Charting the Future of Configural Frequency Analysis: The Development of a Statistical Method

This issue of the journal, Volume 45 from 2003, collects work devoted entirely to Configural Frequency Analysis, or CFA. Alexander von Eye and Erwin Lautsch open the collection by tracing where the method came from and where its authors expected it to go. The rest of the volume splits into two halves: one on how CFA gets applied to real problems, and one on technical questions about the method itself.

The applications begin with a reanalysis of older data. Marcus Ising and Wilhelm Janke return to a placebo experiment Janke ran between 1963 and 1966, comparing what multiple regression and CFA each reveal about who responds to a placebo. Ising contributes a second piece arguing that CFA can serve genetic association studies of complex diseases, where the usual tools often struggle.

Several authors push the method into assessment and social measurement. Michaela M. Wagner-Menghin uses CFA to sort people into achievement-motive types, reading the results through a test battery called Work Style that she treats as a fresh instrument for measuring the achievement motive. Erwin Lautsch and Ulf Thöle apply the same analytic logic to the 14th Shell Youth Study of 2002, classifying and then explaining the different conceptions of life that young respondents held.

One of the more playful entries comes from Mark Stemmler and Günther Bäumler, who test whether a man's family name predicts which athletic discipline he excels in, looking specifically at men named Smith or Tailor. The point is less the names than the question of whether a statistical association is really there once you examine the pattern of cells.

Two comparative studies measure CFA against its rivals. Martin Reuter, Michael Hüppe, Petra Netter and Juergen Hennig set Structural Equation Modeling beside CFA on the same empirical material. Lautsch and Michael M. Plichta widen the field, putting CFA next to Multiple Correspondence Analysis and Latent Class Analysis to see how the three handle the same data.

The second half turns inward, toward the mechanics. Joachim Krauth asks a blunt question in one paper: is it even permissible to dichotomize a variable at its median before running CFA? In another he works through the notion of type structures. Stefan von Weber, Lautsch and von Eye examine where the method breaks down when the tables are small, and in a companion paper they propose continuity corrections tailored to the specific table being analyzed.

Testing procedures get close attention too. Von Eye compares the tests ordinarily used in 2 by 2 tables with those used in two-sample CFA. Lautsch and von Weber introduce a new test procedure outright. Jörg Betzin and Peter Bollmann-Sdorra take up the question of a determination coefficient for CFA, the kind of fit measure that quantifies how much the configurations explain.

Stemmler appears again, once with C. Raymond Bingham on using log-linear modeling and two-sample CFA to search for discrimination types, and once with von Eye on analyzing change by swapping in different base models, since the base model is what decides which configurations count as surprising. Lautsch, von Eye and von Weber round out the technical section with a survey of the software available for running CFA at the time.

The volume closes with something archival. Krauth prints three manuscripts on the mathematical principles of CFA, numbered I, II and III, that had not appeared in print before.