tag:blogger.com,1999:blog-6926413038778731189.post680670283196662267..comments2023-10-16T07:13:12.123-05:00Comments on A plain blog about politics: Notes on a NerdfightJonathan Bernsteinhttp://www.blogger.com/profile/15931039630306253241noreply@blogger.comBlogger7125tag:blogger.com,1999:blog-6926413038778731189.post-39197433585117274862012-03-31T19:06:56.867-05:002012-03-31T19:06:56.867-05:00When I read Nate Silver, I often get the impressio...When I read Nate Silver, I often get the impression that he has made significant money on predictive models of how certain sports events turned out.... especially with his emphasis on the importance of point spreads. He expresses no opinion on policy or partisanship. Politics has interest to him because it is a new arena for his tradecraft, and apparently pays well. He's a breath of fresh air.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-6926413038778731189.post-17303821386560556012012-03-30T23:43:26.569-05:002012-03-30T23:43:26.569-05:00Warning: EXCEDDINGLY long answer follows.
Now we&...Warning: EXCEDDINGLY long answer follows.<br /><br />Now we're getting deeper into my area. For my money, the model to beat all models is Jacobson's. He gets R squares north of .8 with just a few variables. And the one that does just a TON of heavy lifting is a proxy for a local variable: candidate quality. What Jacobson does is total up the % of Dem challengers who are quality (ie, have won an election for SOMETHING before) and subtract the % of the GOP challengers that are similarly qualified. That quality difference just explains a ton. Now, itself, it's also predicted strongly by the state of the economy (transformed right: so the bad economy in 2010 is good for the Reps, and the bad economy in 2008 is good for the Dems, because voters anchor on the party of the president).<br /><br />What's really funny is that my own work (unpublished) and that of Eric McGhee (also unpublished, and you'll see why in a second) trying to use this same variable measured better get it to NOT work. (This is unpublished because it's a null finding that contradicts accepted theory and what we KNOW to be true: incumbents win their elections. Either we've done something wrong, or we've got the wierdest emergent property ever...one that is emergent only in the way we count it, not in the simpler, less sophisticated ways that OTHERS count it! Until I figure out how to square this circle, this paper remains on the shelf.) <br /><br />Now, does it matter what happens in Congress? Well, the more popular models don't include that. Rather, what you have are contributions looking at specific election years that often find effects, usually in the form of people who voted for X were more likely to lose. So, there's a few people that have found that for votes on the ACA, for example. (It's naturally not very practical to do the hypothetical JB often poses: how bad would it have been if it hadn't passed? Just as it will be difficult to prove effects of voting for the Ryan budget because they already voted for it (unless anyone changed their votes, but your leverage is going to be SO TINY on that, that you won't find anything)<br /><br />But, to return to the question, does it matter what happens in Congress? My dissertation says no. Sarah Binder's research suggests no. What I found was the passing major legislation helps presidential approval. It has no real impact on congressional approval besides that moderated through presidential (when we ask people what they think of Congress, we're doing even more of a "how are things going" than we are with the presidency, and so the "right track/wrong track" element carries through both measures). And, studying the effects on congressional elections (which, I modeled using some really cool data that took me MONTHS to collect, so I have more proxies for previous electoral margin than anyone else I'm aware of)....found no effects. So, that's overall productivity of major laws, and over the entire postwar period. Individual votes on individual bills can still have an effect, even if aggregate productivity doesn't. Which makes sense, since how do you credit invidual members for the productivity of the system?Matt Jarvisnoreply@blogger.comtag:blogger.com,1999:blog-6926413038778731189.post-68641428802667902112012-03-30T23:07:12.738-05:002012-03-30T23:07:12.738-05:00Thanks for the clarification. In other words, (or...Thanks for the clarification. In other words, (or maybe pretty much the same words) Abramowitz's model just isn't built in such a way that it can take into account the effect of the things we have been talking about. Fair enough. <br /><br />Still, the general question remains, is there any consensus, or informed body of opinion at all, with regard to the general accuracy of Congressional models versus Presidential models? It would seem that Congressional models would be a lot trickier, since local factors would presumably loom much larger in the individual races. But maybe that is just a naive view.Anastasiosnoreply@blogger.comtag:blogger.com,1999:blog-6926413038778731189.post-47754725047939743792012-03-30T22:43:05.628-05:002012-03-30T22:43:05.628-05:00The idea is that Abramowitz's predictor doesn&...The idea is that Abramowitz's predictor doesn't include (except in the generic ballot so far) the effects of GOP self-damage. So, first, if House Republicans were doing more popular things, they might be doing better on the generic ballot and thus predicted to do better. But, second, it's at least possible that the self-damage could wind up making Republicans underperform compared to the predictors.Jonathan Bernsteinhttps://www.blogger.com/profile/15931039630306253241noreply@blogger.comtag:blogger.com,1999:blog-6926413038778731189.post-50609479304407401252012-03-30T22:26:06.655-05:002012-03-30T22:26:06.655-05:00I have to say, if these predictors are parlor game...I have to say, if these predictors are parlor games, and viewed so by the poli sci community, then there is an enormous amount of gassage going on in the press and blogosphere about things that don't amount to a hill of beans (not that such should be a surprise to anyone).<br /><br />To pick up on doc's point, it would be interesting to know the status of Congressional predictors vs Presidential predictors. There has been a lot of talk the past couple of years, including on this blog in the last couple of weeks, about the inherent illogic of the GOP strategy in the House. Taking tough votes on things that can't pass, backing wildly unpopular policies, following a delusional strategy, etc. Yet, if Abramowitz is right with his Congressional model, this has hurt them ... not at all. He is predicting holding their losses in the House to three and a 6-7 seat pickup in the Senate. That certainly doesn't sound like a party that has been hurt. Is his model flawed (I believe he himself admits that it has a very large error on the Senate side)? Or do all these votes in the House we have discussed on this blog ... actually matter not at all? Does the debt ceiling debacle that we talked about damaging Congress ... actually matter not at all? Is the delusional attachment to unpopular policies actually a delusion on our part (about the impact of that) and not theirs?<br /><br />Maybe Paul Ryan is right, and there simply isn't a penalty to be paid for going off the deep end. Or maybe the penalty is only paid once the policies actually get passed? If that is so, we are in for a world of hurt, since the Republicans will pay no penalty for their policies until they have actually passed regressive tax cuts and damaged Medicare. Bummer.Anastasiosnoreply@blogger.comtag:blogger.com,1999:blog-6926413038778731189.post-29756224712289143422012-03-30T22:10:29.192-05:002012-03-30T22:10:29.192-05:00I don't do election forecasts for a couple of ...I don't do election forecasts for a couple of reasons (a) it's not my field and (b) I don't want to look stupid in public. I will, however, say a couple of other things, based mostly on having a little knowledge of models and of statistics. (1) You probably want to look at different levels of elections differently (House of Representatives separately from Senate separately from Presidential separately from state legislatures...). (2) If you're trying to create a model, in most cases you have relativelyfew observations. Take Presidential elections. This year is, what, number 56? 57? If you want to use any sizable number of explanatory variables, you're going to lose all of the 18th century elections and probably all of the 19th century elections and at least some of the 20th century elections. Good luck developing a useful model from what's left.<br /><br />But I should really let people who create these models defent themselves...Don Coffinhttps://www.blogger.com/profile/07198988872512792834noreply@blogger.comtag:blogger.com,1999:blog-6926413038778731189.post-78444231337610862272012-03-30T17:24:41.501-05:002012-03-30T17:24:41.501-05:00Good points.
The point that I would make to Nate i...Good points.<br />The point that I would make to Nate is that you have to actually get under the hood of the models. Just looking at predictions and RMSE or R2 or whatever your metric is misses the point. What would be much more interesting is a comparison of the beta weights in models that include both polling and fundamentals. And, methinks that the fundamentals are going to really swamp the polls.<br /><br />It's not that polls don't add useful details for the prediction. It's that the basic nature of the race partially determines them, too. I liken it to golf. Yes, the short game is very important. But, if you face 10 degrees in the wrong direction off the tee, you're hosed. In golf, knowing which general direction to face off the tee is easy...but it's also fantastically importnat. Drive to the east on a northerly fairway and you're not going to do very well.Matt Jarvisnoreply@blogger.com