Table 7. Test of the hypothesis that constant errors within single models should be random errors
within a set of models, measured by different participators
v = degrees of freedom. L = level of significance in ?/o.
Coordi and Strip I Strip II Strip III
dist. v t L V t L v t L
X 15 +0.76 — 21 --1.04 — 13 — 0.06
Y 15 — 1.08 — 21 +691 01; 13 181 —
Z 15 —2.24 5 20 — 1.67 — 11 + 0.19
LA+B 7 +122 — 10 +455 1 5 +018 —
La—B 7 —1.06 — 10 —+ 1.48 - 5 +294 5
It must be concluded, that no constant errors could be indicated within the set of models measured by
different participators.
6.24 So the standard deviations of the »constant errors» (within set of models) could be calculated as
weighted standard deviations of M (= the means of the total errors). Tables 1: I—III. They were put to-
gether in the following table.
Tabl. 8. Standard deviations in mm of the means of the total errors within set of models.
Strip X | Y Z | L
I 26 | 52 120 | 52
II 54 150 154 63
III 58 | 109 156 | 39 |
6.3 The systematic errors.
6.80 The analysis of regression might be considered somewhat uncertain because of the low number of
checkpoints in every model (58—95 points per model) and of the calculation with only »single degree of
accuracy in the computor» (= 10—8 sedecimal digils). A result of the analysis was the values of S, and
r?,. No acceptable function could be obtained as to Lí.pg and L4 jy, which could imply that no syste-
matic errors occurred for the distances.
6.81 T'he systematic errors within single models were tested with the aid of the multiple correlation coeffi-
cient, The results were given in the following table.
Table. 9. The number of models with systematic errors, tested with the multiple correlation coefficient
at different levels of significance.
The number of models | |
. — | Average number
Coor- RON ; RS :
Strip dinate Not sig- _ Level of significance in 9$ Total of checkpoints
nificant 5 | 95 | 1 0.5 € per model |
; ET |
X l 0 | 1 1 13 16 |
I Y 2 0 l ] 12 16 58 |
Z 0 1 | l 0 14 16 |
i | x : |
X 0 oa! 2 0 20 22 |
I Y 0 031: 9 LL | A | 22 68 |
Z 2 0 | 3 1 15 21] |
| is A eiie
X 0 0 | 1 0 17 18 |
III Y 0 0 0 0 18 I8 ..| 95 |
Z 0 |} Q0 0 I 15 16 | |
The conclusion must be, that systematic errors occurred in models.
6.32 The question wether there were systematic errors also within a set of models was tested with a run-test
(Dixon & Massey — Introduction to Statistical Analysis. Page 254) on the signs of the regression coeffi-
cients. The nullhypothesis was that all the coefficients within a strip were assumed to be random between
models. The values of the testfunction z normal (0.1) and the corresponding levels of significance in per
cent were given in the following table.
14
The
How
6.33
s
6.4 7
6.40
error
6.41
was
tistic
the f
In th
tion.
Ther
fican
rence
6.5 |
6.50
rand
mati
pare
view
tice.