Non Parametric Tests

Comparing Three or More Groups: Kruskal-Wallis Test and Friedman Test0:04:36

When comparing three or more independent groups with non-parametric data, the Kruskal-Wallis test serves as the non-parametric ANOVA. It determines if there's a significant difference based on average ranks among groups. For repeated measures (same group under multiple conditions), the Friedman test is used, ranking conditions for each individual and summing ranks to check for significant differences between conditions.

Measuring Relationships: Spearman's Rank Correlation0:06:06

When looking for relationships between two variables with unconventional data, Spearman's rank correlation is the appropriate tool. Unlike Pearson correlation, it doesn't assume a linear relationship and is ideal for data already in ranks or when the connection is non-linear. It assesses if higher ranks in one variable correspond to higher (or lower) ranks in another.

Conclusion: Choosing the Right Tool for Scientific Honesty0:07:00

The video concludes by emphasizing that for every common statistical scenario, a robust non-parametric counterpart exists. Choosing the right statistical tool that respects the true nature of the data, however messy, is crucial for scientific honesty and allows the data to speak for itself.

The Problem with Parametric Tests in Real-World Data0:00:00

Traditional parametric tests like t-tests and ANOVA assume data follows a perfect bell curve, which is rarely the case in health research. Real-world data is often messy, skewed by outliers, or comes from small samples, making parametric tests unsuitable and potentially misleading. Non-parametric tests are designed for these real-world scenarios.

Introduction to Non-Parametric Tests: The 'Distribution Free' Solution0:01:25

Non-parametric tests are flexible and robust alternatives that do not require data to fit a specific distribution. They handle messy data by ignoring actual values and focusing on their ranks. Data is sorted from smallest to largest, and each value is replaced by its rank, neutralizing the effect of outliers and skewed distributions.

Comparing Two Groups: Wilcoxon Signed-Rank Test and Mann-Whitney U Test0:03:00

For 'before and after' studies with non-normal data, the Wilcoxon Signed-Rank test is used. It ranks the differences between paired measurements to see if improvements systematically outweigh negative changes. For comparing two independent groups (e.g., treatment vs. placebo) with skewed data, the Mann-Whitney U test is the non-parametric equivalent of an independent t-test, comparing ranks across combined groups.