midterm elections - polls

so we all know the polls were off 2 years ago and i cant trust wapo/cnn anymore.

what polling sources are you using to gauge the possible outcomes for the midterms?

Doesn’t look good for Dems. Seems like the whole platform has fallen to sh*t for them. Just the observation of a white male middle class tax payer though, not a demographic you would want to pay any attention to.

^a minority population pretty over time!

Use 538. Really the best source there is

There seems to be a Conservative lie factor. People afraid to 'admit ’ they are Republicans

Except that to Igor’s point, 538 was also dead wrong 2 years ago.

yep lol. i was checking 538. those damn libs got it wrong.

Dead wrong? I expect this sort of analysis from others, but not the quant who posts crazy math theorems to judge the star rating of Yelp restaurants

538 had an above consensus probability for a Trump winning. And surely you don’t assume a probability of 30% means Trump never wins?

From what I gather, lots of the polls were pretty accurate. However, prediction markets and a lot of so called “data scientists” still gave Crooked Hillary 90% chance. They all had some weird justification, probably based on personal bias, as to why the polls, which directly addressed likely voters, were false.

you should never bet against a winner!

to be fair, was anybody dead right two years ago?

Trump was and still is!

No, and I never indicated that anybody was. Everyone was wrong. OP is perhaps looking for something that doesn’t exist; i.e., a poll with classical predictive power in the modern voting environment.

In a poll between two potential candidates, if one stands at 30%, are we having a serious discussion about what the poll is supposed to be saying about the true chance of that candidate winning? The value of the poll assessing likelihood goes out the window if that is our analysis. We could argue that a candidate standing at 1% also does not preclude the possibility of that candidate winning, however remote. But that is not what the poll is designed to do. The poll attempts to assess the probability, not the possibility. Those are two different concepts.

The model is supposed to be accurate over a range of predictions. Sounds like you want a binary model result. Trump won with like 70k votes. It’s not a stretch to think if we ran the election 100 times, he would lose more than he wins

Actually I don’t want anything at all. The models are faulty for collection bias and input sampling issues, not the back-end output calibration. This is a known issue with current political sampling. All I’m saying is that in a poll assessing the likelihood of one of two candidates winning, if the net interpretation is that it says, hey, either candidate can win, well, then I would say that poll is of questionable value.

I don’t think the polls were off by much in 2016. From memory they had Hillary ahead by around 3% at the end and she ‘won’ by just over 2%. The weird electoral college system of course means that Trump is president, but overall 538 did good job of assessing the probabilities.

Just because a 30% underdog wins a football match, doesn’t mean the other team shouldn’t have been rated a 70% fav. There’s no such thing as a 100% certainty in football or politics.

Gundlach called it in the primaries.

Why did all the “independent” polls, which supposedly collect their own data and did their own analysis, lean the same direction?

I don’t know how everyone could have missed it. It was so damn obvious that Trump would win. I backed Trump all the way from primaries; I don’t back losers. Winners aren’t losers. If you exclude the illegal votes, he would when won by an even greater margin.