186 



THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1956 



Table IV — Average Experiment Time and Probability of a 



Correct Selection — Procedure Type R3 



(a* = 2, P* = 0.800, ^ = 0) 



(Multiply each average time entry by d^) 



For example, we might choose a* = 2 and P* = 0.800. For procedures 

 of type R3 the probability of a correct selection is again completely in- 

 dependent of n; here it depends only on the true value of the ratio a. 

 The average experiment time depends strongly on n and only to a limited 

 extent on the true value of the ratio a. Table IV gives these quantities 

 for a = 1, a = 2, and a = 00 for the particular specification a* = 2, 

 p* = 0.800 and for the particular value ^ = 0. 



efficiency 



We are now in a position to compare the efficiency of two different 

 types of procedures using the same value of n. The efficiency of Ri rela- 

 tive to R2 is the reciprocal of the ratio of their average experiment time. 

 This is given in Table V for a* = 2, P* = 0.800, r = 4 and n = 4, 10, 20 

 and 00 . By Table I the value P* = 0.800 is not attained for n < 4. 



In comparing the sequential and the nonsequential procedures it was 

 found that the slight excesses in the last column of Table I over 0.800 



Table V — Efficiency of Type Ri Relative to 



Type R2 

 {a* = 2, P* = 0.800, r = 4:,g = 0) 



{ 



I 



