Rerandomization is a strategy for improving balance on observed covariates in randomized controlled trials. It has been both advocated and advised against by renowned scholars of experimental design. However, the relationship and differences between stratification, rerandomization, and the combination of the two have not been previously investigated. In this paper, we show that stratified designs can be recreated by rerandomization and explain why, in most cases, stratification on binary covariates followed by rerandomization on continuous covariates is more efficient than rerandomization on all covariates at the same time.

Introduction. This paper presents a study protocol for a comparative effectiveness evaluation of abiraterone acetate against enzalutamide in clinical practice, two cancer drugs given to patients suffering from advanced prostate cancer. Method and analysis. The protocol designs a comparative-effectiveness analysis of abiraterone acetate against enzalutamide. With the substantial number of covariates a two-step procedure is suggested in choosing relevant covariates in the matching design. In the first step, an exploratory factor analysis reduces the dimension of a large set of continuous covariates to nine factors. In the second step, we reduce the dimension of the covariates, interactions and second order terms for the continuous covariates using propensity score estimation. The final design makes use of a genetic matching algorithm. The study protocol provides a detailed statistical analysis plan of the analysis sample derived from the matching design. The analysis will make use of linear regression and robust inference adjusted for multisignificance testing. Discussion. As in a randomised experiment the focus is on the design of the assignment to treatment. This allows the publication of this preanalysis plan before having access to outcome data. This means that the p values will be correct if the maintained assumption of uncounfoundedness is valid. Given that is p-hacking is substantial problem in empirical research, this is a substantial strength of this study. However, while design yields, balance on the observed covariates one cannot discard the possibility that unobserved confounders are not balanced. For that reason, sensitivity tests for the maintained assumption of uncounfoundedness are presented. Ethics and dissemination. The study was approved by the Regional Ethical Review Board in Uppsala, Sweden (Dnr 2017/482). Results will be published in a peer-reviewed journal and distributed to relevant stakeholders in healthcare.

In linear regression of Y on X(∈ R^p) with parameters 𝛽(∈ R^{p+1}), statistical inference is unreliable when observations are obtained from gross-error model, F_{ϵ,G} = (1 − ϵ)F + ϵG, instead of the assumed probability F;G is gross-error probability, 0 < ϵ < 1. Residual’s influence index (RINFIN) at (x, y) is introduced, with components measuring also the local influence of x in the residual and large value flagging a bad leverage case (from G), thus causing unmasking. Large sample properties of RINFIN are presented to confirm significance of the findings, but often the large difference in the RINFIN scores of the data is indicative. RINFIN is successful with microarray data, simulated, high dimensional data and classic regression data sets. RINFIN’s performance improves as p increases and can be used in multiple response linear regression.

Double-blind randomized controlled trials are traditionally seen as the gold standard for causal inferences as the difference-in-means estimator is an unbiased estimator of the average treatment effect in the experiment. The fact that this estimator is unbiased over all possible randomizations does not, however, mean that any given estimate is close to the true treatment effect. Similarly, while predetermined covariates will be balanced between treatment and control groups on average, large imbalances may be observed in a given experiment and the researcher may therefore want to condition on such covariates using linear regression. This article studies the theoretical properties of both the difference-in-means and OLS estimators conditional on observed differences in covariates. By deriving the statistical properties of the conditional estimators, we can establish guidance for how to deal with covariate imbalances.

In a Data-Generating Experiment, the observed sample, x, has intractable or unavailable c.d.f., F_θ, and θ’s statistical nature is unknown; θ is element of metric space (Θ, dΘ). Matching estimates of θ are introduced, learning from the “best” x-matches with samples X^∗ from F_{θ^∗}, θ^∗ ∈ Θ. Under mild conditions, these nonparametric estimates are uniformly consistent and the upper bounds on their rates of convergence in probability have the same rate and depend on the Kolmogorov entropies of an increasing sequence of sets covering Θ. When Θ ⊆ Rm and the observations are i.i.d. the upper bounds can be, \sqrt{log n}/\sqrt{n} when m is known, and \sqrt{m_n·log n}/\sqrt{n} when m is unknown; m ≥ 1, m_n ↑ ∞ at a desired rate. Upper bounds can also be obtained for dependent observations. These rates hold for observations in R^d, complementing recent results obtained for real, i.i.d. observations, under stronger assumptions and using weak probability distances; d ≥ 1. In simulations, the Matching estimates are successful for the mixture of 2 normals and for Tukey’s (a, b, g, h) and the (a, b, g, k) models. Computers’ evolution will allow for more and faster comparisons, resulting in improved Matching estimates for universal use in Machine Learning.