GROSS EARNINGS IN DISTRICT I 273 
occupied in 1924? There are various ways of securing answers 
to these questions, and because of the importance of the problem 
each of them is set out in detail. 
Let us (1) take as standards of reference the average of the 
ratios for the 410 banks in each of the years 1924 and 1925; 
(2) compute the percentage amount and direction by which 
the ratio for each bank each year deviates from the yearly 
average for all banks; (3) determine suitable groups for the 
deviations above and below the averages for the respective 
years; (4) by signs and by percentage amounts, distribute the 
deviations each year for the several banks into the groups 
provided for in (3) above; (5) for each percentage group, 
determine for the first year the arithmetic mean deviation, and 
for the second year the arithmetic mean net deviation, the 
latter amount being secured by adding the amounts of the 
deviations algebraically and by dividing by the number of devia- 
tions; and (6) for each of the groups and for those above and 
below the averages for the respective years, compare the amounts 
as secured in (4). Then the differences will represent the amounts 
of regression to type as measured by differences in mean disper- 
sion—minus (=) amounts for ratios above the average, and 
plus (+) amounts for those below the average in the first year 
indicating regression toward the average in the second year. 
That is, measures of change, in terms of dispersion, are secured 
for ratios which are differently placed, plus and minus, relative 
to the average levels for all banks in each of the years. 
When this process is carried through it is found, as shown 
in Table 157, that the net result for ratios which were high 
in 1924 (relative to the average in that year) is a decrease, 
and for those which were low in 1924 (relative to the average 
in that year) it is an increase, the changes being determined by 
the differences in the amounts of dispersion in the first and in 
the second years with respect to the averages in these years. 
Ratios which are high or low in the first year tend to be high 
or low in the second year, but the net dispersion is less the 
second year. That is, regression to type varies inversely with 
the signs but directly with the percentage amounts of 
dispersion. 
It is apparent, therefore, that the unmistakable tendency for