A FIRST EXPERIMENT 3 1 could be shown, the tests would be of little worth to the employment office. Therefore, it was decided to take as the basis of each girl’s standing, an average of her work for four weeks. These averages were obtained by making a detailed statement, drawn from the separate daily pro duction slips of each girl, showing the number of pounds done for every day in the week, together with the exact number of hours taken to do them. The total number of pounds inspected by a girl, divided by the total number of hours worked, gave the average number of pounds per hour for the particular girl and became the basis for her ranking. In making out these statements, all work other than that on a single kind of shell was discarded. This was done for the sake of uniformity, it being manifestly unfair to judge the relative speed of different girls on a basis of pounds when one girl was inspecting large shells which went very quickly while another was inspecting small shells which went very slowly. After the average hourly production of each girl for a period of four weeks had been determined, the results were compared with the performance of each girl in each of the tests. This was done to obtain the degree of correspond ence or the correlation, as it is technically called, between the tests and the actual production. Now the method by which the correlation between the performance of the girls in the tests and their rate of production was deter mined was not by guesswork or by rough observation, but by an exact statistical process. This process is very simple a ud can easily be described. Let us suppose that girls A, B, C, D, E, F, G, H, I, and J are ten girls who have been tested (see table on page 32). After the tests have been given, it is necessary to rank them in the order of their ability; that is, the girl who turns out the most work