Abstract:Simple correlations play a large role in the analysis of psychiatric data. They are used to predict outcome, validate new instruments, establish treatment efficacy and find symptom patterns. Researchers and data analysts often face a question about which correlation coefficient to use in a study but are often unaware of the strengths and weaknesses of the alternative correlation measures. The presence of outliers, nonconstant variance, skewed distributions and unequal n are common in psychiatric data and this poses severe problems for many classic statistical methods. We compare Pearson, Spearman and Kendall's correlation coefficients using a large sample of subjects with schizophrenia spectrum disorders who were evaluated with 7 different psychiatric rating scales. Samples sizes ranging from 8 to 50 were evaluated using bootstrapping methods. The criteria for evaluation of the correlations were the type I error rates, power, bias and confidence interval width. Pearson's r did not always control for false positives at the nominal rate and was often unstable. Spearman's r performed better than Pearson's but provided a biased estimate of the true correlation. Spearman's r was also difficult to interpret. Our results suggest that Kendall's tau(b) has many advantages over Pearson's and Spearman's r; when applied to psychiatric data, tau(b) maintained adequate control of type I errors, was nearly as powerful as Pearson's r, provided much tighter confidence intervals and had a clear interpretation.

Comparison of Spearman's rho and Kendall's tau in Normal and Contaminated Normal Models

An adjusted-range based self-normalization test for correlation change

On the Non-degeneracy of Kendall's and Spearman's Correlation Coefficients

Asymptotic Properties of Pearson's Rank-Variate Correlation Coefficient under Contaminated Gaussian Model

Adjusted-range based self-normalized tests for correlation constancy

Monotonicity properties of the asymptotic relative efficiency between common correlation statistics in the bivariate normal model

Correlating and predicting psychiatric symptom ratings: Spearman's r versus Kendall's tau correlation

Comparison of correlation-based measures of concordance in terms of asymptotic variance

A robust Spearman correlation coefficient permutation test

Mean Test with Fewer Observation than Dimension and Ratio Unbiased Estimator for Correlation Matrix

A Short Note on Spearman Correlation: Impact of Tied Observations

Comparison of Values of Pearson's and Spearman's Correlation Coefficients on the Same Sets of Data

The effect of normality and outliers on bivariate correlation coefficients in psychology: A Monte Carlo simulation

A Comparison of the Pearson, Spearman Rank and Kendall Tau Correlation Coefficients Using Quantitative Variables

An Extreme-Value Approach for Testing the Equality of Large U-Statistic Based Correlation Matrices

A short note on the maximal point-biserial correlation under non-normality

On Testing the Mean Equivalence of Treatments from Correlated Normal Populations

A Novel Kernel Correlation Coefficient with Robustness Against Nonlinear Attenuation and Impulsive Noise

Regression models of Pearson correlation coefficient

Pearson's chi‐square test and rank correlation inferences for clustered data

On the distribution of the correlation coefficient when sampling from a mixture of two bivariate normal densities: Robustness and the effect of outliers