22 



EMPIRICAL STUDIES OF MEASUREMENT 



Age of Husbnd. 

 11-11 M M-3/eK. 



I I 



' I I 



I 



' I 



I I 



I i 



I I 



I I 



I I 



TABLE X. 



Fig. 6 gives the regression of y (wife's age) on x (husband's 

 age) in terms of averages of arrays of y and also of medians of 

 arrays of y. To give the regression by single modes for the arrays 

 would be fallacious, for each array is more or less clearly a bimodal 

 distribution. This is shown in Fig. 7, where the s/'s are grouped 

 in four large arrays. It should be clear that any single figure is 

 inadequate to express this relationship. The Pearson Coefficient of 

 correlation is .20 and the regression of y (wife's age) on x (hus- 

 band's age) calculated from it is .25. But this would lead one far 

 astray concerning the real regression, as we see by Fig. 6. The 

 relationship is closer for early deaths than for late. The form of 

 distribution of the relationship is, apart from this, skewed in gen- 

 eral from a mode of close resemblance toward very great diversity, 

 and is in the third place complicated by the submodal tendency of 

 a wife to die at about 35 more often than at 30 or 40. Jguch a case 

 illustrates the fact t.ha.f. panTi typp nf measure of a relationship meas- 



ures some particular aspect thereof and also the fact of the extreme 

 {jbstractness from realityjjf the Pearson Coefficient, which in this 



