Running Regressions for the New York Times

David writes:

When I read a New York Times article (or any newspaper or magazine for that matter) based on a survey they conducted, I always lament the fact that they only offer simple crosstabulations in their analysis. For example, the article will discuss whether primary and caucus voters weigh issues versus electability more when selecting a candidate (see previous post), yet they only provide the percentage of voters who say issues and electability are important, respectively. Wouldn’t it be nice if we could just put both (scaled) predictors in the same vote choice model to see which has a larger impact on vote choice? Since I’m sharing an office with Andrew Gelman, while working on our Red State Blue State Paradox book, I mentioned to him that we should offer our services, free of charge, to the New York Times to run regressions and write up a short article that a general NY Times reader (without any statistical knowledge) could understand. The short article could be available just online at the NYTimes.com. Bill Keller, any thoughts?

Seems like a good idea for me. This woul also work with Gallup etc. These polling organizations have research departments, but I don’t see these sort of in-depth reports coming out; I’m not sure why. Ideally they’d hire some Ph.D. in quantitative political science full time to do this, along with someone like me as a consultant to keep an eye on things. Polls themselves are fairly expensive so this doesn’t seem like such an outlandish idea to me, not that I know anything about the business end of anything,

2 thoughts on “Running Regressions for the New York Times

  1. Someone recently suggested there may be economic interests that discourage polling organizations from doing things that may make them seem exchangeable ( i.e. that one could for instance pool over similar polls done by different polling organizations )

    Very likely they do not perceive this type of work as being in their interests to promote

    I would "hope" a more educated customer base would be able to change that

    Keith

  2. You see this in several areas. A friend who has worked with bankers said they only ever want to know simple bivariate relations. They are not interested in more complicated multivariate models. This is probably partially due to the difficulty of explaining these models well, and the fact that people have a hard time thinking in multivariate terms. Or maybe the reality is, more complicated models aren't useful for their practical decisions. All else in rarely held constant.

Comments are closed.