The models of Taylor and Russell (1939) or Abrahams, Alf and Wolfe (1971) can be used to predict structure in a selected group provided there are only two groups and the correlation between predictor and group membership variable is known. They are usually used in applied psychology to predict the proportion of successful candidates in a selected group, but they can also be used to predict any dichotomous structure (e.g., sex structure). This paper gives two models for the generalized case of m‐groups. The first model assumes that the Ns, means and standard deviations of the normally distributed predictor in m‐groups, are known, i.e., 3m parameters. The second model assumes that the distribution of the predictor and group membership variable is bivariate normal (and the regression is, therefore, linear), and that the total mean, standard deviation and correlation are known, i.e., 3+m parameters. When tested with a practical example whose distributions strongly violate the assumption of normality, the models seem to be quite robust.
Notes
The work on this paper was supported by research funds of Raziskovalna skupnost Slovenije and Zveza skupnosti za zaposlovanje that also made data available. The help of Meta Jurc and Tony Buatti is acknowledged in preparing the English version of this text.