Comparison of Neyman-Pearson and Evidential Approaches inSeparate One-Sided Tests Based on Model Misspecification Error

model misspecification, the preferred (or more accurate) hypothesis is the one that is closer to reality.This implies that the hypothesis should be chosen such that the Kullback-Leibler divergence between the models under that hypothesis and the true model is minimized.Under model misspecification, the probability of hypothesis test errors is redefined as the probability of true errors.Finally, based on the true error probability, the performance of both the Neyman-Pearson and evidential approaches is examined in separate one-sided hypothesis testing. Resuts an isussionResults indicated that under model misspecification, the true Type II error probability of the Neyman-Pearson test approaches zero as the sample size increases; however, the true Type I error probability may approach one as the sample size increases.In other words, under model misspecification, the Neyman-Pearson test tends to reject the null hypothesis as the sample size grows, regardless of which hypothesis is closer to reality.On the other hand, results showed that in hypothesis testing based on the evidential approach, even under model misspecification, both the true probability of misleadingly strong evidence and the true probability of weak evidence approach zero as the sample size increases.In other words, under model misspecification, in evidential testing, the probability of strong support for the correct hypothesis approaches one as the sample size grows. onusionThe findings of this research suggest that under model misspecification, the evidential approach outperforms the Neyman-Pearson approach in hypothesis testing.