Red Mosquito

Simple Torture wrote:

McParadigm wrote:

--- wrote:<<says something>>

We really are living in an upside-down universe now, huh?

I suspect we actually agree on much, much more than we disagree on.

4/5 wrote:

tragabigzanda wrote:
My point is that personal narratives play a much bigger role in how people vote, and those narratives are extremely sensitive to day-to-day fluctuations in the news, and to what extent people are feeling seen and heard. And I believe that line of thought has much more importance for the rural states where ideas and demographics are rapidly evolving.

I mean isn't that exactly why polls are useful? To measure the day-to-day or week-to-week, etc. changes in the mood and opinion of the electorate?

Yeah also a good point.

I just don't know how anyone can build any type of reliable model right now.

Our country is so "in flux." Our suburbs are changing (dramatically in some cases), demographics are exploding, then there are identity shifts going on in both parties. Add to that unreliable polling and I just don't see why anyone would put much stock in a model. People just aren't voting the way "they should."

Like I said, just give me a list of the latest polling numbers and I'll filter that through what I see on the ground.

I spoke about Sinema losing weeks ago and someone piped up that she was solid in polls. That's not what I saw with my eyes. I know that's unscientific, but I drive through Tempe/ASU on my way to work and there are a lot of Red Team stickers on cars. I knew Sinema was in trouble.

--- wrote:

Strat wrote:The doks just appreciates a really solid excel spreadsheet.

Excel? Pffft.

What's your go to program for creating a linear regression analysis?

tragabigzanda wrote:Yeah, and I should add that the other thing that still bothers me about 538 (and we have covered this ad nauseam), is how their 2016 data showed Hillary with a win by a fair margin (less than other popular outlets, yes, I know). I'm convinced the polls directly contributed to voter apathy in that race, and had people opened their fucking EYEBALLS in their COMMUNITIES, rather than relying on the polls, perhaps they would have gotten more people out on election day. The numbers created a false sense of security for many, and while 538 was certainly more conservative than all the other outlets, they were wrong all the same.

You're gonna make 4/5's head explode.

run2death wrote:I spoke about Sinema losing weeks ago and someone piped up that she was solid in polls. That's not what I saw with my eyes. I know that's unscientific, but I drive through Tempe/ASU on my way to work and there are a lot of Red Team stickers on cars. I knew Sinema was in trouble.

Fwiw, this was exactly my experience with Gillum and Desantis and I live in Broward County which is supposed to the Democratic party's stronghold, so I was pretty sure Gillum was in trouble as well.

But that said, both of those races were very close which is precisely what the polls and good models indicated.

tragabigzanda wrote:

McParadigm wrote:

What is this 99% shit??? There are still 178 precincts outstanding, plus lots of partial precincts.

http://mtelectionresults.gov/resultsSW. ... ED&map=CTY

The latest I see is 88% in.

tragabigzanda wrote:

McParadigm wrote:

What is this 99% shit??? There are still 178 precincts outstanding, plus lots of partial precincts.

http://mtelectionresults.gov/resultsSW. ... ED&map=CTY

News outlets almost always have results up before official sites.

run2death wrote:

tragabigzanda wrote:Yeah, and I should add that the other thing that still bothers me about 538 (and we have covered this ad nauseam), is how their 2016 data showed Hillary with a win by a fair margin (less than other popular outlets, yes, I know). I'm convinced the polls directly contributed to voter apathy in that race, and had people opened their fucking EYEBALLS in their COMMUNITIES, rather than relying on the polls, perhaps they would have gotten more people out on election day. The numbers created a false sense of security for many, and while 538 was certainly more conservative than all the other outlets, they were wrong all the same.

You're gonna make 4/5's head explode.

Quite possibly.

I agree completely about other outlets, but not 538 in this case. Some had Hillary at >99% which was just moronic and should have been disregarded ex ante.

538 ended with her at 71%, him at 29%. That's not nothing. A couple weeks before the election it had her down to about 55% but then polling turned in her favor up until the Comey stuff 4 days before the election which wasn't fully baked into the polling by election day. Even still, 538's model gave Trump the same chance as a .290 hitter coming to bat has of a getting a hit. And their model was very clear that if polls were off in Pennsylvania (in Trump's favor) that it would quite likely mean that they were similarly off throughout the rust belt and that their models were quite possibly underselling Trump's true chance of victory. Again, their output can only be based on the inputs they receive through polling.

And yeah, it's probably true that if people could have voted again on Wednesday realizing that Trump could actually win that Hillary probably would have won. But it's not 538's fault people don't understand percentages. I suspect that's why they switched to 1-in-4 type of stuff for this cycle.

I'd argue that people who read 538 should have been less confident in a Clinton victory and more inspired to go vote for her (if that was their preferred outcome) than if you looked only at the polls or read most other predictive models.

tragabigzanda wrote:

--- wrote:So, in general, is it just the quantification of the electoral process that's got you so upset? Maybe I'm just wired differently, but I don't understand how a pop pollster's observations or predictions in any way cloud an individual's voting decision(s). Curious about your experience with this.

...And the experience I shared re: Kathleen Williams here in MT is a fine example: Nearly every Dem I talked to here in MT was pulling for one of the big-money guys to get the nom, and it was often based on the "electability" according to the polls.

"All models are wrong, but some are useful."

One of the most important - and most neglected - parts of managing expectations when developing and putting models into production is communicating that models can and do get it wrong. Often. I'm not claiming otherwise: the world is always changing, and there are real limits to what models are able to accomplish.

However.

The counterfactual - a world in which decisions are based entirely on heuristics or gut feelings or eyeball tests or whatever you want to call them - does not eliminate error either. On the contrary, actually. Models get it wrong, but so do single human brains. For every Kathleen Williams that a model fails to identify as a strong candidate, there are likely two or three other candidates it would correctly identify that a gut feeling or eyeball test wouldn't. All a model has to be is better than the model-less alternative.

This, of course, is an empirical question. I do not actually know how these kinds of models perform vis-a-vis a single human's non-algorithmic predictions, but I suspect their performance is quite favorable. I wonder whether there are any pundits willing to go on-record with a comparison of their performance vs any of Silver's models.