Decision Tree in Statistical Analysis
Overview of Nonparametric Techniques
Nonparametric techniques are used to analyze data that do not require a normal distribution. The two primary nonparametric tests discussed are the Wilcoxon and Mann Whitney tests for comparing conditions.
Mann Whitney Test
Definition: A nonparametric test used for comparing two independent groups (between subjects).
Example Scenario: Comparing scores from boys vs. girls where participants in one group cannot be the same as in the other.
Ranking Method: All scores from both conditions are ranked together. Ties share average ranks.
U Statistic Calculation: To determine significance, compute the U statistic using:
U = n1 \times n2 + \frac{n1(n1 + 1)}{2} - T_1
Where:
$n_1$ = number of participants in group one
$n_2$ = number of participants in group two
$T_1$ = sum of ranks for group one
Significance Testing: Compare U value to critical values in Mann Whitney tables to determine statistical significance.
Wilcoxon Test
Definition: A nonparametric test used for comparing two related groups (within subjects).
Application: Suitable for repeated measures, where the same individuals are assessed under different conditions.
Friedman Test
Purpose: Used for analyzing nonparametric data across more than two related groups (within subjects). *(shows differences)
Ranking Method: Data is ranked within each subject across multiple conditions.
Statistic Calculation: Involves using ranks to compute a test statistic which reveals the variance among conditions.
Significance Testing: Compare computed statistic to relevant tables to determine significance.
Kruskal-Wallis Test
Purpose: Used for comparing more than two independent groups (between subjects).
Ranking Method: All groups’ scores are ranked together, and total rank sums are calculated.
Statistic Calculation: Utilizes total ranks to compute a Kruskal-Wallis H statistic.
Significance Testing: Compare H statistic against critical values for nonparametric testing.
Page L Test
Definition: Conducted after establishing differences via the Friedman test to find trends in related groups. (* to show directions )*
Ranking Method: Order groups based on their total scores to evaluate if there’s a significant trend.
Conclusion
The discussed tests (Mann Whitney, Wilcoxon, Friedman, Kruskal-Wallis, and Page L) help establish statistical significance in various experimental designs without relying on parametric assumptions. They cater to both within-subjects and between-subjects comparisons, thereby enhancing the analysis of non-normally distributed data.