A METHOD OF GRAPHICAL ANALYSIS 79 



Preliminary Analysis 



After the data have been collected and summarized in accordance 

 with the general plan of the study, the graphical phase of the analysis 

 begins with trial setups in which the dependent variable is plotted 

 against each of the independent variables in turn. Such charts are 

 intended only to give a general idea of the types of relationships and to 

 determine which of the factors tested are most closely related to the 

 dependent variable. Factors which do not vary with the dependent 

 variable are not necessarily to be discarded permanently since the 

 effect of one factor may obscure that of another. It is not to be ex- 

 pected that the data plotted on any of these charts will fall along 

 smooth curves. They will probably be widely scattered but in the 

 case of the more important factors a general trend is usually evident. 



On the next series of trial charts, several of the more important 

 factors are considered simultaneously. If a qualitative factor is 

 under consideration, separate charts are plotted for the different 

 classes. If these charts are essentially similar, the qualitative factor 

 may be disregarded for the time being and the data considered as a 

 whole. If, however, the qualitative factor appears to influence the 

 relationships in a logical manner the data must be sub-divided 

 and a number of practically independent studies carried on. In fact, 

 the analysis of the effect of a qualitative factor is intended to 

 determine whether or not the data forms an essentially homogeneous 

 whole. If there is a discontinous variable, it is often convenient to 

 hold it constant, i.e., a separate chart may be plotted for each value 

 or group of values of this factor. The factor, which from the pre- 

 liminary charts, seems most important is usually plotted against 

 the dependent variable. One or two other factors are coded. The 

 codes may be either in colors or symbols or both. The color codes are 

 usually the more easily distinguished and are, therefore, the better 

 for working charts. For final charts, however, color codes are not 

 usually practicable because of the difficulties of reproduction. Both 

 colors and symbols may be used when two coded factors are to be 

 tested simultaneously. 



In these preliminary sets of charts, it is well to test as many differ- 

 ent factors and combinations of factors as appear logically to 

 vary with the dependent variable. It is usually best, however, to 

 consider not more than three or four independent variables at a time, 

 one plotted against the dependent variable with one or possibly two 

 coded and one held constant on each chart. An attempt to hold 

 constant a greater number will often sub-classify the number of data 

 points so far as to obscure the real trends. Furthermore, the com- 



