and it can be assumed that this present tendency will even accelerate with the increasing availability and proliferation of Joereskog's LISREL program, which is designed to estimate such models.
In contrast to almost all writers dealing with the problem of unmeasured variables, the author of this article is convinced that this development is highly undesirable. It seems that because most authors are mainly concerned with discussing and/or solving formal problems of identification and estimation, fundamental flaws have been overlooked. In fact, the whole approach using information about measured variables to make inferences about the relationship between unmeasured and measured variables as well as between unmeasured variables themselves is, as I believe, untenable on substantial grounds. This is of course a very general thesis. I am not able to defend it in this short article; rather, ! will only argue in favor of the weaker thesis that there are at least some inferences, such as those mentioned above, generally believed to be acceptable which are in fact not so. lI To reach this objective, I chose to discuss one simple example which typifies the kind of inferences under criticism. Actually, the model