A "polls only" model with a Dems at a 93% chance of winning the Senate today is [censored]. No way reasonable assumptions get you there.
— Nate Silver (@NateSilver538) September 19, 2014
What you see above are the two "money shot" prediction forecasts from Princeton Election Consortium (top) and FiveThirtyEight (bottom). While these are not precisely identical (the PEC visual is where the race now-stands, the FiveThirtyEight is the November election prediction) the fact is that they are (a) pretty similar and (b) bitterly disputed.In the smack-talk war between Nate Silver and Dr. Wang, the battle ground is Twitter and the Battle Space is over whose model is better.
For right now, not a shred of difference between "polls" and "polls+" Senate models. There's a difference between Wang's and everyone else's
— Nate Silver (@NateSilver538) September 19, 2014
Open to critique of our simple model: polls will do what they've done Jun-Sep. But @NateSilver538 critique inaccurate http://t.co/odR2tUv4LO
— Sam Wang (@SamWangPhD) September 18, 2014
The big difference? The meaningful difference? Despite Democratic numbers kinda "crashing" right now PEC gives Election Day Probability of Democrats keeping the Senate at 62%.FiveThirtyEight gives them 40.7%
Wait A Minute--Which Is Right??
This is math, science, and reality--right? Objectively one of the two models has to be better. One of the two has to be right. Maybe. The problem is, The Omnivore can't tell you which better. Firstly, they were both very, very accurate in 2012. Secondly, their methods have changed slightly since, say, 2008 (which would give us a large enough sample size for comparison). It's also important to note that in 2012 PEC had a larger confidence of victory than 538--so this is the same pattern repeating here.
Thirdly, while PEC's model is pretty transparent and 538's explicitly contains secret sauce, by the time we reach November the models will have more or less converged (the secret-sauce is supposed to tell us where the polls will go--but The Omnivore doesn't have an understanding of Silver's mathematical degree of confidence in his prognostication to be able to say if the eventual convergence will be within that boundary). If they both wind up at more or less the same place in the end, is there really a difference?
Finally, The Omnivore is no quant. This is a battle for quants.
But that's not what interests The Omnivore. What interests The Omnivore is the weaponization of good news.
The Weaponization of Good News
In 2012 the polling aggregates--even within the Romney campaign showed he was behind--was pretty much always behind (save for just after the 1st debate) and really never had a chance anyway. What this "sensor data" (which is what polling is for our purposes here) was telling Republicans was that they were gonna lose.
What the Republican rank and file, pundit class, and bloggers did was respond by making up good news. They "unskewed the polls" to adjust the numbers until, hey, Romney was gonna crush Obama. Then they weaponized it.
This "weaponization" was taking the phrenological science of poll-unskewing (or just ignoring polling altogether) and turning it into a mass-media message. This gave it the cable-news credibility to penetrate deeply. Today there are people who feel certain there was unprecedentedly massive levels of election fraud so that Obama could win. How can you blame them? Karl Rove, at 10:30 PM on election night, seemed to believe the same thing.
The weaponization of this--taking what is essentially a placebo-talking point--and amplifying its signal-strength and credibility to the highest possible levels was driven in part by human nature (denial, observation bias, etc.)--but also in part by money. If The Omnivore were discussing a phenomena that wasn't just a pernicious lie, it would be called monetization.
The drive to monetize fake good news is also pretty central to human nature. If there is a demographic hungry for something, in a capitalist society someone will come and fill it. That's a feature, not a bug--so long as filling it isn't done with toxic food, exploding Pintos, and the like. It's also a very powerful motivation. People in a position to fill a need are generally driven to do so--there's profit in it.
So The Omnivore doesn't really blame the pundits who predicted a Romney victory based on fundamentals. The Omnivore doesn't quite blame the masses who were told--from the highest levels--that the GOP was poised to win. But The Omnivore does blame the people who unskewed the polls in the first place with the veneer of scientific "mysticism."
The Omnivore totally blames the media outlets that promoted this thinking--major bloggers, senior pundits, and television personalities who promoted this thinking. Whether it was really believed or not, (a) it should not have been and (b) the behavior was predatory.
So the real question is: Is this happening with the poll aggregators? Is PEC choosing its algorithm to "beat" 538 on the good-news front? Today, facing a prospect of (likely) losing the Senate, are the Democrats also "weaponizing" good news?
It Doesn't Look Like It
The Omnivore doesn't think so. Nate Silver has the actual monetized property to lose--he doesn't seem to be altering his percentages to "keep up" with PEC's sunnier forecast. He's unquestionably a Democratic sympathizer (as is Dr. Wang of PEC) so there isn't a partisan leaning keeping him from making more positive prognostications.
Indeed, while PEC's formula is out in the open, 538 uses internal adjustments which he could alter without it being obvious. If 538 felt it was competing with PEC (and it does--just, maybe, not for money) it could change things to compete.
The fact that this is not happening should be troubling for Republicans. While we might argue that the burn the GOP took in 2012 would have frightened off the Democrats from doing the same--but that isn't how people's brains really work. We might postulate that the mathematical nature of poll aggregation doesn't lend itself to lying--but poll unskewing was mathematical. It took place down in the cross-tabs. It took issue with the behind-the-scenes numbers. It was bad math--but it was math. Silver could convince himself to change some values and pop the victory % up to ... 55% to a Democratic Senate? That'd put him in the "good news" category.
No--it appears that for whatever reason the (democratic) poll aggregators are sticking to their guns. They're arguing about models--and not competing for eyeballs. The eyeballs are a result of the models, yes--PEC's rise in popularity is most likely partially because of its improved chances--but the positive feedback loop where it improves chances as eyeballs go up isn't happening. And 538 isn't increasing to "try to keep up."
Why not?
Why did a reaction that was pervasive amongst Republicans fail to take place with Democrats?
The Omnivore isn't sure yet (intellectual integrity is possible--but The Omnivore is loathe to ascribe that to people in general). There are a few weeks left--and as Democratic chances dwindle (and they probably will), The Omnivore will be watching closely to see if anything changes.
Two reasons, at least.
ReplyDelete1. World view
2. Echo chamber
These are tightly intertwined, but the world view of many on the right is informed by (ironically) feelings of persecution and victimization. They believe they are being attacked on all sides - by the "MSM" or "Libruls" - and this leads to a defensive posture where empirical evidence and fact are just words brandished by those attacking them (though not clear on which is cause and which is effect). This is then re-affirmed by the Fox/Talk Radio echo chamber, and the disdain/distrust for "empirical" thinking is further eroded.
I have found, anecdotally, but anecdotally across hundreds and hundreds of cases, that those on the left are more willing to question their convictions and beliefs when presented with evidence to the contrary, whereas on the right you get the defensive reaction described above.
The Omnivore is very skeptical of non-data-driven Us-vs-Them ("Democrats -are- or Conservatives Tend to --") arguments. Sure, some do seem to hold up (urban vs. rural)--but once you get to "feelings of victimization" The Omnivore would rather hear a conservative saying this than a Democrat.
DeleteThat said, I think that the Echo Chamber effect was, yeah, at least a major factor in signal amplification. People share news they like and major news outlets are responsible for that.
But if it were not vices vs. virtues that's driving this, what -else- might it be? One serious possibility is that the poll aggregators are driven by intellectual pride in their predictions: wanting to be right is more important than getting eyeball-money / web hits.
In this sense Dr. Wang and Silver are essentially -athletes- and this is a sports competition. The public is in the position of "attempting to pay someone to throw the game" (not really--but in the metaphor, Democrats paying more attention to more positive positions is kind of like that)--but ultimately not paying enough.
The political pundits are not in the same class--because although in theory they have to be right to be respected, in practice they are respected because of their history and their job is appearing on TV. To appear on TV they have to deliver good news.
If we want to add a vice in here, it may also be that the current RINO hunting has made -any- apostasy -deadly- on the right.
-The Omnivore
"This is math, science, and reality--right? Objectively one of the two models has to be better. One of the two has to be right. "
ReplyDeleteNo, they are both models. Models are inherently simplifications of a complex reality. Models are never right, because the underlying truth is always more complex than the model.
Ok--well, fair enough--but the point is that there has been a track-record of predictions based on each model and even evaluating which track-record is better is still difficult.
Delete-The Omnivore
Silver and Wang are real scientists. They might vote Democrat, but their first allegiance is to reality.
ReplyDeleteSo you can expect them to behave much better than partisan hacks do.
Far as I can tell, that's all there is to it.