How to Read the Polls

Polling is the lifeblood of modern data-driven election analysis. Yet if you find the endeavor perplexing, you’re hardly alone! Some poll analysis is truly complex. Nevertheless, there are some basic facts about interpreting polls that are fairly easy to grasp, and will greatly enhance your experience at RealClearPolling.

First, why do we poll? The answer comes in two parts, the motive and the method. The motivation is that we’re seeking a snapshot of what Americans are thinking about any given issue and any given campaign as a specific point in history. The second part of the answer is a practical one: We can’t talk to everyone. If we could, we would, and we wouldn’t need to sample. But that would take a very long time and cost massive amounts of money. So instead, we talk to some people, who can give us an idea of how everyone else is voting.

This leads to the most common reason for skepticism: How can you draw conclusions about 300 million Americans by talking to 500 people?

The answer is pretty intuitive. If we want to know if a coin is fair, i.e. equally weighted, we could get a pretty solid answer by tossing it 300 million times. About 150 million of those times it would land heads. But the truth is, we wouldn’t demand that. If it were extremely biased, by the 10th heads toss in a row, most people would suspect something was up. Could we have tossed a fair coin 10 times in a row and only gotten heads? Sure. It’ll happen about one time in 1,000, in fact. It’s just not the best explanation. Certainly, if we tossed it 500 times and got all heads, all but the most fervent skeptic would conclude the coin was rigged.

The same thing happens with polls. If President Biden’s “true” job approval were 50%, it’s possible that a sample of 500 people would contain just 30% Biden approvers. It just wouldn’t be very likely.

What do we mean by “not very likely”? That’s where error margins come in. Say that the error margin for a poll is 4%. What that means is that for every 100 polls taken, we expect that in 95 of them the “true” output will be within 4 points (in either direction) of the reported number. Is it possible for polls to be farther off? Absolutely. It just isn’t very likely.

So why do we average polls at RealClearPolitics? The reason is that by averaging polls we effectively increase the number of respondents and decrease the error margins. In any given election there may be a pollster who is more accurate than the RCP Average. But over time, you’ll do better following the RCP Average than by following any single pollster.

A few final notes on reading polls that a lot of people get wrong. The first is that the error margin attaches to both candidates. So if a poll is Biden 46%, Trump 42%, that means we’re fairly confident Biden is somewhere between 42% and 50%, while Trump is between 38% and 46%. Second, as we get close to 0% or 100%, the error margins shrink. This requires some involved math to explain, so just trust us when we tell you that a candidate receiving 4% of the vote has a smaller error margin than one with 50% of the vote.

If you want to determine whether a race is outside the margin of error, take the reported error margin and multiply by 1.6. So if the error margin is 4%, a candidate who led by more than 6 points would have a lead “outside the error margin.”

Finally, we should remember that polls are only snapshots of public opinion at a given point in time. Opinions can change, and frequently do! We can’t tell you what people will think a year from today, but we can improve your ability to understand what is going on now.