No matter how many times you say it, they are, in reality, the same. The reason the simple example is supposed to help you understand is because it doesn’t lead you to be confused about whether a parameter*should *or should not have predictive power.
In practical applications there will be parameters seem like they should, like unemployment, and parameters that seem like they should not, like average daily temperature. However, even the most logically obvious or absurd assumption is just that until you test it with actual training data.
I believe you agree up to this point.
The problem is if you indiscriminately toss in as many parameters as you like, even if they all seem like they *should *have predictive power, your model will be over fit. Exactly like my earlier example and Jragon’s example. It’s not different because the parameters appear on their face like they should be predictive. You will find parameters that appear to fit your training data very well, and also logically explain people’s voting power. But because you have too many parameters you’re just mistaking noise for pattern - or you might be - you can’t tell because you have too many parameters.
Thems are some interesting figures. “Independents” must be identifying as such at record low numbers (or filtered out as not likely to vote) for all likely voters to be coming out 52 to 46% in favor of Obama, with Democratic and Republican spreads being about the same (97 and 96%) in favor of their party’s candidate, and that big of a spread for Romney among those who call themselves “independent.” The alternate explanation is that many more Democratic voters are likely voters, which is never the case, and in fact has been estimated to be even less the case this cycle than in many others.
Interesting in those cross tabs is not the 94 to 6% Obama lean among liberals, but that the lean is 59 to 37% among moderates and that Obama is even getting the nod from 23% of those who identify as conservative. When you lose the middle and fail to excite your base you are pretty much in an underdog position.
I expect these Ronulan delegates would hold their noses and vote for Mitt if it actually matters, but if he gets blown out anyway they’ll be more inclined to go ahead and raise the middle finger to the party establishment.
You realize that a lot of these “independents” are tea party whackos right?
In any event. Its going to be a close race and anyone who thinks that Obama has got this in the bag this far out from the election is nucking futz. Right now, Obama is ahead in the horse race.
I’m pretty sure that the assumed enthusiasm gap goes the other way.
Not if your base is women, minorities, the elderly, the middle class and the young.
2010 was a combination of very high turnout among the tea party and very low turnout among the Democrats.
An enthusiasm gap is denoted as the difference in likelihood that persons of one party will vote versus persons of another party. It is not the difference between a party’s registered and likely voters. As it stands, not only am I pretty sure you didn’t understand my last post (because there’s nothing in 538’s newest post which explains why I’m wrong) but you haven’t read any of my prior posts, either.
I don’t know what polls you’ve been looking at, but for the entire election season there has been a distinct enthusiasm gap between Republicans and Democrats in the Republicans favor. That’s, historically, how it turns out. Republicans tend to be more likely to vote than Democrats but there are simply more registered Democrats than Republicans, so Democrats could conceivably win by just having all of their base come out and vote (as I said on the very first page). Obama could lose Independents and win the election while Romney really cannot. Now, if Romney wins Independents by 10+ points, he will win the election as-- by my totally non-expert calculations-- it would take a D+4/D+5 turnout to offset being trounced in the Independent category. 2008 was an anomaly; Democrats won’t see record high turnout and deflated Republican turnout like they did then.
[QUOTE=iiandyiiii]
You had to dig pretty deep in those polls to find some good news for Romney.
[/quote]
No, I didn’t. All you have to do is click on the demographic breakdown. It takes, literally, twenty seconds to do and I’ve been doing it for quite a long time now.
An independent is only an independent if they have no lean/opinions on each political party? Hmmm, what? Surely you can see what’s wrong with your above statement, correct? It would cut both ways. I’m not so sure how many times I need to point this out, but voting characteristics are not immutable. Solely because some person votes one way today, does not mean that person will vote the same way tomorrow or that because a group votes differently than they did yesterday that the group must be comprised on different individuals. It’s an inference fallacy to assume so.
That’s great. Now what do most moderates identify themselves as?
No. I’m referring to the difference between the support for a candidate in a poll of registered voters and a poll of likely voters. It has nothing to do with party affiliation. The one thing you are correct about is I almost always skip your posts.
As I said in my last post, for this entire election season the enthusiasm gap has been unusually large. Finally, since the convention, the enthusiasm gap has close to what it typically is. You’re right that it favors the Republicans. The point is it has finally start to close to a normal gap while it had previously been extra large.
The problem for republicans though is that even with the extra large enthusiasm gap, they were still losing.
The math doesn’t add up for that poll, which has roughly the same margin overall for Obama as he had in 2008, if the same independents who leaned Obama in 2008 are now breaking for Romney by 14 points. The best explanation for it is some combination of these two possibilities:
Some of the Independents from 2008 who were for Obama now call themselves Democrats
Some of the Republicans from 2008 who were against Obama now call themselves Independents.
Without looking out how many of those polled were independents (and I couldn’t find this), it’s hard to tell exactly. But this is the most logical explanation.
Do you have an explanation for why the poll shows an overall result similar to 2008, but wildly different results for independents?
Omg- you won’t bet money, but how about this, if you’re so confident? If Romney wins, then on the day after the election I will post (in this forum or the MPSIMS forum, whichever would be more appropriate) something to the effect of “Omg a Black Conservative has superior political knowledge to me. I weep in the presence of his wisdom. That is all.” (you can modify this if you want) And if Obama wins, then you post something I want- something like “iiandyiiii’s political knowledge is superior to mine. I acknowledge that the country is changing politically and leaving me behind, and my political beliefs are as obsolete as a cassette recorder”. And I’ll bet you could get plenty of takers to this bet- plenty of people that would be willing to risk their humiliation against yours (to avoid cluttering up the forum we could have one Omnibus Presidential Election Schadenfreude ). What do you say?
Yes, I know, it’s the same(-ish) thing. Multivariate fitting is more or less normal trend-fitting for single-dimensional cases WITH MATRICES AND VECTORS! Dependence relationships complicate things, and while the answer may turn out much different and you have to use a couple more methods than normal, it’s still effectively the same. In prediction and trend-fitting model creation is much more difficult than the math. The math is pretty uniform between models, it’s making a good model that counts, and it’s all potentially subject to the same problems like overfitting.
I haven’t read precisely what they did, but my guess is that they did normal multivariate fitting and analysis as you’d do for a case with a continuous output, and then they said that if the predicted output value is bigger than <x> it maps to either Dem or Pub, and if it’s less than it maps to the other (this is called thresholding, and is a relatively common way to map a continuous – single or multi-variate – input to a discrete or finite output).
All the problem I linked said was “If I know X, I want you to tell me Y”. In this case all it’s saying is “If I know X,Y,Z,A,B,C,D – some of which are interdependent, I want you to tell me E” (and I’ll map E to Dem or Pub). It’s susceptible to the same problems – more problems, actually, since it’s more complex, they’re just harder to see in more dimensions than 3.
Evidently, you did not read what it was you responded to. As a general rule, I tend to not like having to go in detail because it’ll be ignored anyway, but I have some free time. So here goes. Just to reiterate, here is what I said:
Registered voters do not matter; likely voters do. Guaging political affiliation split of likely voters is a bit of a gamble. Most polling firms simply look at past election turnout to measure how likely voter turnout.
1.) For the ABC/WaPo poll, the split is D+6 at 33/27/36. That’s not realistic for a presidential election. Again, check out the historical voter splits since 1984. Outside of the 2008 election, Democrats haven’t had more than a four point edge in percentage of votes cast. Republicans casting only 27% of the vote would be the worst in the shown 24 years while Independents casting 36% of the vote would be the highest recorded. Even the Democratic sample is a bit underweighted as they’ve cast at least 37% of the votes in every election since 1984. I say a split of 37/35/28 would be more in-line with historical voting trends, which would put both Obama and Romney around 48% each. Either way, if Romney continues to lead among Independents then the election comes out to turnout, in which case I believe Romney will win.
2.) CNN doesn’t post their splits, but it has to have severely undersampled Independents in order for Obama to come out at +6 among likely voters while losing among Independents by 14%. Either that, or they filtered a lot of Independents into the Democratic column. Again, I’m no Nate Silver, but I’m apt to think that the actual likely voter split is a lot closer than what was posted by CNN.
I have a simple explanation; Independents aren’t party line voters like Democrats and Republicans are.
That doesn’t explain that poll’s results- again, the overall results match the 2008 election- roughly 52 to 46 for Obama. Obama won independents in 2008. So for Obama to be losing independents by 14, and still be ahead overall by 52-46, then the number of independents is probably smaller, which means that some of those who were independent in 08 likely now consider themselves Democrats.
So do you have an explanation for how the overall result matches 2008, but the preference of independent voters is so different from 2008?
Some may be, but not only is that not what you said the first time, that shift would be very minimal, at best. Go back and click on the link I gave you.
I did. For one thing, it’s from different polls. For another, the two together just don’t make sense mathematically, unless one of them is wrong. I guess your view is that the overall result of the CNN poll is inaccurate, but the independents’ results inside the CNN poll is accurate, and the state polls you cited are accurate.
Considering the many polls that show Obama with a decent margin ahead of Romney, I think the CNN poll overall results are pretty accurate. If his overall margin is similar to 08 (and I think the polls show it is, though perhaps a couple of points closer), and the margin for independents is wildly different than 08, then the composition of the electorate must be somewhat different. The logical explanation that fits the data would be that there are more Democrats and/or fewer independents and/or fewer Republicans as compared to 08.
Unless you don’t see most of the overall results of the polls as accurate, then I don’t see another explanation.