A study of alternative approaches to non-normal latent trait distributions in item response theory models used for health outcome measurement

It is often unrealistic to assume normally distributed latent traits in the measurement of health outcomes. If normality is violated, the item response theory (IRT) models that are used to calibrate questionnaires may yield parameter estimates that are biased. Recently, IRT models were developed for dealing with specific deviations from normality, such as zero-inflation (“excess zeros”) and skewness. However, these models have not yet been evaluated under conditions representative of item bank development for health outcomes, characterized by a large number of polytomous items. A simulation study was performed to compare the bias in parameter estimates of the graded response model (GRM), polytomous extensions of the zero-inflated mixture IRT (ZIM-GRM), and Davidian Curve IRT (DC-GRM). In the case of zero-inflation, the GRM showed high bias overestimating discrimination parameters and yielding estimates of threshold parameters that were too high and too close to one another, while ZIM-GRM showed no bias. In the case of skewness, the GRM and DC-GRM showed little bias with the GRM showing slightly better results. Consequences for the development of health outcome measures are discussed.

Generalized Linear Mixed Models for Randomized Responses

Abstract. Response bias (nonresponse and social desirability bias) is one of the main concerns when asking sensitive questions about behavior and attitudes. Self-reports on sensitive issues as in health research (e.g., drug and alcohol abuse), and social and behavioral sciences (e.g., attitudes against refugees, academic cheating) can be expected to be subject to considerable misreporting. To diminish misreporting on self-reports, indirect questioning techniques have been proposed such as the randomized response techniques. The randomized response techniques avoid a direct link between individual’s response and the sensitive question, thereby protecting the individual’s privacy. Next to the development of the innovative data collection methods, methodological advances have been made to enable a multivariate analysis to relate responses to sensitive questions to other variables. It is shown that the developments can be represented by a general response probability model (including all common designs) by extending it to a generalized linear model (GLM) or a generalized linear mixed model (GLMM). The general methodology is based on modifying common link functions to relate a linear predictor to the randomized response. This approach makes it possible to use existing software for GLMs and GLMMs to model randomized response data. The R-package GLMMRR makes the advanced methodology available to applied researchers. The extended models and software will seriously improve the application of the randomized response methodology. Three empirical examples are given to illustrate the methods.

Multiple sensitive estimation and optimal sample size allocation in the item sum technique

For surveys of sensitive issues in life sciences, statistical procedures can be used to reduce nonresponse and social desirability response bias. Both of these phenomena provoke nonsampling errors that are difficult to deal with and can seriously flaw the validity of the analyses. The item sum technique (IST) is a very recent indirect questioning method derived from the item count technique that seeks to procure more reliable responses on quantitative items than direct questioning while preserving respondents' anonymity. This article addresses two important questions concerning the IST: (i) its implementation when two or more sensitive variables are investigated and efficient estimates of their unknown population means are required; (ii) the determination of the optimal sample size to achieve minimum variance estimates. These aspects are of great relevance for survey practitioners engaged in sensitive research and, to the best of our knowledge, were not studied so far. In this article, theoretical results for multiple estimation and optimal allocation are obtained under a generic sampling design and then particularized to simple random sampling and stratified sampling designs. Theoretical considerations are integrated with a number of simulation studies based on data from two real surveys and conducted to ascertain the efficiency gain derived from optimal allocation in different situations. One of the surveys concerns cannabis consumption among university students. Our findings highlight some methodological advances that can be obtained in life sciences IST surveys when optimal allocation is achieved.