A maximum statistic for the one-sided location-scale alternative in the two-stage design

Abstract An increase in location is typically accompanied by an increase in variability. Subsequently, the heteroscedasticity can indicate a treatment effect. Therefore, it may be appropriate to perform a location-scale test. A common statistic for a location-scale test is the sum of a location and scale statistic. As demonstrated by Neuhäuser (Biometri J 43:809–819, 2001), weighting the sum increases the power. Although weights cannot usually be reasonably selected a priori, a weighting is possible in an adaptive design using the information obtained in an interim analysis. Here, we propose an adaptive statistic that increases and stabilizes the power. The power performance in various situations for continuous and discrete distributions is investigated using Monte Carlo simulations, which reveal that the proposed statistic increases and stabilizes the power, thus rendering it a strong competitor to existing location-scale statistics. The new statistic is illustrated using real data.