An Equivalence Test Approach to Rounding Parameter Estimates

Empirical researchers routinely face a simple but consequential reporting decision: how many digits to display for estimated coefficients and associated standard errors? We first document current practice, which consists in always reporting a fixed number of decimal digits, most often three, for all parameter estimates and standard errors. We propose an alternative approach to determine the number of digits to report. Specifically, we introduce the concept of a “distinct” digit and recommend reporting digits in parameter estimates up to the first non-distinct digit. Our rounding heuristic is straightforward to implement and already embedded in a widely used R package.