478 
Table 4. Investigation 2. F-Test of Differences of Setting Precision. 
Operator 
no. 
F 
Test 
P(F) 
10 log P(F) 
1 
0.96 
not sign 
0.65 
0.81 1 
2 
1.42 
almost sign 
0.025 
0.40 - 2 
3 
0.97 
not sign 
0.60 
0.78- 1 
4 
1.00 
not sign 
0.50 
0.70 - 1 
4 
0.73 
sign 
0.993 
0.00 
According to Fisher’s omnibus test : 
rViJ2 — 4 60s • ( ? — 10 f\A 
- 2.31 
Chi 2 - Chi 2 38 Vo (10) 
model, the correlation coefficient has been computed for every coordi 
nate in investigation 1. The pair of values refer to the same time. 1 he 
correlation coefficients are found in table 3. They show a great dis 
persion and have different signs. A positive value indicates a lower 
precision when the blood alcohol concentration is high, while a negative 
value indicates a high precision under the same condition. 
As the precision shows a great variation in investigation 1, further 
tests are difficult to apply. 
Influence Upon Precision the Following Day 
The precision is uniform in investigation 2 and a pooled value of 
the standard deviation for each operator was computed every day. A 
two-sided F-test (see |1] 14.2) has been used to investigate whether 
the standard deviations are significantly different if alcohol has or 
has not been taken the previous evening. The significance levels are 
the same as in the earlier tests. F is defined as the ratio between the 
variance influenced by alcohol and the variance without this influence. 
The degrees of freedom are 360 for both numerator and denominator, 
except for operator 2 where the degrees of freedom for the numerator 
is 72. (He became ill with appendicitis.) The results of computations 
and tests can be seen in table 4. With the aid of Fisher’s omnibus test 
(see [1] 15.6) these values are combined to a single value, which is 
not significant i. e. roughly speaking: the alcohol taken had no influ 
ence upon the setting precision the following day.