KELIABILITY OF TRAFFIC MEASl'REMIONTS 3()3 



■I. Uiordiiu has shown' that the standard tlcviaticjn of the average 

 traffic load for any one period is given by 



~-'(^- 1 + «-"'') (3) 



\vli(M-(' a = the average soui'ce load 



/ = a\'erage liolding tunc per call 



T = lcn_ti;lh of ohsciA'ation period 



(Assumptions are as liefoi'c with an additional one that all periods are 



in statistical equilibrium) 

 When ? T « 1 this reduces to the form also given by F. W. Rabe^ 



or expressed in a per cent of the average 



I' (4) 



V, = 100 ^^ (a) 



Wlien A^ periods of length T are obser\'ed the coefficient of variation is 

 reduced further to: 



In the example of the previous section, 



A^ = 10 

 T = 1 

 t = 1/20 

 a = 5 



-' - /i^: = ^-% 



COMBINATION OF ERRORS 



Evidently if switch count readings are used to estimate the average 

 which may be expected in other periods, the two errors described above 

 should both be taken into account. The errors are probably correlated 

 but this correlation is weak and at present no method of allowing for it 



