|
>I cited this >Emerson poll because it's reputable (FiveThirtyEight >continually rates their polling with high-marks; I believe the >latest score was a "B+").
Whether a firm has a good historical accuracy score or not has nothing to do with whether you should take an individual poll result seriously. That "B+" rating is an estimate of a few different things, but the main thing it's estimating is "systematic error." Systematic error is the error associated with flawed measurement techniques and data analysis. Emerson apparently does reasonably well on these counts, hence the B+ rating from FiveThirtyEight.
But that's not the only kind of error, and it's definitely not the kind that's most relevant in understanding the meaning of a single poll. The other kind is "random error," or "sampling error." This is the error associated with the fact that Emerson, in this case, polled 430 people -- four hundred and thirty people -- to estimate the mood of a nation with nearly a million times that many voters.
There's nothing wrong with them doing that. It's how polling works. This is the incredible power of statistics. If Emerson happens to come up unlucky and calls the wrong four hundred and thirty people, then that error gets a little better when it's averaged against the next poll, and the next, and the next. The more polls are combined, the larger the overall sample is, and the more accurate the average is (at least as far as random error is concerned). There's a theorem in statistics called the central limit theorem that underlies this fact.
The existence of sampling error is such an important point, you'd THINK the polling firms would help us out with this kind of context. If you follow the link, you find, holy shit, they give us a nice easy-to-digest estimate of their sampling error. They call it the "margin of error." In this case, it's 4.7%.
What this number means is that there's roughly a 95% chance that Bernie's number according to their measurement is somewhere between 20 and 30. It could even be outside of that range, on either side (and it often is; under ideal assumptions even this padded number will be wrong in about one out of 20 polls). Biden's number, they estimate, is probably somewhere between 22 and 32. Warren's number is probably somewhere between 16 and 26.
So looking at this poll in isolation, what it tells us is that all three of them are in overlapping confidence intervals. This is not news. It's where the race has been, according to national tracking polls, ever since, ever since. (The fact that national tracking polls are a dumb thing to look at, is another story. And it's literally why aggregator sites like FiveThirtyEight exist.)
In Emerson's own press release, they said that their September tracking poll had Biden at 25, Warren at 23, and Sanders at 22. Note: all three of these numbers are less than 4.7 points away from what they are in this new BOMBSHELL poll.
In other words, this poll does not show any statistically significant change since last month.
|