RELIABILITY OF HOLDL\G TIME MEASUREMENTS 379 



and 



(Tv 



e 2< 



\ — C I 



(9) 



The frequency distribution of the sum of a' such discrete variables v is 

 readily found to be 



/(iV.) 



(' - ."0 



/W — 1 IVy, 



2 4 6 8 10 12 14 16 18 20 22 24 26 



W = NUMBER OF CALLS EXTENDING BEYOND NO. 60 SWITCH COUNT 



Fig. 9 — Loss of true switch counts at the end of the observation period 



The parameters of this sum distribution are 



.s.f = \vv 



we *■ 



1 - 



Ow = s/'WOv = \/i 



e 2i 



1 - e~ 



(10) 



(11) 



A check on these last formulas is shown in Fig. 9 where the numbers of 

 switch counts lost on w calls carried beyond the last (or r -\- 1st) switch 

 count are recorded for 30 busy hour periods. Two of the hours show results 

 outside the theoretical ±3 a normal curve limits but the falling of almost 



