I was astonished to read the following text in this article:
"How do you know if polling works? If a candidate wins?
COX I think it’s more complicated than that. People do sometimes use that metric and I think it’s a bad metric."
What is the use of polling if it cannot successfully predict who wins? Knowing who will actually vote is important, but predicting outcomes is, or should be, the goal of the polling exercise.
@Ro Ma If a poll says someone is going to win by 10, and they win by 1, was it accurate? If another poll says someone was going to win by 1 and they lost by 1, is it more or less accurate than the first poll?
Polls are always estimates of some underlying value in the population. The way to gauge the accuracy of the poll is how close the estimate comes to the actual value it's attempting to model. Not whether the poll and the actual value are both on the same side of some arbitrary line, regardless of how close they are.
Knowing how well the sample matches the greater population on *other* values - e.g. capturing an accurate portrait of the demographics of the actual voting population - is useful in evaluating the overall effectiveness of a poll, and diagnosing why the estimate might or might not have been accurate. It's also valuable for developing future polls.
I just want to say a formal thank you for this project. It informed me, it entertained me, and changed my emotional state with every blue and red dot which danced (or crept) across the screen. The depth and breadth of information for each poll was unprecedented and incredibly helpful to data nerds like me. I will never forget the night in September driving the forty miles to Racine Wi to phonebank for Randy Bryce in WI01 and watching his early polling lead evaporate. Or the final tense moments of polls where my preferred candidate tried to hold on to a 1 point lead (especially last week where two candidates lost their lead or tie on the 500th call!). Anyhow, I look forward to getting back to driving with my eyes on the road rather than on my cellphone- at least until Tuesday when the needle makes its long awaited return.
Sounds like a good idea!
The Times Sienna exhaustive and comprehensive polling is welcome and appreciated. The “live” aspect, not so much. Publishing poll results before full sample size is misleading at best. So if I ask 100 people and say Beto is in the lead, but as I get to a reasonable sample size of 500, Cruz is favored, what value is that? The only value could be for entertainment. Not a good trade off.
3