Posts
Every party would like to find a way to find the perfect candidate to run for office. In some way, this is the goal of primaries in the United States. But in most states, this only determines the preferred candidate among sympathizers, but the most electable in the general population may be different. Any other criteria we could use?
Scott Armstrong and Andreas Graefe looked at detailed biographies of US presidential candidates and claim to have found the formula that works 25 times out of 28. You can start now looking for the perfect candidate: coming from a political family, first-born, single-child, lost a parent in childhood, is still married with children, some adopted, went to a military academy, then received a graduate degree from an Ivy League school, is a member of Phi Beta Kappa, held political office (the more the better) and was never defeated in an election, has written books, was a movie or sports celebrity, has military experience, survived a major disease, is tall and heavy, has common first and last names, is attractive and looks competent, comes from a large state and is affiliated with a large region. All in all, the paper mentions 34 criteria. My score is 15, so I am afraid politics is not for me.
The unpredictable ones? Truman, Carter and Clinton (first term).
Monday, September 14, 2009
Subscribe to:
Post Comments (Atom)
2 comments:
The method used here gives equal weights to 34 criteria. I am puzzled that it would outperform methods that are optimized with flexible weights (a ... regression). How can that be?
Regression is useful to fit a model to training data (i.e. explaining existing data). That is, the fit of a model (i.e. estimating "optimal weights") will be better than equal weighting. However, "optimal weights" also incorporate noise and thus do not generalize well when predicting new data. For further reading, I recommend "Simple heuristics that make us smart" by Gigerenzer et al. (1999).
Post a Comment