This paper shows that more reliable data can improve the estimation of a relationship. Reliability of the measure of economic well-being was improved in a number of ways. The authors then examined the effect on the estimate of the relationships to three variables: evaluations of the nationís economy, evaluations of Ronald Reaganís performance as president, and the vote choice between Reagan and Mondale. Here are some of the findings:
The authorís claim that the use of five rather than three categories for the economic well-being question produced Ďstartlingí results; the estimated relationships were almost twice as large. Most significant for the forecasters, however, is the finding that when these different measures were used to develop a predictive model, they yielded negligible improvements in accuracy (when tested in the calibration samples). This is my conclusion from the standard errors reported in the paper; Rosenstone told me that he disagrees. No examination was made of the effect of these alternative measures on out-of-sample forecasts.