's good.

However, the press often makes the mistake of saying this also means it's a "dead heat" or "statistical dead heat".  Well, no.  That is not what these results mean.

For example, if one candidate (call that person A) leads by 2-percentage-points with a 3-percentage-point MOE that means we cannot reject the hypothesis that the candidates are in fact tied, or in fact candidate B is ahead.  

However it is also means, with 95% confidence, we can say the LIKELY OUTCOME is between +5% for Candidate A and -1% for Candidate A (or a 2% victory for Candidate B). In other words, a 6-percentage-point margin around the estimate of 2% difference.

Take the California Democratic primary right now. Several polls show Clinton ahead by narrow margins of about +2%.  Most of these have MOEs of about +-3%.  So again this means the likely confidence interval is +5% Clinton to +1% Sanders, but most of the margin is in favor of Sanders.

In fact if you look at the RealClearPolitics results over time, in every single poll, Clinton has led in the polls, sometimes by huge margins.