Rows (1) and (2) of Table 3 report estimates of regressions in which the propensity to pay does not control for risk. In these specifications, the dividend premium is 1 % significant and the regression R2 is as high as 30%. These results are similar to those reported in BW. In Rows (3) and (4), the propensity to pay is adjusted for risk in the first-stage logit model. The dividend premium is no longer significant and the regression R2 is essentially zero. Thus, controlling the propensity to pay for risk eliminates the explanatory power of dividend premium. This result is obtained regardless of whether we control for M/B in the stage 1 logit. Panel B reports similar results for the B W sample period from 1963 to 2000.