Right Track/Wrong Track foolishness

Jay Cost, who was on the button in predicting the outcome of the 2004 election warns on Real Clear Politics that the pundits' reliance on right track/wrong tracks polling as predictors of the mid term outcome is misplaced:

Changes in the final Gallup "right track/wrong track" only anticipate 11.6% of changes in the President's party's share of the two—party vote between 1982 and 2004. They anticipate only 11.25% of changes in the President's party's share of House seats. Why is that the case? Here are some fun examples. Between 1990 and 1992, voter assessments of the state of the nation actually worsened by about 8%. However, the Republicans increased their share of the two—party House vote by 1.3%. Between 1996 and 1998, there was a dramatic 41% turnaround in "right track/wrong track." How much did this help the Democrats in Congress? They lost half a point. Between 1984 and 1986, voters started to feel nice and sunny about the state of the nation. Net positive ratings rose by about 15%. Did that help Reagan and the GOP? No. Their share of the two—party House vote dropped by 2.3% — and they lost the Senate.

Well, one might respond, perhaps "right track/wrong track" is just one of many factors that explain changes in votes or seats. When we combine it with the other factors, it will be salvaged. This seems to me to be extremely unlikely. The reason for this is that, in statistics, for every explanatory factor that we add, we need to "penalize" our coefficient of determination. We lose a "degree of freedom." This is done to prevent us from simply adding factors to explain little tiny bits of variation. Because of the penalty, such additions diminish, rather than enhance, the explanatory power of a model; the loss imposed by the penalty is greater than what the factor adds. "Right track/wrong track" explains so little variation that, if its explanatory power is greater than the penalty (and that is a big "if"), it is not much greater. In other words, it is quite unhelpful in predicting congressional election outcomes — by itself or in conjunction with other variables.

Why is right track/wrong track such a poor anticipator of changes in the balance of power? The reason is simple: this is not a parliamentary system. Our votes for Congress are not proxies for our evaluations of the state of the nation. They are, rather, proxies for our evaluations of the candidates that the parties present to us. Voters, quite obviously, do not have the state of the nation in mind when they cast their ballot for Congress — at least not in any way that implies that one party is punished and one party is rewarded.

This means that, this year, it is utter foolishness to pin your prediction of seat changes in the House upon "right track/wrong track." The worst flea market astrologer could do better than 11.6%.

Clarice Feldman   8 19 06

Jay Cost, who was on the button in predicting the outcome of the 2004 election warns on Real Clear Politics that the pundits' reliance on right track/wrong tracks polling as predictors of the mid term outcome is misplaced:

Changes in the final Gallup "right track/wrong track" only anticipate 11.6% of changes in the President's party's share of the two—party vote between 1982 and 2004. They anticipate only 11.25% of changes in the President's party's share of House seats. Why is that the case? Here are some fun examples. Between 1990 and 1992, voter assessments of the state of the nation actually worsened by about 8%. However, the Republicans increased their share of the two—party House vote by 1.3%. Between 1996 and 1998, there was a dramatic 41% turnaround in "right track/wrong track." How much did this help the Democrats in Congress? They lost half a point. Between 1984 and 1986, voters started to feel nice and sunny about the state of the nation. Net positive ratings rose by about 15%. Did that help Reagan and the GOP? No. Their share of the two—party House vote dropped by 2.3% — and they lost the Senate.

Well, one might respond, perhaps "right track/wrong track" is just one of many factors that explain changes in votes or seats. When we combine it with the other factors, it will be salvaged. This seems to me to be extremely unlikely. The reason for this is that, in statistics, for every explanatory factor that we add, we need to "penalize" our coefficient of determination. We lose a "degree of freedom." This is done to prevent us from simply adding factors to explain little tiny bits of variation. Because of the penalty, such additions diminish, rather than enhance, the explanatory power of a model; the loss imposed by the penalty is greater than what the factor adds. "Right track/wrong track" explains so little variation that, if its explanatory power is greater than the penalty (and that is a big "if"), it is not much greater. In other words, it is quite unhelpful in predicting congressional election outcomes — by itself or in conjunction with other variables.

Why is right track/wrong track such a poor anticipator of changes in the balance of power? The reason is simple: this is not a parliamentary system. Our votes for Congress are not proxies for our evaluations of the state of the nation. They are, rather, proxies for our evaluations of the candidates that the parties present to us. Voters, quite obviously, do not have the state of the nation in mind when they cast their ballot for Congress — at least not in any way that implies that one party is punished and one party is rewarded.

This means that, this year, it is utter foolishness to pin your prediction of seat changes in the House upon "right track/wrong track." The worst flea market astrologer could do better than 11.6%.

Clarice Feldman   8 19 06