What question did this study set out to answer?

To compare the regression constants derived from different transformations of binomial data using maximum likelihood fitting.

synapse

⌘+K

synapse

⌘+K

March 1, 1963

Comparisons of regression constants fitted by maximum likelihood to four common transformations of binomial data*

Key Points

To compare the regression constants derived from different transformations of binomial data using maximum likelihood fitting.
Utilized an electronic computer for analysis of various transformation methods.
Fitted regression constants to arcsine, logit, log-log, and inverse normal transformations.
Applied complex data with multiple independent variables.
No practical differences were found in analyses using the four transformations.
Results lead to strong assumptions regarding the interchangeability of these methods.

Abstract

SUMMARY An electronic computer was used to fit regression constants of arcsine, logit, log‐log and inverse normal (probit) transforms of binomial data. The data ranged over a variety of types, biologically and mathematically, but were ‘complex’(with several independent variables). This empirical study leads one to assume, very unequivocally, that there is no practical difference in analyses using these four transformations. I thank Drs Daphne Trasler and Dorothy Warburton for permission to quote unpublished data and the McGill Computing Centre for advice and assistance.

Mark Helpful

Bookmark

Relay