OFFICE MEMORANDUM e STANFORD UNIVERSITY © OFFICE MEMORANDUM @ STANFORD UNIVERSITY ¢ OFFICE MEMORANDUM

Date: November 13, 1973

To: Dr. Joshua Lederberg
FROM: L. L. Cavalli-Sforza
a emma
SUBJECT: Chronology of research on phylogenetic analysis

1961

1963
1964
1965
1966

1966
(unpub1.)
1970

1973

1973

Work with Anthony Edwards in Pavia begun.

Three methods--all approximate--for analysis in the case of

gene frequencies or anthropometric traits (least squares,

minimum path, cluster analysis. All use matrix of distances

between population pairs).

Paper at the Hague Genetics Congress.

Anthony and I start on maximum likelihood solution of problems.

Fitch reinvents independently method 1.

Camin and Sokal independently find approximate method for qualitative

traits and accept principle that phylogenetic analysis differs from

standard taxometry.

J. Felsenstein proves covariances between populations are proportional

to evolutionary time. Starts on maximum likelihood solution.

Anthony proves that full maximum likelihood solution is practically

impossible. Gives paper at Royal Statistical Society of London,

all big brass of British statistics attending. No useful advice

obtained.

J. Felsenstein publishes iterative maximum likelihood solution

using matrix of distances.

September, I get interested in measuring "treeness", that is how

well data are represented by a model of independent evolution. On a

problem concerning distribution of non-independent correlation

coefficients, I consult with Ingram Olkin. He mentions a formally

related problem was solved by a student of Wilks and he himself has
provided an extension designed to solve a psychological testing

problem. Ingram gives me references to Wilks, Votaw's papers. On

this basis, the problem of estimation of tree parameters and of

treeness can be solved by the likelihood ratio criterion (a somewhat

different procedure from classical maximum likelihood but almost

as widely accepted). It is basic to use matrix of variances and

covariances, not of distances. Solution would have been easy since

the beginning if the Wilks' paper~-a classic, but rarely read in the

original even if it is very readable--had been reinterpreted as an

example of a tree with all populations splitting at once.

Some lessons:

1) Fundamental papers should be read in the original and not learned from textbooks.
2) There is no universal method best for everything.
3) One never considers things from sufficiently different angles.

DEC 24 1980 Len cp-s/o.2 Ly

WNONVYOWIW JD1ddO © ALISYTAINN GYOINVIS © WNGNVYOWIW 3Dd1ddO © ALISUTAINN GYOA4NVIS ®© WNANV¥OW|W FD1ddO © ALISHZJAINN GYOANVIS ©