177 
However, it is very difficult to understand synthetically this accordance of behav 
iour when observing the mean values in their totality. 
We think that if a systematic common behaviour exists, it should appear to have 
a certain resemblance of the mean curve, that is the curve parting the values M higher 
than their mean M m from the smaller ones. 
Independently from eventual overall asymmetries, this curve approximatively 
determines the form of the block characterizing the distribution of the mean errors. 
For the six above mentioned tests, we drew up the figures 32 and 33 which 
reproduce the behaviour of the mean curves according to the preceding definition. 
Test 13 Z 
♦ 
O' 
T 
+ 
+ 
- 
+ — 
+ 
r 
4- 
- 
\ 
o 
A + 
/ - 
4 
+ 
, 
- 
- 
A 
+ 
+ / 
- 
4 
- 
-\ 
- 
- 
* N 
Test 14 X 
Test 14 Y 
Test 14 Z 
Test 15 X 
Test 15 Y 
Fig. 33 
Test 15 Z 
For the first three tests we have only the curves for the errors in X and Y, while 
for the other three tests we have the curves for the errors in Z. In the spaces delimited 
by the curves, we have written + or — according to whether the means had a value 
greater or smaller than the value of the mean curve.