Academic journal article Psychologische Beiträge

Identifying Types, Response Errors, and Unscalable Respondents from Personality Questionnaires

Academic journal article Psychologische Beiträge

Identifying Types, Response Errors, and Unscalable Respondents from Personality Questionnaires

Article excerpt


Latent class analysis provides a toolbox useful for a variety of research questions. With reference to diagnostics in psychology and related fields, unconstrained latent class models may be used to identify types of respondents. Constrained models may be used for scaling items and respondents, for identifying unsealable respondents whose response behavior is not governed by a certain scaling model, and for assessing response errors (unexpected solving behavior in the sense of omissions, intrusions, guessing, and cheating). For these purposes, some submodels of the latent class family are presented: older ones, for example, the latent distance model, and recently developed ones, for example, so called hybrid models combining the Rasch or Birnbaum model with an additional unscalable class. Three examples illustrate the practical application and its possible results a) using data from the Marburg Behavior Inventory, b) evaluating diagnostic criteria for mania, and c) investigating fear of HIV infection.

Key words: restricted latent class models; hybrid models; behavior disorders; criteria for mania; fear of HIV infection.

1. Introduction

Latent class analysis (LCA) does not belong to the standard statistical methods in psychological research. Developed since 1950 (cf. Lazarsfeld & Henry, 1968), it was seen primarily as a tool for scale construction, as a factor analysis for qualitative data, and as a method for finding types (clusters, "classes"). Today, latent class theory is embedded in the family of finite mixture models (Everitt & Hand, 1981; McLachlan & Basford, 1987; Titterington, Smith & Makov, 1985) and is recognized as a useful method for modeling observed heterogeneity. Its domain of applications has considerably enlarged from the social sciences, especially sociology and psychology, to biology and medicine; for an overview, see Formann and Kohlmann, 1996.

The purpose of this paper is to demonstrate the utility of latent class models in the analysis of personality questionnaires. More specifically, three types of research problems will be considered where employing latent class models may be helpful: first, the classic classification problem, second, the problem of detecting and of quantifying response errors, and third, the problem of identifying subjects with aberrant response behavior. To tackle the first problem, the basic, i.e., unconstrained latent class model may be appropriate. Both remaining problems may be solved using constrained latent class models whose structured component expresses a specific scaling model (including response errors) and whose unstructured component represents the unscalable respondents. Their response behavior is not governed by the scaling model, but follows a random mechanism Some models providing for unscalable respondents are described, among others the well-known latent distance model going back to Lazarsfeld (1950), but also recently developed models being related to the models by Ranch (1960) and Birnbaum (1968).

Three empirical examples will illustrate the goals, the model selection process, typical results and their interpretation. The first example referring to the classification problem uses data from the Marburg Behavior Inventory (Ehlers, Ehlers & Makus, 1978); it demonstrates the usefulness of LCA for differentiating children with behavioral disorders from unproblematic children. The second example investigates the validity of diagnostic criteria for mania; employing the latent distance model with unscalable respondents, it is shown that most people are scalable on the basis of the five symptoms under investigation, but that assuming an additional class of unscalable subjects allows an essentially better description of the data observed. A similar conclusion has to be reported for the third example concerning fear of acquiring HIV infection. However, this conclusion is based on the application of the more recently proposed latent class models of the Rasch and Birnbaum type. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.