Overview of Statistical Tests

All of these tests assume quantitative variables. For statistical tests on counts or frequencies of categorical variables, consider Chi-squared tests.

• T-tests assume a normal distribution
• Z-tests (test based on z-scores)
  • Do not assume a normal distribution
  • Require larger samples (typically N > 30)
• Non-parametric tests
• Do not assume normal distribution
• May work on ordinal data
• Use Mann-Whitney test for two independent samples
• Use Wilcoxon Signed Ranks Tests for paired data
• Test for a non-zero correlation coefficient (example SAS program)

Datasets for demonstrating statistical tests

• Educational software data using Electric Field Hockey (EFH). The data values are student performance on a post-test. The experiment used two conditions: students played EFH as a game and students interacted with EFH with no specific goals. This is an unpaired (independent sample) study.
• User interface redesign data. The data values measure the time in minutes to complete the task. The experiment used two conditions: the original design and the revised design. This is a paired study, where each test user interacted with both designs. The study was counter-balanced to mitigate order effects.