146 THE SOCIAL LIFE OF ANIMALS 



nesota. Forty-two hundred and five students were 

 observed in large classes, and 1,854 in small ones; 

 of these 1,288 were paired as to intelligence, sex 

 and scholarship before the experiment began. One of 

 each pair was assigned to a large and one to a small 

 class in the same subject taught by the same instruc- 

 tor. In this way the obvious variables were controlled 

 as well as is humanly possible, unless we could have 

 a large number of identical twins with which to 

 experiment. 



In 78 per cent of the experiments a more or less 

 decided advantage accrued to the paired students in 

 the large classes, and at every scholarship level tested, 

 the paired students in the large sections did better 

 work than their pairs in the smaller ones; the excel- 

 lent students appeared to profit somewhat more from 

 being in large classes than their less outstanding 

 fellows. 



Of the available data, a re-examination of the sum- 

 maries indicates that there is on the average a dif- 

 ference in the means in the final grade of 4.1 points, 

 favoring the students in the larger classes. There 

 is a statistical probability of matching this by 

 random sampling of four chances in ten million 

 (P = 0.0000004), and this despite the fact that the 

 majority of the class comparisons did not give signifi- 

 cant differences when considered alone. 



