I saw a note that Romney could finally start spending money he raised for the general election campaign now that he is officially the candidate. Up til now he has only been able to spend that which was raised for the primary campaign. I presume Obama has the same issue.
Now we’ll see the yard signs and bumper stickers just explode all over the place and the tv ads in the swing states increase to an unbearable level.
Thanks for the NYT link. My wife says she heard Chuck Todd on MSNBC say that the conservative superpac American Crossroads was reducing its presence in Ohio…but you’re right, that’s obviously not representative of the campaign as a whole. Perhaps she misheard, or Chuck is wrong…or, maybe it’s true, because the Romney campaign prefers to concentrate its direct campaign resources in such a crucial state, and leave the “indirect” superpac stuff for other regions.
Again post-dictions do not need to be limited to fitting the training data. If a hypothesis generated out of “the training set” leads to results in other past data sets that were not considered in developing the hypothesis then the hypothesis can be considered strengthened.
The 538 model traditionally is a forecast of what will be on election day; he has just added a “now cast” that states what would be if the election was held today. The forecast factors in that a current poll will be influenced by a transient factor, like a convention bump, that the now cast does not adjust for. He of course cannot factor for unforeseen events; he can merely state that the odds of one occurring that is large enough to change the results are below a certain percent.
I believe the primary spending is limited by state. That could limit some of the spending. But overall, there’s really no difference. The candidates are going to spend lots of that primary money on general election stuff, like ads attacking their opponent from the opposite party, anyway. So it doesn’t matter that much.
They’re probably going to spend most of it now because voters don’t pay as much attention until after Labor Day.
He’d prefer if we stiff had slavery or separate water fountains.
I read it as a statement of how little power Blacks have - if they as a group all want something to occur bet on the opposite to happen. The only other read is that if they all wanted something to occur the rest of America would find it s threatening they’d all vote against it.
It could also mean the electorate did not like McCain as a candidate. You can’t just assume it’s about the economy.
I read it to mean that if 100% of blacks are for something it means they are all group thinking and incapable of evaluating objectively that one candidate has made it clear he doesn’t give damn about them as a group…I’d like to clarify.
I’ve seen the argument before that blacks are racists since they voted for Obama (of course, white people voting for our few hundred years straight of white presidents are of course, voting on the issues only).
I’d like to hear the explanation of what he actually meant, which is why I asked…
The math in this thread is all weird.
Obama won 365-173. 193 is not twice 173. If Obama wins by 60 electoral votes he wins 299-239, which would be a pickup of 66 electoral votes for the GOP.
Strangely, that would mean he’s saying that MORE support means it should carry LESS weight.
This nonsense again? These people did not, in fact, predict anything in 1980, or 1984, or 1988, or… well, which part of this progression escapes you? Retroactive “predictions” are nothing more than “Bible Code” twaddle.
That’s fine, but until you demonstrate that it actually works it’s still just a hypothesis. As it happens I think your hypothesis on ice cream is likely correct, but then we have a rather enormous amount of data, not just one day of sales every four years.
I am far less impressed by the “we looked at a bunch of past elections and concluded from some disparate economic indicators who will win” is as sure a bet as summer and ice cream when the model isn’t proven,** especially when current polls state Obama will win the election.** I said it in the other thread and will again here; all this nattering over this or that demographic don’t matter when the current polls, and a good, solid analysis of how the polls line up by state, say Obama’s winning. However Romney is doing with a given demographic, he’s not doing well enough. He has to do BETTER than this or he’s going to lose. **Elvisl1ves **notes:
The math is really simple and it has nothing to do with how Romney does with white lesbian homeowners in the top income bracket or Hispanic males who cheer for the Dolphins and have car loans; it’s about swing states. 35 states are out of play, realistically, and those give Obama about 230-240 EVs and Romney about 190-200. Romney must win the remaining states, the big ones, and win almost all of them.
Pennsylvania is probably lost, which means Romey must win Ohio, Florida and Virginia, and that still wouldn’t be enough; he also has to flip Wisconsin, Colorado, AND Nevada, or if he misses one of those, get both Iowa and New Hampshire. Right now all those states are polling slightly for Obama. He is losing. He’s losing even after the convention bump. It ain’t over yet, here’s two months to go, but a model that says he’s going to win is flying in the face of the fact that he is currently set to give a concession speech.
Well, then, pay attention and learn. As has been repeatedly explained in this thread, a “model” that “predicts” the data used to create the model, and only the data used to create the model, (e.g. the “Bible Code”, the U of CO “model”) is worthless.
So you can’t ever look at past data relationships and use them to predict the future?
That’s just science.
You look at the relationship between CO2 levels and global temperature in the past. And then you look at CO2 level trends today. Boom - global warming theory.
Not all models are good, but I don’t see why this is the only reason it wouldn’t be.
Simple:
Bickers-Berry = “predictions” of the events used to build the model = junk science
Silver = predictions of new events not used to build the model = real science
Of course.
It’s a bit disappointing to learn that elections are that simple, but perhaps they are.
I agree. I wouldn’t pick the CO model over polls, or even over other models. I’m just defending the idea of models in general.
Nope.
If you build a model that explains past events, it’s very good at explaining past events.
Why did the dinosaurs die out? A meteor? If it fits, it’s a model. Doesn’t matter that it isn’t (and can’t be) tested because we have no new dinosaurs to watch go extinct.
Sigh.
It simply means that it fits the data that was used to create it. Which is understood. They aren’t saying they actually predicted anything. If the model had been around in 1980, it would have predicted the elections. That’s all. That makes it a decent model.
Yes, you have to use one set of data to develop a model that predicts another set of data (e.g. compare temperatures and CO2 levels for 1970-1990 and see how it predicts observations for 1990-2010). You can’t use the same data set for both, or you’re just Bible-Coding a circularity into your logic.
That’s true enough, for things we can’t observe future events for. Ideally what we would do is build a model based on known data for a subset of past events and then see if it predicted other known past events. That is, fit to a lot of data and then use the resulting model to make predictions against other known data sets (or predictions about past events we don’t yet have evidence of but might some day - new fossils, for example).
What you don’t want to do, however, is fit to all of the known data (particularly if you have more variables than data points) and then use it to predict future events. It just doesn’t work well that way.
I would really like to see Silver do an analysis of the full paper when it comes out - he generally is very fair, but tough, on these sorts of things. One particular thing I’d like him to try is to remove one election from the procedure, run the regression, and then use the result to predict the “removed” election - how well does this match up with the true result. This is the kind of robustness I’d require before giving the model any weight in making future predictions.