Please note, due to essential maintenance online transactions will not be possible between 09:00 and 13:00 BST, on Monday 20th January 2020 (04:00-08:00 EDT). We apologise for any inconvenience.
The t-test is a work horse of a lot of statistical analysis in HCI. There are a lot of myths about how robust it is to deviations from normality and other assumptions. However, when faced with practical data, particularly those coming from usability studies, the claims of robustness do not stand up. This chapter reevaluates the t-test as a test for an effect on the location of data. This leads to considering robust measures of location, such as trimmed or Winsorized means and associated Yuen–Welch test as a robust alternative to the traditional t-test.