now predicted correctly with respect to the win for Carter. It is not clear, however, that much confidence should be placed on this result. Each election has special features, and without too much work one could rig the equation to predict most of the observations perfectly within sample. Third, the editor3 has expressed concern about the number of coefficient estimates in, say, equation 6 that are insignificant by conventional standards. Given the small number of observations, this is not necessarily surprising, and my general procedure is to retain insignificant vafiables if their coefficient estimates are of the right sign. I would rather risk the loss of some efficiency by incorrectly including a variable than risk the loss of consistency by incorrectly excluding it. In the present case there is no question that the likelihood function is fairly flat in the vicinity of the estimate of X, although the size of the estimate seems reasonable. The only observations that really matter for the estimate of X are those in which at least one of the candidates has run before, and so the number of observations involved in estimating A is very small. There is also considerable collinearity among the estimates of A and the coefficients of DPERt and I. (The coefficient of 1, is merely the constant term in the definition of the measure of performance (see Fair (1978, p. 164)), and it is not very important.) There is also some collinearity between the time trend and the inflation variable. Equation 9 is the same as equation 6 except that the time trend, I, and X have been excluded. The main change is that the estimate of the coefficient of DPER, has increased from 0.0493 to 0.0637. The results are otherwise quite similar. The 0.0637 estimate of the coefficient of DPER, in equation 9 seems somewhat high to me, and I am inclined to stay with equation 6 even though there are insignificant estimates. Tastes differ on this, however, and there is clearly no right answer. As a final exercise, equation 6 was used to predict the 1984 election for alternative values of g*, and P2,j. The results are in table 2. For an inflation rate of, say, 8.0%, the break-even point (0.5) corresponds to a growth rate of about 0.0%. One standard deviation (0.0352) below 0.5 corresponds to a growth rate of slightly more than 3.0% for an inflation rate of 8.0%. If Reagan does not run, the Republicans do not have the person advantage, and the break-even points are less fivunrnhk1 for the-m