defined from Förstner (1983) 
  
Especially for onedimensional alternative hypothesis we get 
vos; = 09.60 7 /n'PQ. Ph; . 
Having a gross error, we obtain for diagonal weight matrix P 
a lower boundary value: 
Vol, = oiu // P1 4 (20) 
Only a gross error vl,> v,1,can be found by the test with a 
probability g > gg. 
  
  
  
  
  
  
  
Now as a measure of internal reliability for two multidimen-: 
sional alternatives we obtain the lower Bounds of two distin- 
guishable model errors (detail derivation see Li,1985); 
Vas 3 = 3j:90°80° ky, // Si(Pss)s; = Vas, ko, (21) 
2984; represents a minimum parameter vector vs, in direction s 
  
  
  
  
  
i 
which can be Statistically detected with a test power not smals 
  
ler than 8j and can be Separated from the parameter vector vs, 
in direction s,with à probability not smaller than í(l-y$). Here 
kp, jis defined as a separability multiplying factor and it ig 
à function of statistical parameter 239,80 ,v0 and correlation 
coefficient Pij : 
= k SO YO 2a). 
UF (ag » B80 » Yo Py) (22) 
The values of koij are calculated according to different corre- 
lation coefficients and listed in Tab.2. 
1 
If the value k, is taken according to the maximum correlation 
of two alternatives, the corresponding lower bound is called 
The maximum lower bound of Separability. 
Especially for two gross errors in uncorrelated observations 
  
we get a lower boundary value of loc@bility: 
  
UT = 93 30 koi; / Jr, = Vol;ko;, sa (23)