Search

While there appears to be a general consensus in the literature on the definition of the estimand and estimator associated with the Wilcoxon-Mann-Whitney test, it seems somewhat less clear as to how best to estimate the variance. In addition to the Wilcoxon-Mann-Whitney test, we review different proposals of variance estimators consistent under both the null hypothesis and the alternative. Moreover, in case of small sample sizes, an approximation of the distribution of the test statistic based on the t-distribution, a logit transformation and a permutation approach have been proposed. Focussing as well on different estimators of the degrees of freedom as regards the t-approximation, we carried out simulations for a range of scenarios, with results indicating that the performance of different variance estimators in terms of controlling the type I error rate largely depends on the heteroskedasticity pattern and the sample size allocation ratio, not on the specific type of distributions employed. By and large, a particular t-approximation together with Perme and Manevski's variance estimator best maintains the nominal significance level

Multiple and combined endpoints involving also non-normal outcomes appear in many clinical trials in various areas in medicine. In some cases, the outcome can be observed only on an ordinal or dichotomous scale. Then the success of two therapies is assessed by comparing the outcome of two randomly selected patients from the two therapy groups by 'better', 'equal' or 'worse'. These outcomes can be described by the probabilities $p^-=P(X~Y)$. For a clinician, however, these quantities are less intuitive. Therefore, Noether (1987) introduced the quantity $\lambda=p^+ / p^-$ assuming continuous distributions. The same quantity was used by Pocock et al. (2012) and by Wang and Pocock (2016) also for general non-normal outcomes and has been called 'win-ratio' $\lambda_{WR}$. Unlike Noether (1987), Wang and Pocock (2016) explicitly allowed for ties in the data. It is the aim of this manuscript to investigate the properties of $\lambda_{WR}$ in case of ties. It turns out that it has the strange property of becoming larger if the data are observed less accurately, i.e. include more ties. Thus, in case of ties, the win-ratio looses its appealing property to describe and quantify an intuitive and well interpretable treatment effect. Therefore, a slight modification of $\lambda_{WR} = \theta / (1-\theta)$ is suggested, namely the so-called 'success-odds' where $\theta=p^+ + \frac12 p_0$ is called a success of a therapy if $\theta>\frac12$. In the case of no ties, $\lambda_{SO}$ is identical to $\lambda_{WR}$. A test for the hypothesis $\lambda_{SO}=1$ and range preserving confidence intervals for $\lambda_{SO}$ are derived. By two counterexamples it is demonstrated that generalizations of both the win-ratio and the success-odds to more than two treatments or to stratified designs are not straightforward and need more detailed considerations.

There are many different proposed procedures for sample size planning for the Wilcoxon-Mann-Whitney test at given type-I and type-II error rates $\alpha$ and $\beta$, respectively. Most methods assume very specific models or types of data in order to simplify calculations (for example, ordered categorical or metric data, location shift alternatives, etc.). We present a unified approach that covers metric data with and without ties, count data, ordered categorical data, and even dichotomous data. For that, we calculate the unknown theoretical quantities such as the variances under the null and relevant alternative hypothesis by considering the following `synthetic data' approach. We evaluate data whose empirical distribution functions match with the theoretical distribution functions involved in the computations of the unknown theoretical quantities. Then well-known relations for the ranks of the data are used for the calculations. In addition to computing the necessary sample size $N$ for a fixed allocation proportion $t = n_1/N$, where $n_1$ is the sample size in the first group and $N = n_1 + n_2$ is the total sample size, we provide an interval for the optimal allocation rate $t$ which minimizes the total sample size $N$. It turns out that for certain distributions, a balanced design is optimal. We give a characterization of these distributions. Furthermore we show that the optimal choice of $t$ depends on the ratio of the two variances which determine the variance of the Wilcoxon-Mann-Whitney statistic under the alternative. This is different from an optimal sample size allocation in case of the normal distribution model.

Rank-based inference methods are applied in various disciplines, typically when procedures relying on standard normal theory are not justifiable, for example when data are not symmetrically distributed, contain outliers, or responses are even measured on ordinal scales. Various specific rank-based methods have been developed for two and more samples, and also for general factorial designs (e.g., Kruskal-Wallis test, Jonckheere-Terpstra test). It is the aim of the present paper (1) to demonstrate that traditional rank-procedures for several samples or general factorial designs may lead to paradoxical results in case of unbalanced samples, (2) to explain why this is the case, and (3) to provide a way to overcome these disadvantages of traditional rankbased inference. Theoretical investigations show that the paradoxical results can be explained by carefully considering the non-centralities of the test statistics which may be non-zero for the traditional tests in unbalanced designs. These non-centralities may even become arbitrarily large for increasing sample sizes in the unbalanced case. A simple solution is the use of socalled pseudo-ranks instead of ranks. As a special case, we illustrate the effects in sub-group analyses which are often used when dealing with rare diseases., Comment: 19 pages, 0 figures

Existing tests for factorial designs in the nonparametric case are based on hypotheses formulated in terms of distribution functions. Typical null hypotheses, however, are formulated in terms of some parameters or effect measures, particularly in heteroscedastic settings. Here this idea is extended to nonparametric models by introducing a novel nonparametric ANOVA-type-statistic based on ranks which is suitable for testing hypotheses formulated in meaningful nonparametric treatment effects in general factorial designs. This is achieved by a careful in-depth study of the common distribution of rank-based estimators for the treatment effects. Since the statistic is asymptotically not a pivotal quantity we propose three different approximation techniques, discuss their theoretic properties and compare them in extensive simulations together with two additionalWald-type tests. An extension of the presented idea to general repeated measures designs is briefly outlined. The proposed rank-based procedures maintain the pre-assigned type-I error rate quite accurately, also in unbalanced and heteroscedastic models.

For general repeated measures designs the Wald-type statistic (WTS) is an asymptotically valid procedure allowing for unequal covariance matrices and possibly non-normal multivariate observations. The drawback of this procedure is the poor performance for small to moderate samples, i.e. decisions based on the WTS may become quite liberal. It is the aim of the present paper to improve its small sample behavior by means of a novel permutation procedure. In particular, it is shown that a permutation version of the WTS inherits its good large sample properties while yielding a very accurate finite sample control of the type-I error as shown in extensive simulations. Moreover, the new permutation method is motivated by a practical data set of a split plot design with a factorial structure on the repeated measures.