"Vs 
Wiz mn : | (23) 
tetes RI 
9 NAvivi 
The acceptance interval for Ws is then 
1 1 
MFR 30319 )é Se E qd22^91,0)25; | (24) 
0 1 0 
where a, is the type I error size for the test criterion Ws. 
The significance level a for the test of individual residuals (23) is dependent 
on the significance level ao of the global test (13). Baarda harmonizes a with a, 
by putting 8 (3p) = 8(31), i.e. by using equal type II error sizes for the global 
test and the individual test. For the relations between a and a, see the nomograms 
in /3/, pp. 21-23. As far as the author could understand this, the harmonizing 
between o and a, Was based on the assumption that the single events wj are inde- 
pendent on each other. Hence a mostly reliable system is required for an effec- 
tive application of the data-snooping technique; otherwise the danger of false 
inference is too great. At least we have to exclude those observations from the 
test procedure - or better: we have to change the network arrangement in a way 
that those observations don't occur - which are correlating too much with others. 
This is valid especially for "spur" observations, i.e. for observations which 
lead to Ov = 0 (zero variance problem), since in the extreme case of spur obser- 
vations the test criterion Wi= $ is not defined. 
In. (21) yi UU was denoted as an individua] reliability indicator. The amount for 
the computation of all individual Nj = Pi9vjv; (if P is diagonal and if only one 
gross error is assumed) may sometimes prevent of individual computations. However 
we are able to make some statements on the global reliability of a system (see 
FPorstnep /7/,0Gȟm /11/, Pope /17/). 
In Grün /11/ the average diagonal term of Qyy was proposed to serve as global. 
reliability indicator: b 
tr(Q 
RI(T) = in) == : (25a) 
2tr (Ql) 2tr (QJ) 
RI(x')s 77 9 RI(S) * — 7 ‘ (25b) V 
r = redundancy, n'= number of observations 
Of course global reliability indicators can be constructed which do more refer to 
the "external" reliability (Baarda /3/), that means to the effect of gross errors 
upon the final product of the adjustment - the object point coordinates. Those 
would correspond even more with the following accuracy indicators. But here the 
term "reliability" should be stronger connected with the detection than with the 
effect of gross errors. tr(Q, 
An analogous global accuracy indicator would be —A Te yet we use the average 
standard deviations of the point coordinates (with gi" 1): 
tr (ax) 
AI(T) = ca , k = number of points, including (26a) 
the control points 
tr (fax, 
AI(X) sr aol