Detection of outliers in test, questionnaire and survey data

Wobbe Zijlstra

Project: Department of Methodology, Faculty of Social Sciences, Tilburg University

Financed by : NWO

Supervisors: Prof. dr K. Sijtsma, dr A. Van der Ark

Project running from: 1 September 2005 – 1 September 2009

Summary:
Outliers, often defined as observat ions, which are inconsistent with the remaining data, are influent ial to the degree in which they affect stat ist ical results. Out lier research is often done with cont inuous variables, such as income and age. This project defines out liers for highly discrete data in test , quest ionnaire and survey data, and proposes several methods for detect ing out liers. I t is also invest igated when an out lier is influent ial, and how to deal with influent ial cases. Simulated and numerous real data sets are used. The project will result in user – friendly software for handling data quality problems, including out lier analysis.

Date of defence:  4 December 2009

Title of thesis:  Outlier detection in test and questionnaire data for attribute measurement

ISBN/EAN:  978-90-5335-232-8