A note on the acceptability of regression solutions: another look at computational accuracy

This note examines the experiment performed by Beaton, Rubin, and Barone (1976) to study the effect of rounding errors in published figures, when these data are used in regression analysis. The experiment could be vitiated by the fact that the error introduced in the trend variable is by no means trivial when one measures the data in deviation from the mean. For this reason, the results presented in Beaton, Rubin, and Barone (1976) do not contain enough evidence to suggest "that it is extremely unlikely that the unperturbed solution" (p. 161) of the Longley model is the correct one.Publicad