Doing Better Statistics in Human-Computer Interaction

Each chapter of this book covers specific topics in statistical analysis, such as robust alternatives to t-tests or how to develop a questionnaire. They also address particular questions on these topics, which are commonly asked by human-computer interaction (HCI) researchers when planning or completing the analysis of their data. The book presents the current best practice in statistics, drawing on the state-of-the-art literature that is rarely presented in HCI. This is achieved by providing strong arguments that support good statistical analysis without relying on mathematical explanations. It additionally offers some philosophical underpinnings for statistics, so that readers can see how statistics fit with experimental design and the fundamental goal of discovering new HCI knowledge.

Getting started; Part I. Why We Use Statistics: 1. How statistics support science; 2. Testing the null; 3. Constraining Bayes; 4. Effects: what tests test; Part II. How To Use Statistics: 5. Planning your statistical analysis; 6. A cautionary tail: why you should not do a one-tailed test; 7. Is this normal?; 8. Sorting out outliers; 9. Power and two types of error; 10. Using nonparametric tests; 11. A robust t-test; 12. The ANOVA family and friends; 13. Exploring, over-testing and fishing; 14. When is a correlation not a correlation?; 15. What makes a good Likert item?; 16. The meaning of factors; 17. Unreliable reliability: the problem of Cronbach's alpha; 18. Tests for questionnaires.

Paul Cairns is a reader in Human-Computer Interaction at the University of York and Scholar-in-Residence for The AbleGamers Charity that helps people with disabilities combat social isolation by making videogames more accessible. He has taught statistics at all levels of education for nearly twenty years. His particular research interest is in players' experiences of digital games, and his expertise in experimental and statistical methods was developed through working in this area.