SEMAINE D'ÉTUDE SUR LE ROLE DE L ANALYSE ECONOMETRIQUE ETC. 
40% 
expense of doing violence to the model. There are now several 
techniques for dealing with such problems, and the fact that they 
may not be fully satisfactory does not justify the use of a wholly 
inappropriate technique in a context which is really simultaneous 
KOOPMANS 
If I understand Prof. FISHER correctly, use was made of the dat: 
two or three times in order to select the set of predetermined va 
riables that is drawn on for help in estimating a particular equation. 
Should one worry about the effect of repeated use of the same data 
and if so what kind of errors or lack of efficiency could have beer 
produced thereby? 
FISHER 
Professor KoopMaNs wishes to know whether my recommenda: 
tions for the choice of instrumental variables does not involve the 
double use of the same set of data. I think that it only does so 
superficially. One does indeed use the data in applying a stopping 
rule in the regression of endogenous variables on instruments, but 
the regressions obtained by applying the stopping rule are just the 
ones which are used in estimating the equation of interest. I do 
not see that this involves using the data twice any more than does 
two-stage least squares, choice of instrumental variables by princi- 
pal components, or any simultaneous equation technique in which 
reduced form coefficients have first to be calculated. 
MAHALANOBIS 
I have a purely technical question. Can you break up into two 
or more random partitions? And secondly, it would be of interest 
61 Fisher - pag. 79