Permutation methods for the analysis of matched-pairs experimental designs

Area of Science:

• Statistics
• Biostatistics
• Experimental Design

Background:

• Conventional statistical tests like the t test can be sensitive to outliers.
• Matched-pairs experimental designs are common in various scientific fields.
• Robust statistical methods are needed to ensure reliable analysis in the presence of extreme values.

Purpose of the Study:

• To compare the performance of Fisher-Pitman and powers of ranks permutation tests against the conventional t test.
• To evaluate the resistance of these permutation tests to extreme values in matched-pairs data.
• To determine the suitability of permutation tests for robust analysis of experimental data.

Main Methods:

• The study involved analyzing matched-pairs experimental data.
• Comparison of the Fisher-Pitman permutation test and powers of ranks permutation test with the conventional t test.
• Assessment of the tests' sensitivity and resistance to extreme values.

Main Results:

• The Fisher-Pitman and powers of ranks permutation tests demonstrated substantial resistance to extreme values.
• The conventional t test showed greater sensitivity to extreme values compared to the permutation tests.
• Permutation tests offer a more reliable analysis when extreme data points are present.

Conclusions:

• Fisher-Pitman and powers of ranks permutation tests are recommended for analyzing matched-pairs designs, especially when outliers are suspected.
• These permutation tests provide a robust alternative to the t test, enhancing the reliability of statistical inferences.
• The findings highlight the importance of choosing appropriate statistical methods for data analysis to mitigate the impact of extreme values.