Analysing and treating test item nonresponse

M. Assessment & Eval.This study uses an empirical and normative methodology to analyse the effects and various treatments of nonresponse on the measurement of student ability in large scale educational achievement studies. The primary focus of the study is an empirical analysis of disturbances introduced into psychometric measurement by the presence and treatment of nonresponse. This focus is balanced by a normative analysis of the possible implications of a series of psychometric treatments of nonresponse on students' response strategy and on the conduct of standardised testing. Initially, the scope, focus and research questions of the current study are outlined. Second, the research perspectives used to study nonresponse and used in this thesis are reviewed. Third, nonresponse is defined and distinguished from other forms of test item missing data. Methods proposed for analysing and treating test item nonresponse are considered, and the six treatments examined in this thesis are presented. The methodology underpinning the current analysis is then presented. This study investigates the empirical performance of each of these treatments using data from a crossnational test of Reading ability. After presenting the characteristics of the test and sample, the assumptions and methods underpinning analysis of the treatments are described. First, item response theory and Rasch measurement are discussed. Second, the nonresponse treatments are considered in more depth. Finally, the psychometric and statistical methods used to evaluate the treatments are presented. Results describing the distributions of nonresponse in the samples and the characteristics of student ability estimates under each of the treatments are presented. Next, the psychometric, response strategy and situational implications of each treatment are discussed. The relative merits of these treatments are summarised in conclusion, and implications for research and practice are considered