78 BELL SYSTEM TECHNICAL JOURNAL 



other hand, there are many problems to which these methods may 

 properly be applied and in which it is practically impossible to derive 

 a reliable and satisfactory basis for making estimates without some 

 such methods of analysis. Certain economic problems, in particular, 

 because of the impracticability of experimentation and because the 

 complex reactions of a group of individuals are involved, are not 

 adapted to solution by the statistical methods which have proved 

 useful in biometric sciences, but may be dealt with by graphical 

 methods. This has been found particularly true in problems in- 

 volving local telephone message use, and throughout the following 

 discussion, illustrations are drawn from analyses of this type. 



Data 



Since the ultimate aim of a graphical analysis of this type is to 

 provide a basis for making estimates, the first step is to determine 

 the estimates which will be required and the type of cases and condi- 

 tions under which they will be used. In this way the aim and scope 

 of the analysis is clearly defined. The unknown factor (the dependent 

 variable) is to be estimated from certain known factors (independent 

 variables). Various factors, quantitative and qualitative, which 

 might logically appear to be indices of conditions controlling the 

 dependent variable are, therefore, considered. 1 Only factors as to 

 which data are available at the time and place where estimates are to 

 be made are useful as independent variables. It is usually advisable 

 to test a suggested factor by means of any data, even in small amounts, 

 which may be available before a complete body of data is collected. 

 Such preliminary investigations are useful in indicating the scope and 

 detail in which data should be secured. In general the data should: 



1. Be adequately representative of the type of cases for which 

 estimates must be made, 



2. Be adequate from a sampling standpoint for each situation, 



3. Be as nearly homogeneous as practicable, i.e., cases having any 

 outstanding peculiarities should be excluded, 2 



4. Include what appear to be the important factors or indices for 

 each case. 



1 It should be noted that such relationships need not be those of cause and effect. 

 If two factors vary together (as do, for instance, different effects of a common cause) 

 the values of the one which are hard to determine can be estimated from the more 

 easily measured values of the other. 



2 For instance, if estimates are to be made for small exchanges, it would not be 

 advisable to include data from large exchanges in the analysis. 



