84 



BELL SYSTEM TECHNICAL JOURNAL 



The smoothing process must, therefore, be repeated back and forth a 

 number of times until both sets of curves appear to be smooth, reason- 

 able and consistent families. During this process, it is important 

 that the various families of curves be tested against the original data. 

 If this is not done, it may happen that a series of small changes will 

 accumulate in such a way as to bring portions of the curves outside 

 the limits of the original data. Furthermore, the factor or factors 

 held constant on each chart must not be lost sight of. These factors 

 should be plotted against the dependent variable holding constant 



CODE 



PRELIMINARY CURVES FROM FIG. 2 



REVISED CURVES FOR 90 ALLOWANCE FROM FIG.3-B 



W 

 (0 



u 



< 

 (0 

 (0 



LU 



2 



X 



h 

 z 

 o 



u 

 < 



u 

 > 

 < 



10 20 30 40 50 60 70 



AVERAGE RANK OF SERVICE 



Fig. 3A 



all other factors (values being read from the smoothed curves) to see 

 that these relationships also are being made reasonable, consistent 

 and smooth. 



The process of smoothing described above is a long and laborious 

 one involving at every step the making of special decisions based 

 upon knowledge of the data and the logic of the situation with regard 

 to the particular problem. Various methods of facilitating the work 

 have, however, been devised some of which are described below. 

 Figs. 3A and 3B illustrate the advantage of having both sets of curves 

 on the same chart with the same scale for the dependent variable so 



