Key points are not available for this paper at this time.
Some twenty years ago, not long after the appearance in Sankhya of Sukhatme's tables of the 5% and 1% significance levels of the BehrensFisher test (since republished in Fisher and Yates' Statistical Tables 1963), Professor Snedecor asked me to supply an approximation to these significance levels for use in practical work by biologists who did not have access to Sukhatme's tables or who wanted probability levels, such as the 2% or 10%, for which no tables were available. By inspection of Sukhatme's tables I produced an empirical approximation, based on the Student t-table. In the interveninig period this approxinmation has found its way into a number of text-books, and I should like consequently to give some account of its accuracy.
William G. Cochran (Sun,) studied this question.