Mathematics
8 years ago

T Test Not Always Reliable for Unequal Variances, New Guidelines Recommended

Journal of Educational Statistics
Exact Type 1 Error Rates for Robustness of Student's <i>t</i> Test with Unequal Variances
Philip H. Ramsey

Paper Summary

Related papers

Theme background image
Mathematics
8 years ago

T Test Not Always Reliable for Unequal Variances, New Guidelines Recommended

Clinical Trials
Missing Data
Causal Inference
1 view
Theme background image
Mathematics
8 years ago

New bootstrap test improves accuracy of variance equality assessments

Clinical Trials
Regularization Methods
Financial Volatility
Theme background image
Mathematics
8 years ago

Permutation tests show more reliable results in statistical analysis accuracy.

Clinical Trials
Missing Data
Skew Distributions

Disagreements exist about how reliable the t test is when comparing groups with different variances. By using Hsu's method, it was found that the t test may not always be reliable when variances are unequal, even with equal sample sizes. To use the t test confidently at a significance level of 0.05 with unequal variances, sample sizes of at least 15 are needed. In cases where sample sizes are unequal, it is recommended to use Welch's procedure instead.