A METHOD OF GRAPHICAL ANALYSIS 81 



When several different sets of charts such as are described above 

 have been scrutinized, definite trends will usually be fairly clearly 

 established. It will often be found that while these trends are well 

 defined, nevertheless a number of points may scatter widely. Such 

 points are studied carefully. If, after the original data are checked, 

 the points are found to be correctly plotted, each case is investigated 

 in detail to account for the observed divergence. Sometimes it will 

 be found due to a factor which has not been taken into account, the 

 inclusion of which will often improve the results of the study as a 

 whole. On the other hand, peculiar local conditions or history may 

 give rise to such divergence. These cases are not really a part of the 

 similar group under consideration. If they are sufficient in number 

 and similar with respect to each other they may be studied inde- 

 pendently. If not, they are either excluded entirely or given slight 

 weight in the general study. Because of wide differences in problems 

 and material, it is not practicable to describe in detail the process of 

 analyzing such preliminary charts in arriving at decisions as to data 

 and process. 



Curve Drawing 



The next step is the construction of curves through these data 

 which will truly represent the relationships involved. This can be 

 facilitated by plotting the average values of the dependent variable 

 for all cases having the same values (within certain limits) for all the 

 independent variables. 



On Fig. 2 the data points are the same as those plotted on 

 Fig. 1. The closed symbols which have been added are average 

 points representing the data points of the same symbol. The 

 abscissa of each average point is the mid-point of an interval of 

 rank of service (0-20, 10-30, etc.) and the ordinate is the average 

 of the message uses of all the points falling within that interval. 



The average most often used on such charts is the median not only 

 because it is most easily located but because it is usually the most 

 representative, giving little weight to extreme cases. Whatever 

 average is used, it is well to make it a moving average, i.e., covering 

 overlapping intervals such as 20-30, 25-35, 30-40, etc., rather than 

 20-30, 30-40, 40-50, etc. 



These averages serve as a guide for drawing the preliminary curves 

 through the data but the actual data points are considered at the 



