90 James F. Crow
view grew out of this; I shall return to this subject later.
3 STATISTICS AND PATH ANALYSISWright had an interest in statistics from his earliest scientific studies. He was fa-
miliar with the work of Karl Pearson and soon became adept in statistical methods,
especially in the use of the correlation coefficient. This proclivity lasted through
his life; Wright preferred to attack almost any problem through correlations. An
example, inbreeding, has already been mentioned. His derivation of the inbreeding
coefficient by correlations seems awkward and indirect to many modern workers,
but that was how he worked. In fact, he defended his procedure, since he liked
the idea that inbreeding could be negative, if mates were less closely related than
random pairs. Correlations can be negative; probabilities, of course, cannot.
Wright was ahead of the field in two early statistical studies. In his early study
of guinea pig weights in 1917, he was able to subdivide product moments into
between- and within-group components[Wright, 1917]. In effect, he had discovered
what was later called the analysis of covariance, a subject invented independently
and carried farther by R. A. Fisher. Another example: In a study of the extent
of white-spotting in guinea pigs he found a transformation to linearize cumulative
percentage data[Wright, 1926]. This later came to be known as probit analysis
[Finney, 1971].
Wright’s most important contribution to statistical methodology is his method
of path analysis[Wright, 1921]. The algebra is that of partial regression analy-
sis, but the approach is unusual and typical of Wright. He was not particularly
interested in correlation and regression as descriptive or predictive. Rather, his
main interest was interpretative; he was especially interested in causal analysis.
He assumed that the causal pathways were known but the relative magnitude of
different paths was to be determined. His procedure was to draw diagram with a
sequence of paths. A path of influence was indicated by an arrow while unana-
lyzed correlations were indicated by two-headed arrows. This was a natural way
to diagram pedigrees, but Wright extended it to other problems. Each step in a
pathway was associated with a “path coefficient”, a partial regression coefficient
standardized by being measured in standard deviation units. A path coefficient
is then a measure of the relative contribution of this particular step in the path-
way. Wright devised simple rules from which it is easy, from the path diagram, to
write the relevant equations, which can then be solved for the unknown variables.
A virtue of the method is that it immediately shows whether there are sufficient
measurements and equations to permit a unique solution.
In 1925 Wright published a monumental analysis of the production and prices
of corn and hogs in the period from the Civil War to World War I[Wright, 1925].
This involved no less than 510 correlations. Wright did the calculations himself,
using the primitive card-sorting and calculating equipment available at the time.
It would be a quick calculation with modern computers, but at the time this
required an enormous amount of time. Remarkably, he accounted for some 80