record should have been between these two scale lines. As actually 
measured it was between the two scale lines for 8.03 minutes. This 
is a discrepancy of about 6% between the theoretical and observed 
values. 
Table 18 shows the other values as computed from the theory and 
as checked by measurement. The greatest error in minutes is 0.45 
minutes between the predicted and observed values. Thus the error 
in prediction is only about 2% of the total pressure record length. 
For the greater departures from the mean, the percentage errors are 
larger, but the whole table shows remarkable agreement between pre- 
dicted and observed values. The last row, for example, predicts that 
the record will be more than three positive scale divisions from the 
mean for about eight seconds out of twenty-five minutes and that the 
record will be more than three negative scale divisions from the mean 
for another eight seconds. Actually the record never went below 
three scale divisions and it was above three scale divisions for 
ten seconds. 
What has just been done should be reemphasized. Points were 
taken at random from a pressure record. The standard deviation of 
these points in terms of scale units on the paper was then computed. 
Then the total time that the record would occupy a certain range of 
values was computed on the basis of the fact that the record was 
Gaussian. The predicted and observed values were found to agree 
remarkably well out to 2.5 standard deviations of the distribution. 
Usually statisticians are well pleased if an observed distribution 
fits a normal curve two standard deviations away even crudely, and 
in this case the agreement is even good 2.5 standard deviations away. 
B2 
