IN [1], I PROPOSED a model to interpret the behavior of beta coefficients over time. Without going into every detail, that model hypothesized that the beta coefficients of a set of securities at two points in time could be interpreted as drawings from a multivariate normal distribution. The beta coefficients were defined as strictly stationary between two points in time if (a) the marginal distributions of the beta coefficients in each cross section had the same mean and standard deviation and (b) the correlation coefficient between the betas in the two cross sections was 1.0.1 The article [1] itself contained evidence that beta coefficients were not strictly stationary over time and tended to regress towards 1.0 over time.2 Haltiner, Elgers, and Hawthorne [2] have hypothesized that one of the reasons for this regression tendency is that the cross-sectional standard deviation of the beta coefficients for a given set of firms may decline over time. They point out that could cause such a decline even if the cross-sectional standard deviation of the betas of all securities in existence at each point in time were constant due to a "survivorship" bias. Although this type of phenomenon would not create any post-selection bias in the tests conducted in [1], which were carefully formulated to avoid such a bias, it would have real economic meaning to an investor who wished to select a portfolio of extreme risk, either high or low, on the basis of historical data. Such an investor would find subsequently that his portfolio would have a less extreme risk than originally thought. This result would follow even if the correlation coefficient between the historical betas and the subsequent betas of those companies which survived were 1.0. This hypothesis deserves to be tested, and Haltiner et al. do examine it over the twenty-one years ending in June 1975. After removing the effect of measurement error, they find some support for it; but their results are not strong (significance at the 5 percent level in one case and at the 10 percent level in two other cases3). To provide more extensive evidence, I have estimated, after removing the effect of measurement error, the cross-sectional standard deviations of the true beta coefficients for NYSE stocks in successive seven-year periods from 1926 through 1975 as a function of the length of prior listing on the NYSE: