Related Experiment Videos
A Monte Carlo evaluation of tests for comparing dependent correlations
James B Hittner1, Kim May, N Clayton Silver
1Department of Psychology, College of Charleston, SC 29424, USA. hittnerj@cofc.edu
The Journal of General Psychology
|May 30, 2003
Summary
This study found that statistical test performance for dependent correlations depends on sample size, distribution, predictor intercorrelation, and effect size. Dunn and Clark
Area of Science:
- Statistics
- Psychometrics
- Quantitative Psychology
Background:
- Comparing dependent correlations is crucial in various research fields.
- Previous simulation studies have explored statistical tests, but gaps remain.
- Understanding Type I error rates and statistical power is essential for accurate conclusions.
Purpose of the Study:
- To evaluate the Type I error rates and statistical power of eight statistical tests for dependent zero-order correlations.
- To investigate the influence of sample size, population distribution, predictor intercorrelation, and effect size on test performance.
- To identify the most robust statistical tests for applied researchers.
Main Methods:
- A Monte Carlo simulation was employed to assess eight statistical tests.
- Simulations were conducted across various sample sizes (20-300), population distributions (normal, uniform, exponential), predictor-criterion correlations (.1-.7), effect sizes (.1-.6), and predictor intercorrelations (.1-.6).
- Type I error rates and statistical power were systematically analyzed.
Main Results:
- Both Type I error rates and statistical power were significantly influenced by sample size, population distribution, predictor intercorrelation, and effect size.
- The performance of statistical tests varied considerably under different conditions.
- O. J. Dunn and V. A. Clark's (1969) z-test and E. J. Williams's (1959) t-test demonstrated superior overall statistical properties when considering both Type I error rate and power.
Conclusions:
- The choice of statistical test for comparing dependent correlations is critical and should account for multiple factors.
- Dunn and Clark's z-test and Williams's t-test are recommended for their robust performance across various simulation conditions.
- These findings provide valuable guidance for applied researchers, enhancing the reliability of statistical inferences in studies involving dependent correlations.