SEMAINE D ETUDE SUR LE ROLE DE L’ANALYSE ECONOMETRIQUE ETC. 
387 
zations. The second includes: two-stage least squares; limited- 
information maximum likelihood; the other members of 
Theil’s k-class; Theil’s h-class; and Nagar’s double k-class (2). 
All the estimators in this group have the common property that 
whereas (unlike ordinary least squares) they take account of 
the simultaneous nature (if any) of the equations in the model 
to be estimated, they use only a priori restrictions on one 
equation at a time. Accordingly, we shall call such estimators 
« limited information » methods. The last class of estimators 
consists of those methods which do use information on all equa- 
tions at once, what we shall term « full information » methods 
Among these, of course, is full-information maximum likeli- 
hood, but the class also contains A. ZELLNER and H. THEIL’s 
three-stage least squares, an estimator recently proposed by 
T.J. ROTHENBERG and C.T. LEENDERS called « linearized 
maximum likelihood », and the simultaneous least squares 
estimator of T.M. Brown (3). 
In principle, all of the above estimators make use of all 
exogenous and lagged endogenous variables in the model as 
predetermined instruments. As indicated above, for reasons to 
be discussed below, this cannot always be done or is not always 
desirable, and in such cases other methods which so employ 
only some of the exogenous or lagged endogenous variables 
must be used. We shall discuss the problems raised in such 
situations below, observing here only that, given the choice of 
variables to be treated as predetermined, most of the estimators 
just classified have exact counterparts in such circumstances. 
(?) See THEIL [32, pp. 353-354] and Nacar [23]. 
‘\ See ZELLNER and THEIL [37]. ROTHENBERG and LEENDERS [26j, and 
[M Brown (TF 
bi 
Fisher - pag. 3