Randomization tests: Understanding Experimental Results

This lecture introduces randomization tests as an alternative to t-tests when the normal, independent, and identically distributed assumption is violated. By creating fake random data, researchers can test for significance without external reference data. The process involves mixing up labels and analyzing variations to determine if the null hypothesis holds. The lecture also covers paired experiments, where before and after data or correlated samples can be used for randomization. Through examples like a tomato fertilizer experiment, students learn how to apply randomization tests to assess the effectiveness of treatments.