COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	\input basic
C00009 ENDMK
C⊗;
\input basic
\ctrline{Critique of Path Analysis}
\vskip 36 pt
\ctrline{\sl by Samuel Karlin and Ed Cameron}
\vskip 2.54 cm
{\bf I. Introduction}
\vskip 2.54 cm
The purposes of this paper is to offer criticism of the technique of linear
path analysis, especially as applied in the context of genetic epidemiology.
The method of path analysis is currently in wide use and is vigorously 
promoted by such investigators as N. Morton, D. C. Rao, T. Reich, R. Cloninger,
O. Eaves, O. Nance, and others for objectives of discerning genetic, environment
al, and cultural components contributing to familial resemblances.  Their 
studies are generally data consisting of biochemical, physiological, 
environmental, and cultural measurements collected from individuals and
families and pedigrees of various structures.  The procedures involved are 
so wide spread and established in application to genetic epidemiological
problems that they have become available in the form of packaged computer
programs.

Any methodology which makes claim to be coming an established method 
of procedure, especially in an important area of concern such as 
epidemiology should be subject to the most severe scrutiny, as many 
investigators are likely to use the method, especially since it is 
available in the form of a packaged computer program, without stopping
to evaluate its basic assumptions and tenets.  (The assumptions and tenets
may even become blurred to those who are most involved with its use and
formulation).  In fact what tends to happen is that due to inertia in the
flow of creative ideas and the desire for practitioners to "have answers",
no matter how spurious they might be, once a method becomes established as 
the method to use these is a tendency to hold it up as sacrosanct and a 
skepticism toward criticism of its basic tenets and inclinations 
and develops those who are most involved in promoting the techniques
may automatically tend to resist criticism of it because they personally
will have a vested interest in its maintenance as an established methodology.

It is in this context that we offer our criticism of the use of linear path
analysis in genetics epidemiology with full criticizing a method which is
now being held up as the established and accepted one and the one with 
which most investigators and practitioners have become comfortable and 
which they tend to trust, desite whatever extensive limitatins or even
implicit and explicit assumptions inconsistencies might exist within its
framework.

Our conclusion is that in actuality there are so many serious and major
problems involved with the formulation of, application of, and use of
statistical analysis in linear path analysis that a major reevaluation
of its usefulness in and place in the treatment of genetic 
epidemiological data is in order.  As a result, we especially feel it 
is premature and unadvised to use results of such analysis in the 
assessment of disease risk and similar applications by the medical 
profession.

In Section II we present a general multidisciplinary perspective on the
problem of determining casual relationships between measurable
variables in the context of the assumption of the existence of so-
called latent variables which are assumed to exist and influence
the measurable variables but are themselves assumed to be unmeasurable
({/sl at least directly unmeasurable as their values may become more 
delineated through measurement of other related variables}).  We do
so to place the issue of path analysis in a broader perspective of
approaches taken in many fields to linear causal analysis which may
shed some light on its limitations or strengths relative to other methods.



In Section III we give a description of the basic path analysis model
using as a descriptive example a model suggested by D. C. Rao for analysis
of the Lipid Research Clinic Collaborative Family Data.  In Section IV 
we offer a large host of general criticisms of the path analysis formulation
with an especially extensive analysis of the ({/sl implicit and explicit})
role of normality assumptions.  

Section V consists of an extensive description and criticism of the
estimation and hypothesis testing methodologies used in path analysis.

Finally, we summarize our objections of attempt to offer a fresher
perspective on the problem in Section VI.


\vfill\end