Subject: Phony Science? Try Phony Journalism
From: Jim Campbell, Professor of Political Science, University of Buffalo, SUNY
Date: Jun 5 2:52 p.m.
A core value of election forecasting research is accuracy. I would commend this as well to Karl Eisenhower and Pete Nelson (and to the editors of Slate). Their piece on election forecasting is replete with errors and cheap shots. Let's get specific:
1. Eisenhower and Nelson say that the forecasters predict the obvious. At least E&N's hindsight is 20-20. First, in retrospect, all election results look predictable. However, when you compare the forecasts to polls or pundits, the record of the forecast models is terrific. As is documented in Before the Vote, the seven forecasts of the 1996 election made two-to-three months before the election were at least as accurate as a group as the national polls reported in USA Today on Election Day. Are the forecasting models perfect? No. And as a result they (like the polls or pundits) may miss very close elections, such as the 1960 and 1968 elections. However, when it comes to reliably accurate forecasts of the national vote split, the hard record clearly indicates the greater accuracy of the models.
2. E&N snidely suggest that the models have "ducked" the hard calls, most notably Truman's 1948 comeback. This, again, is nonsense. All of the models include all elections in which data for their predictor variables is available. If a forecaster excluded a year just to help his model, someone else would estimate the model with the year included. There is some healthy competition here and the data that we are all using is publicly available to anyone who wants it. Moreover, four of the seven models included in Before the Vote include the 1948 election in their analyses. Those who do not include 1948 cannot do so because they need data that was not collected at that time. Again, E&N should read before they write.
3. E&N criticize the forecasting models because they use polls in generating their forecasts. Well some do and some do not. I would think that the use of polls in forecasting would make the forecasts more credible to E&N, but consistency, like accuracy, does not seem high on their list of values. My perspective of forecasting, using polls in the forecast, is that it is a more sophisticated way of reading polls. If a candidate is ahead by 60-40 at a particular point in the campaign, should we take this at face value or try to put it in some historical context and the context of the current economic climate? I think that an intelligent observer would do the latter and that is what several of the forecasting models do. For the models that do not use polls, I think that they are taking on the tougher task of trying to forecast with one hand behind their back. However, if they can obtain good measures of what tends to influence the electorate, then more power to them.
4. E&N criticize the forecasters for making multiple forecasts throughout the election year. Say I do a model based on data available in July, and then another on Labor Day. My analysis indicates that the Labor Day prediction is the most accurate, and so that is the one that I rely on. So long as the specific model is cited as the one producing the forecast, I see no problem other than the fact that it may create some confusion among those who are not paying close attention. Apparently, E&N can be counted in this group.
P.S.: If Eisenhower and Nelson had read Before the Vote, they not only would not have attributed it to Michael Lewis-Beck, but would not have incorrectly listed my affiliation as LSU. It is noted at least four times in the book that I am affiliated with the University at Buffalo, SUNY.
[To read an unedited version of this post, or to reply, click here and scroll to bottom.]
Subject: Re: Phony Science? Try Phony Journalism
From: Karl Eisenhower and Pete Nelson
Date: Thu Jun 15 11:42 a.m.
We don't regret pointing out why Lewis-Beck and Campbell should be modest about the accomplishments of their models, but we do regret a superficial editing error that confused the names of their books. We also don't regret our criticism of forecasters' use of ad hoc variables, but we do regret citing Holbrook's model as an example. Holbrook used an extremism variable to analyze determinants of elections, not to forecast them. A better example would have been Campbell's May 1992 article, "Forecasting the Presidential Vote in the States" (in the American Journal of Political Science), in which he used ad hoc variables to fit his model to the data. Finally, we regret identifying Campbell with his previous employer, Louisiana State University (as his publisher's Web site does), rather than with his current employer, the University at Buffalo, SUNY.
None of these points affects the substance of the piece. The models have a very small sample size. Most of them use poll numbers of some kind (which have a built-in error) as variables. They make a dubious assumption that the statistical relationship between the economy and elections is not going to change in half a century. Campbell's belief that using out-of-sample "pseudo-forecasts" dispenses with the retrospective estimation issue misses the point. The
Campbell claims that "when it comes to reliably accurate forecasts of the national vote split, the hard record clearly indicates the greater accuracy of the models." Let's look at that record. Campbell's model was developed in 1990, meaning that he has used it to predict just two elections. Although he got within a percentage point in 1992, his 1996 prediction, made with the benefit of September polling data, was off by almost four points. What difference does four points make? A lot according to Campbell. In October 1996, he told the Buffalo News that, assuming this forecast was correct, the Democrats would likely recapture the House.
What the forecasting research can reasonably be said to demonstrate is that early horse-race poll numbers don't mean much. Voters start paying attention to the election around mid-summer, and when they do, the economy and overall satisfaction with the incumbent administration become important factors. We leave it to readers to judge whether the profundity of this observation justifies the publicity it has received. In any case, it doesn't justify feeding reporters half-baked, down-to-the-decimal-point predictions in May.
[The reply, click here.]
Subject: The Boss Ain't a Party Man
From: Eric Alterman, Author, It Ain't No Sin To Be Glad You're Alive: The Promise of Bruce Springsteen
Date: Fri Jun 9 12:02 p.m.
1) Has never evinced any interest whatever in electoral politics (save when Ronald Reagan tried to hijack him) and has not voted, according to his own recollection, since 1972
2) Has no interest whatever in the Clintons and has refused repeated entreaties to visit the White House during their term of office
3) Has never written a song about any political campaign
4) Has indeed written quite a few songs in the past inspired by real live political events, such as Three Mile Island and the L.A. riots
5) Does not live in New York, and probably does not care who its senator is
6) Premiered the song after Rudy dropped out, as Tim Noah admits.
Subject: Eminem—A Snuff Film for the Hip Set
Re: " Assessment: Eminem"
From: Benjamin Anderson
Date: Fri Jun 9 12:02 p.m.
Eminem does, indeed, have his charms, but I wouldn't call him "South Park with drum and bass." Parker and Stone's humor tends to be genuinely good-hearted, and finds its targets almost exclusively among the greater and lesser gods of pop culture. Eminem also takes swipes at his fellow entertainers, and these can be very funny. His violent rants against his wife, mother, and sister, however, constitute a different sort of beast, about as humorous as a snuff film. Watching South Park never makes me feel like I need a shower. Listening to Eminem makes me feel like I need to scrape off several layers of contaminated skin.
[To reply, click here.]
Subject:Love's Labour's Camp
Re: " Movies: Love's Labour's Lost"
From: Jeffrey Orlinski
Date: Sun Jun 11 12:31 a.m.
Kenneth Branagh's production of Love's Labour's Lost is actually the first time the play has been filmed. The reason: It's a lousy play. Just as Julie Taymor took what is essentially a giant bloodbath and adapted it to a mix of style and humor in Titus so has Branagh taken a weak and contrived plot about lovers and taken it to its logical, albeit silly, conclusion: The Hollywood Musical. At its heart, that is what it is. A director confronted with material to adapt that is 500 years old is better off capturing the tone and meaning of the play than to stifle the audience by trying to ram Elizabethan tights and every last "whither thou goest?" down their collective throats. Branagh's production is not definitive, but wasn't meant to be. I ask the reviewer to go see a production of Love's Labour's Lost (if he can find one) and then ask himself if Branagh really did the play all that much injustice.
[To reply, click here.]