820 BELL SYSTEM TECHNICAL JOURNAL 



factory and is the instrument ordinarily used for this work. It has 

 the advantage over some of the other instruments that were tried, 

 of being in much more general use on speech circuits. 



Control of other haphazard factors of a more or less psychological 

 character, may best be obtained by taking enough data so as to average 

 out their effects. This involves the number of syllables that are 

 called by each speaker and the number of caller-observer pairs that 

 are used in the test. The variability of caller-observer pairs for a 

 calling unit of 66 syllables may be seen from Table III. The probable 

 error ^ in percentage articulation of a single observation (es) i.e., 

 one caller-observer pair as taken from the data in the table, is ± 9. 

 The probable error of the average articulation (eav.) of the 56 caller- 

 observer pairs is ± 1.2. 



It has been found from a large number of tests that the probable 

 error of a number of crews, each consisting of one caller and one 

 observer, is of an order of ± 12 (per cent articulation) for a 66 syllable 

 unit when the syllable articulation is around 50 per cent. This value 

 tends to decrease with increasing experience in testing, and with 

 increasing or decreasing values of syllable articulation. The use of 

 36 caller observer pairs obviously reduces the probable error to an 

 order of ± 2 in percentage articulation, which is about the order of 

 magnitude of the errors involved in maintaining the testing personnel 

 over a period of time. 



Since as will be shown in a later paragraph, the syllable articulation 

 is equal to the cube of the sound articulation, the probable error in 

 the sound articulation ® for one caller and one observer, or a unit of 

 198 sounds, is of an order of ± 6 when its value is around 80 per cent. 

 Since each individual sound is called only six times, the probable 

 error for each individual sound for a single caller-observer pair is of 



an order of \\ X 6 = ± 35. If a test comprises 4 observers per 



\ 6 



caller and 8 callers, each sound is called 192 times, which reduces 

 the probable error for the articulation of each sound to dr 6. Under 



^ Es = .67 ■ — ^and eav = -= ; where, n = number of caller-observer pairs; d = 

 n — I ' -^n 



difference between the articulation of a caller-observer pair and the average arti- 

 culation of n caller-observer pairs. 



^ Es = ^ ei = 3L^eL 



8, = ± 12, 5 = .5, L = S'i\ 



El = j^, = ± 6. 



Cs = prob. error in syl. art. for one caller-observer pair. 

 Ez, = prob. error in sound art. for one caller-observer pair. 



