Limiting Dilution

Sat Jul 9 14:11:19 EST 1994

Lefkovits uses linear regression for his goodness-of fit test to the 
single-hit Poisson model.  The use of linear regression in this type of
analysis has been criticized by Finney, Fazekas de St. Groth, and Taswell.
The references are: Finney, D.J.: Statistical Method in Biological Assay,
3rd. ed., New York, MacMillan, 1978, ISBN 0028446402; Fazekas de St. Groth,
S., J. Immunol. Methods 49:R11, 1982; Taswell, C., J. Immunol. 126:1614
(1981).  The correct method for goodness-of-fit tests and calculation of
frequencies, according to Fazekas, is the maximum likelihood test.  This
test is described in Porter and Berry, Brit. J. Cancer, 17:583 (1963).  Also,
Fazekas wrote a calculator program to calculate these statistics and listed
the program in his paper.  I have seen a more recent paper, probably in the
Journal of Immunological Methods, advocating a "jackknife" maximum likelihood
analysis, but I don't know the reference.                 Jack Komisar

More information about the Immuno mailing list