ABSTRACT
Victor's recent alternative model for configural frequency analysis (CFA) estimates expected frequencies for hypothesized type cells from a quasi-independence log-linear model. The tests against types use chi-square components. A comparison with the binomial test shows that the standard errors of the chi-square components can be too small when samples are small. From simulation results formulas for the standard error that consider sample size are derived. The paper shows that these formulas do not suggest nonconservative decisions. Empirical data examples from legasthenia research illustrate the methodological findings.
Subject(s)
Psychometrics/methods , Analysis of Variance , Binomial Distribution , Computer Simulation , Dyslexia/diagnosis , Humans , Software , SyndromeABSTRACT
A new concept on configural frequency analysis (CFA) is presented under the title: CFA, Victor's singular types and syndromes. The new concept is suitable and highly effective if one and only one type, a singular type, is present within a population of subjects or patients. It is based on the nullhypothesis of quasi-independence which assumes that a type-population with coincident variables is superimposed to a residual population with independent variables. The singular type CFA is illustrated by a set of data where a subgroup of dyslexics is mixed up with a residual group of non-dyslexics who have been evaluated as to reading, vocabulary and spelling. The resulting type, not to be identified by classical CFA, was shown to be singular and clear-cut with weakness in all 3 verbal tests. The importance of singular types and syndromes for psychopathology is discussed clinically and theoretically.
