Visualization of statistically significant correlation coefficients from a correlation matrix: a call for a change in practice

Correlation matrices are tabular numerical displays of correlation coefficients that provide information on pairwise relationships between variables. Often times, they provide information about the statistical significance of correlation coefficients, usually at multiple levels of significance. In two studies, we provide evidence that commonly used formats for displaying statistical significance, namely, the use of different number of asterisks and the use of absolute values, are inefficient. Using lessons drawn from the literature on visual perception, we propose the use of variations in hue and intensity of numbers to reduce the amount of time and effort taken to glean information from correlation matrices. We also create and describe a web-based engine that can be used to implement these modified approaches to display correlation matrices.