MEASURING THE DIFFERENCE (X-Y) OF SIMULATED DISTRIBUTIONS - A CONVOLUTIONS APPROACH

Reampling or simulation techniques are now frequently used in applied economic analyses. However, significance tests for differences between empirical distributions have either invoked normality assumptions or have used nonoverlapping confidence interval criteria. We demonstrate that such methods generally will not be appropriate, and we present an empirical test, based on the method of convolutions, for assessing the statistical significance between approximate empirical distributions created by resampling techniques. The proposed convolutions approach is illustrated in a case study involving empirical distributions from dichotomous choice contingent valuation data.