Statistirnl fHrentiffntionit 



:«»! 



the big man might be more heavily marker] than the small man, although 

 his combined Bysteni was far lens efficient. To throw light on these matters 

 Galton coiiHtructs "mixed" correlation tables in which he ijives the rank of 

 one variate and the deviate of the other and draws contour lines for the first. 

 Thus for each value of stature — i.e. for every two inches — he constructs a 

 scale of deviate of vit.ii capacity to each graoo. Ho then combines these to 

 give a single dijignini and joins the points of wjual vital capacity. He thus 

 obtains wnat he terms isograms at every 20 cubic inches of vital capacity. 

 He gives a similar system of isograms for strength of stjueeze and weight. 

 We rejiroduco one of these diaixranis. 



Isograms of Stature and Vital Capacitj-. 



Galton remarks that these and similar methods indicate how marks may 

 be determined for physical mejisiirements. It is, perhaps, needless to say 

 that they would apply also to mental measurements, and even to examina- 

 tional results. The ordinary process of marking by adding up the marks 

 acquired by the same boys in a variety of subjects is really a fallacious one, 

 as there is usually a high correlation between success in different subjects, and 

 the true difierence between two individuals cjinnot be measured by adding 

 up a series of differences, highly correlated. We might as well repeat papers 

 in one subject until the first of two boys had nuiltiplied his excess marks a 

 hundred times, but this deviate difference woiUd certainly not be the measure 

 of their relative intelligence. Their true ranking would have to be obtained 

 from the multiple correlation surface, allowing correlation between subjects. 

 The theory of examination marking sadly needs scientific treatment'. Galton 

 himself concludes as follows his discussion of marking: 



' Ranking in multiple marking can be thrown back, through multiple correlation, on the 

 r and X of the 'Goodness of Fit' tables. 



