Well your comment about “enough to criticize without …”, when I actually made no criticism at all, suggests some sensitivity to perceived criticism of the model. But no big deal. At any rate, you’ve not substantiated any of your assertions on this topic, so your comments here apply to yourself as well.
As previous, my comments were accurate, best as I can tell.
I think direction counts a bit more, in light of the plethora of polls. Gallup shows Obama gaining a point each of the last two days. Rasmussen has Romney gaining ground. As does IDB, although they still show Obama with a 2 point lead.
Of course, none of these polls reflect last night’s debate. Or the stock market losing 200+ points today, FTM.
One dark horse wild card is a story someone is peddling involving supposed evidence that Obama dealt cocaine in college. Supposedly it was offered to the Romney campaign and they turned it down. I imagine the mainstream media will bury it, but you never know if someone else will succeed in bringing it to public attention. If that happens it could influence public opinion, although in my opinion it’s a lot more likely to end up helping Obama than hurting him (much in the way the Monica Lewinsky scandal helped Clinton).
Well, misrepresenting the model and then identifying sources of error and instability based on the incorrect methodology is what most people would call criticism.
That doesn’t make any sense. I said “I’m just suggesting that before discussing the model’s methodology, you might want to actually find out what it’s methodology is.” How exactly does that apply to me?
What I refuse to do is cut and paste his methodology statements for you to read when you can just easily look them up. It doesn’t mean I haven’t read them.
Does your best include actually looking at the methodology this time? Because I would be interested to read your thoughts on the actual model.
Well I didn’t misrepresent it, and I didn’t identify “sources of error and instability”. I merely noted that the model is what it is and it’s not what it’s not, and if you misunderstand what it is, then you’ll be apt to be puzzled when it changes in ways inconsistent with your misunderstanding.
You’ve made assertions about what his model does, and upon failing to back them up, you’ve announced that you can’t be bothored to look for sources. [In one case, you resorted to misrepresenting what I said, but that’s another issue.]
I don’t have much trouble reading and interpreting the posts on this board, but for some reason I’m apparently misinterpreting even your clearest statements. Don’t know whether it’s me or you, but it’s obviously not a productive dialogue.
If you decide to read about the model’s methodology, I’ll be curious to hear what you think of it.
I read the methodology article that you linked. Seems extremely thorough. Something of a black box in terms of his statistical modeling (not that I would get into that anyway). Huge asset to poll followers like myself and far more valuable than these purely economic modelers out there. A shame that he’s partially behind a paywall.
Until there is more evidence, I put this in the category of Glenn Beck and that poor girl in 1990 – As far as I know he still hasn’t denied being involved with that.
Well there’s more to it than the Glenn Beck “story”, but that doesn’t mean there’s a shred of truth to it, and there probably isn’t. But that’s another issue. What I’m discussing here is how it - baseless or not - might influence the race if it gained some publicity.
Sure. To start, I’m curious what you and other model-watchers think about how the 538 model incorporates economic data but then slowly phases it out (on the reasoning that at some point that stuff is “baked in.”).
One of my questions is about how to determine at what point it is sufficiently baked in to start decreasing the weight assigned to it. It seems to me that you’d need to know whether a difference in economic indicators between October and early November has ever made an impact on the result. Not sure if anyone’s done that research, but I’d be curious to hear about it if they have.
With an unemployment report occurring just before the election and with all the focus on those numbers, and with the significant uptick in consumer confidence in recent weeks, it seems possible that the economic stuff is not quite baked in yet.
There’s a difference between “economic data” and potential “changes in economic data”. Nate Silver is talking about the former and you’re talking about the latter.
What a lot of these economic modelers are doing is running regression models which show things like “for every point that unemployment is above X% the incumbents loses Y% of the vote”, or something analogous (for a whole host of economic indicators). There is rationale to use economic models in the face of polls early in the process, based on the notion that people are not really paying attention, and the parties have not begun running their campaigns yet and so on, but this becomes less of a factor as time goes on. After a while, if people were really going to oust the incumbent because of unemployment, then polls would show it.
But if the economic numbers change, then all bets are off.
One of the things we’ve been disagreeing about here is that I don’t think Nate’s model can or does predict the likelihood of (known or unknown) future events changing public sentiment, and this includes changes in economic conditions along with any other sort of future events.
[I disagree with you about consumer confidence, though. Any prior changes in consumer confidence reflect present conditions and are already baked in to the current polls. I would think Obama’s chances would be somewhat correlated with consumer confidence, and the fact that his polls have not risen along with those numbers is not a positive for him. (FWIW, the reason I earlier mentioned the stock market tanking is because it has the potential to hit consumer confidence, and hurt Obama.)]
(I’m giving in on not looking this stuff up for you, I hope the paywall forgives me.)
As you can see, the model is calibrated to take into account changing economic indicators. It “predicts” these changes insofar as the weight assigned to each factor at a given distance from election day is a function of how predictive (and therefore stable) that factor was at that time in the typical election.
Or am I misreading what he’s written in that post?
I don’t see where he writes that (& I assume he weights the factors are weighted based on some regression analysis, as above). Perhaps you can quote the part which says that.
He’s not talking about how the economic factors are weighted against each other. He’s talking about how they’re (collectively) weighted against the (collective) polls. What he saying is what I wrote in post #129, second paragraph.
Huh? Who said they were weighted against other economic factors? I said the model reflects changes in economic data, contrary to your *underlined *claim in post #129.
Sorry, this is just getting a bit too bizarre for me. I’m gonna let this one end here.
I thought when you said “each factor” that you meant each economic factor (& that these were weighted based on their predicted stability). Sorry.
But he doesn’t say - at least in the part you quoted - that the model reflects the likelihood of future changes in economic factors. Only that since early on economic factors are a big part of the model, actual changes in economic factors have the potential to sway the model even absent a change in the poll numbers, more so than later on when they carry less weight and won’t impact the model numbers as much.
What that means is that if it’s June 1 and the unemployment rate goes up, the model will show a decline in Obama’s chances, even if the polls don’t move. If the same thing happens on November 1, then the model will show a much smaller decline (unless the polls also move in response). This makes a lot of sense, as described above - since as you get closer to the election the economic number are increasingly reflected in the poll numbers, and thus carry less weight on their own.
What he doesn’t say is that his model incorporates any sort of prediction as to what unemployment (or any other economic measure) will be at any time in the future, which is the issue you raised in speculating about the impact of the early November jobs numbers.
So today we have Ras/Gallup at R+4 or 5, with two new Ohio polls (SUSA and Ras) showing O+3 or tied in Ohio. The national average at RCP is now R+0.9 with Ohio at O+1.7.
Do we really think that Ohio is D+2.6 (or more) from the national average? It’s historically more like R+3. Has Ohio really changed that much, or is somebody wrong?
I think both will revert to the mean. The Ohio number will drop a little and O will gain the national number a bit. Depending on what day in the last four you ask this question, you’d get different gaps. That’s just noise, I think.
Recall that, notwithstanding the Republican advantage in registration, Bush underperformed in Ohio in 2004 by .3%. It wouldn’t shock me in the least if Romney underperformed there by .5-1.5%. I think a bigger gap than that would be surprising, but I don’t think that’s what we’re really seeing. I think the national race is about tied and Obama has a small but consistent lead in Ohio. Getting more detailed than that is asking for more than the polls, and certainly a given snapshot, can provide.
Since 1992, the share of the white vote has never increased in a subsequent election. Moreover, the share of the minority vote in 2008 was exactly in line with the trend, suggesting it was not some sort of anomalous bubble.
If Republicans are banking in the white vote increasing for the first time on 20 years, I think they are going to be quite disappointed.
I’m not sure they are banking on it, since that would seem to be pretty silly, and since not all polls are adjusted by race. As far as I can tell, what demographic characteristics polls use to adjust their samples depends on the poll and the state. Not all of them adjust for predicted turnout by race. Some adjust by gender, or age, or zip code. But the thing is, those factors are all also quite stable.
Is there any evidence that, demographically speaking, 2008 was some kind of outlier from the trend of the last two decades?