Conditional Randomization Tests Research Articles

Multiple conditional randomization tests for lagged and spillover treatment effects

Abstract We consider the problem of constructing multiple independent conditional randomization tests using a single dataset. Because the tests are independent, the randomization p-values can be interpreted individually and combined using standard methods for multiple testing. We give a simple, sequential construction of such tests and then discuss its application to three problems: Rosenbaum’s evidence factors for observational studies, lagged treatment effects in stepped-wedge trials, and spillover effects in randomized trials with interference. We compare the proposed approach with some existing methods using simulated and real datasets. Finally, we establish a more general sufficient condition for independent conditional randomization tests.

Using Machine Learning to Test Causal Hypotheses in Conjoint Analysis

Abstract Conjoint analysis is a popular experimental design used to measure multidimensional preferences. Many researchers focus on estimating the average marginal effects of each factor while averaging over the other factors. Although this allows for straightforward design-based estimation, the results critically depend on the ways in which factors interact with one another. An alternative model-based approach can compute various quantities of interest, but requires correct model specifications, a challenging task for conjoint analysis with many factors. We propose a new hypothesis testing approach based on the conditional randomization test (CRT) to answer the most fundamental question of conjoint analysis: Does a factor of interest matter in any way given the other factors? Although it only provides a formal test of these binary questions, the CRT is solely based on the randomization of factors, and hence requires no modeling assumption. This means that the CRT can provide a powerful and assumption-free statistical test by enabling the use of any test statistic, including those based on complex machine learning algorithms. We also show how to test commonly used regularity assumptions. Finally, we apply the proposed methodology to conjoint analysis of immigration preferences. An open-source software package is available for implementing the proposed methodology. The proposed methodology is implemented via an open-source software R package CRTConjoint, available through the Comprehensive R Archive Network https://cran.r-project.org/web/packages/CRTConjoint/index.html.

Nearest-Neighbor Sampling Based Conditional Independence Testing

The conditional randomization test (CRT) was recently proposed to test whether two random variables X and Y are conditionally independent given random variables Z. The CRT assumes that the conditional distribution of X given Z is known under the null hypothesis and then it is compared to the distribution of the observed samples of the original data. The aim of this paper is to develop a novel alternative of CRT by using nearest-neighbor sampling without assuming the exact form of the distribution of X given Z. Specifically, we utilize the computationally efficient 1-nearest-neighbor to approximate the conditional distribution that encodes the null hypothesis. Then, theoretically, we show that the distribution of the generated samples is very close to the true conditional distribution in terms of total variation distance. Furthermore, we take the classifier-based conditional mutual information estimator as our test statistic. The test statistic as an empirical fundamental information theoretic quantity is able to well capture the conditional-dependence feature. We show that our proposed test is computationally very fast, while controlling type I and II errors quite well. Finally, we demonstrate the efficiency of our proposed test in both synthetic and real data analyses.

Hypothesis testing in adaptively sampled data: ART to maximize power beyond iid sampling

Testing whether a variable of interest affects the outcome is one of the most fundamental problems in statistics and is often the main scientific question of interest. To tackle this problem, the conditional randomization test (CRT) is widely used to test the independence of variable(s) of interest (X) with an outcome (Y) holding other variable(s) (Z) fixed. The CRT uses “Model-X” inference framework that relies solely on the iid sampling of (X, Z) to produce exact finite-sample p values that are constructed using any test statistic. We propose a new method, the adaptive randomization test (ART), that tackles the same independence problem while allowing the data to be adaptively sampled. Like the CRT, the ART relies solely on knowing the (adaptive) sampling distribution of (X, Z). Although the ART allows practitioners to flexibly design and analyze adaptive experiments, the method itself does not guarantee a powerful adaptive sampling procedure. For this reason, we show substantial power gains obtained from adaptively sampling compared to the typical iid sampling procedure in a multi-arm bandit setting and an application in conjoint analysis. We believe that the proposed adaptive procedure is successful because it takes arms that may initially look like “fake” signals due to random chance and stabilizes them closer to “null” signals and samples more/less from signal/null arms.

Learning to increase the power of conditional randomization tests

The model-X conditional randomization test is a generic framework for conditional independence testing, unlocking new possibilities to discover features that are conditionally associated with a response of interest while controlling type-I error rates. An appealing advantage of this test is that it can work with any machine learning model to design powerful test statistics. In turn, the common practice in the model-X literature is to form a test statistic using machine learning models, trained to maximize predictive accuracy with the hope to attain a test with good power. However, the ideal goal here is to drive the model (during training) to maximize the power of the test, not merely the predictive accuracy. In this paper, we bridge this gap by introducing, for the first time, novel model-fitting schemes that are designed to explicitly improve the power of model-X tests. This is done by introducing a new cost function that aims at maximizing the test statistic used to measure violations of conditional independence. Using synthetic and real data sets, we demonstrate that the combination of our proposed loss function with various base predictive models (lasso, elastic net, and deep neural networks) consistently increases the number of correct discoveries obtained, while maintaining type-I error rates under control.

On the power of conditional independence testing under model-X

For testing conditional independence (CI) of a response Y and a predictor X given covariates Z, the model-X (MX) framework has been the subject of active methodological research, especially in the context of MX knockoffs and their application to genome-wide association studies. In this paper, we study the power of MX CI tests, yielding quantitative insights into the role of machine learning and providing evidence in favor of using likelihood-based statistics in practice. Focusing on the conditional randomization test (CRT), we find that its conditional mode of inference allows us to reformulate it as testing a point null hypothesis involving the conditional distribution of X. The Neyman-Pearson lemma implies that a likelihood-based statistic yields the most powerful CRT against a point alternative. We obtain a related optimality result for MX knockoffs. Switching to an asymptotic framework with arbitrarily growing covariate dimension, we derive an expression for the power of the CRT against local semiparametric alternatives in terms of the prediction error of the machine learning algorithm on which its test statistic is based. Finally, we exhibit a resampling-free test with uniform asymptotic Type-I error control under the assumption that only the first two moments of X given Z are known.

A Graph-Theoretic Approach to Randomization Tests of Causal Effects under General Interference

AbstractInterference exists when a unit's outcome depends on another unit's treatment assignment. For example, intensive policing on one street could have a spillover effect on neighbouring streets. Classical randomization tests typically break down in this setting because many null hypotheses of interest are no longer sharp under interference. A promising alternative is to instead construct a conditional randomization test on a subset of units and assignments for which a given null hypothesis is sharp. Finding these subsets is challenging, however, and existing methods are limited to special cases or have limited power. In this paper, we propose valid and easy-to-implement randomization tests for a general class of null hypotheses under arbitrary interference between units. Our key idea is to represent the hypothesis of interest as a bipartite graph between units and assignments, and to find an appropriate biclique of this graph. Importantly, the null hypothesis is sharp within this biclique, enabling conditional randomization-based tests. We also connect the size of the biclique to statistical power. Moreover, we can apply off-the-shelf graph clustering methods to find such bicliques efficiently and at scale. We illustrate our approach in settings with clustered interference and show advantages over methods designed specifically for that setting. We then apply our method to a large-scale policing experiment in Medellín, Colombia, where interference has a spatial structure.

A high-dimensional power analysis of the conditional randomization test and knockoffs

Summary In many scientific applications, researchers aim to relate a response variable $Y$ to a set of potential explanatory variables $X = (X_1,\dots,X_p)$, and start by trying to identify variables that contribute to this relationship. In statistical terms, this goal can be understood as trying to identify those $X_j$ on which $Y$ is conditionally dependent. Sometimes it is of value to simultaneously test for each $j$, which is more commonly known as variable selection. The conditional randomization test, CRT, and model-X knockoffs are two recently proposed methods that respectively perform conditional independence testing and variable selection by computing, for each $X_j$, any test statistic on the data and assessing that test statistic’s significance, by comparing it with test statistics computed on synthetic variables generated using knowledge of the distribution of $X$. The main contribution of this article is the analysis of the power of these methods in a high-dimensional linear model, where the ratio of the dimension $p$ to the sample size $n$ converges to a positive constant. We give explicit expressions for the asymptotic power of the CRT, variable selection with CRT $p$-values, and model-X knockoffs, each with a test statistic based on the marginal covariance, the least squares coefficient or the lasso. One useful application of our analysis is direct theoretical comparison of the asymptotic powers of variable selection with CRT $p$-values and model-X knockoffs; in the instances with independent covariates that we consider, the CRT probably dominates knockoffs. We also analyse the power gain from using unlabelled data in the CRT when limited knowledge of the distribution of $X$ is available, as well as the power of the CRT when samples are collected retrospectively.

Fast and powerful conditional randomization testing via distillation.

We consider the problem of conditional independence testing: given a response and covariates , we test the null hypothesis that . The conditional randomization test was recently proposed as a way to use distributional information about to exactly and nonasymptotically control Type-I error using any test statistic in any dimensionality without assuming anything about . This flexibility, in principle, allows one to derive powerful test statistics from complex prediction algorithms while maintaining statistical validity. Yet the direct use of such advanced test statistics in the conditional randomization test is prohibitively computationally expensive, especially with multiple testing, due to the requirement to recompute the test statistic many times on resampled data. We propose the distilled conditional randomization test, a novel approach to using state-of-the-art machine learning algorithms in the conditional randomization test while drastically reducing the number of times those algorithms need to be run, thereby taking advantage of their power and the conditional randomization test's statistical guarantees without suffering the usual computational expense. In addition to distillation, we propose a number of other tricks, like screening and recycling computations, to further speed up the conditional randomization test without sacrificing its high power and exact validity. Indeed, we show in simulations that all our proposals combined lead to a test that has similar power to the most powerful existing conditional randomization test implementations, but requires orders of magnitude less computation, making it a practical tool even for large datasets. We demonstrate these benefits on a breast cancer dataset by identifying biomarkers related to cancer stage.

The Holdout Randomization Test for Feature Selection in Black Box Models

We propose the holdout randomization test (HRT), an approach to feature selection using black box predictive models. The HRT is a specialized version of the conditional randomization test (CRT) (Candes et al., 2018) that uses data splitting for feasible computation. The HRT works with any predictive model and produces a valid p-value for each feature. To make the HRT more practical, we propose a set of extensions to maximize power and speed up computation. In simulations, these extensions lead to greater power than a competing knockoffs-based approach, without sacrificing control of the error rate. We apply the HRT to two case studies from the scientific literature where heuristics were originally used to select important features for predictive models. The results illustrate how such heuristics can be misleading relative to principled methods like the HRT. Code is available at https://github.com/tansey/hrt.

Randomization tests for multiarmed randomized clinical trials.

We examine the use of randomization-based inference for analyzing multiarmed randomized clinical trials, including the application of conditional randomization tests to multiple comparisons. The view is taken that the linkage of the statistical test to the experimental design (randomization procedure) should be recognized. A selected collection of randomization procedures generalized to multiarmed treatment allocation is summarized, and generalizations for two randomization procedures that heretofore were designed for only two treatments are developed. We explain the process of computing the randomization test and conditional randomization test via Monte Carlo simulation, developing an efficient algorithm that makes multiple comparisons possible that would not be possible using a standard algorithm, demonstrate the preservation of type I error rate, and explore the relationship of statistical power to the randomization procedure in the presence of a time trend and outliers. We distinguish between the interpretation of the p-value in the randomization test and in the population test and verify that the randomization test can be approximated by the population test on some occasions. Data from two multiarmed clinical trials from the literature are reanalyzed to illustrate the methodology.

The Conditional Permutation Test for Independence While Controlling for Confounders

SummaryWe propose a general new method, the conditional permutation test, for testing the conditional independence of variables X and Y given a potentially high dimensional random vector Z that may contain confounding factors. The test permutes entries of X non-uniformly, to respect the existing dependence between X and Z and thus to account for the presence of these confounders. Like the conditional randomization test of Candès and co-workers in 2018, our test relies on the availability of an approximation to the distribution of X|Z—whereas their test uses this estimate to draw new X-values, for our test we use this approximation to design an appropriate non-uniform distribution on permutations of the X-values already seen in the true data. We provide an efficient Markov chain Monte Carlo sampler for the implementation of our method and establish bounds on the type I error in terms of the error in the approximation of the conditional distribution of X|Z, finding that, for the worst-case test statistic, the inflation in type I error of the conditional permutation test is no larger than that of the conditional randomization test. We validate these theoretical results with experiments on simulated data and on the Capital Bikeshare data set.

Randomization Tests that Condition on Non-Categorical Covariate Balance

Abstract A benefit of randomized experiments is that covariate distributions of treatment and control groups are balanced on average, resulting in simple unbiased estimators for treatment effects. However, it is possible that a particular randomization yields covariate imbalances that researchers want to address in the analysis stage through adjustment or other methods. Here we present a randomization test that conditions on covariate balance by only considering treatment assignments that are similar to the observed one in terms of covariate balance. Previous conditional randomization tests have only allowed for categorical covariates, while our randomization test allows for any type of covariate. Through extensive simulation studies, we find that our conditional randomization test is more powerful than unconditional randomization tests and other conditional tests. Furthermore, we find that our conditional randomization test is valid (1) unconditionally across levels of covariate balance, and (2) conditional on particular levels of covariate balance. Meanwhile, unconditional randomization tests are valid for (1) but not (2). Finally, we find that our conditional randomization test is similar to a randomization test that uses a model-adjusted test statistic.

Erratum to: A Conditional Randomization Test for Covariate Imbalance

Erratum to: A Conditional Randomization Test for Covariate Imbalance

A Conditional Randomization Test to Account for Covariate Imbalance in Randomized Experiments

Abstract We consider the conditional randomization test as a way to account for covariate imbalance in randomized experiments. The test accounts for covariate imbalance by comparing the observed test statistic to the null distribution of the test statistic conditional on the observed covariate imbalance. We prove that the conditional randomization test has the correct significance level and introduce original notation to describe covariate balance more formally. Through simulation, we verify that conditional randomization tests behave like more traditional forms of covariate adjustment but have the added benefit of having the correct conditional significance level. Finally, we apply the approach to a randomized product marketing experiment where covariate information was collected after randomization.

Conditional Monte Carlo randomization tests for regression models.

We discuss the computation of randomization tests for clinical trials of two treatments when the primary outcome is based on a regression model. We begin by revisiting the seminal paper of Gail, Tan, and Piantadosi (1988), and then describe a method based on Monte Carlo generation of randomization sequences. The tests based on this Monte Carlo procedure are design based, in that they incorporate the particular randomization procedure used. We discuss permuted block designs, complete randomization, and biased coin designs. We also use a new technique by Plamadeala and Rosenberger (2012) for simple computation of conditional randomization tests. Like Gail, Tan, and Piantadosi, we focus on residuals from generalized linear models and martingale residuals from survival models. Such techniques do not apply to longitudinal data analysis, and we introduce a method for computation of randomization tests based on the predicted rate of change from a generalized linear mixed model when outcomes are longitudinal. We show, by simulation, that these randomization tests preserve the size and power well under model misspecification.

Shift in re‐randomization distribution with conditional randomization test

Proschan, Brittain, and Kammerman made a very interesting observation that for some examples of the unequal allocation minimization, the mean of the unconditional randomization distribution is shifted away from 0. Kuznetsova and Tymofyeyev linked this phenomenon to the variations in the allocation ratio from allocation to allocation in the examples considered in the paper by Proschan et al. and advocated the use of unequal allocation procedures that preserve the allocation ratio at every step. In this paper, we show that the shift phenomenon extends to very common settings: using conditional randomization test in a study with equal allocation. This phenomenon has the same cause: variations in the allocation ratio among the allocation sequences in the conditional reference set, not previously noted. We consider two kinds of conditional randomization tests. The first kind is the often used randomization test that conditions on the treatment group totals; we describe the variations in the conditional allocation ratio with this test on examples of permuted block randomization and biased coin randomization. The second kind is the randomization test proposed by Zheng and Zelen for a multicenter trial with permuted block central allocation that conditions on the within-center treatment totals. On the basis of the sequence of conditional allocation ratios, we derive the value of the shift in the conditional randomization distribution for specific vector of responses and the expected value of the shift when responses are independent identically distributed random variables. We discuss the asymptotic behavior of the shift for the two types of tests.

A General Method for Detecting Interference Between Units in Randomized Experiments

Interference between units may pose a threat to unbiased causal inference in randomized controlled experiments. Although the assumption of no interference is often necessary for causal inference, few options are available for testing this assumption. This article presents an ex post method for detecting interference between units in randomized experiments. With a test statistic of the analyst’s choice, a conditional randomization test allows for the calculation of the exact significance level of the causal dependence of outcomes on the treatment status of other units. The robustness of the method is demonstrated through simulation studies. Moreover, using this method, interference between units is detected in a field experiment designed to assess the effect of mailings on voter turnout in a U.S. primary election.

Sequential monitoring with conditional randomization tests

Sequential monitoring in clinical trials is often employed to allow for early stopping and other interim decisions, while maintaining the type I error rate. However, sequential monitoring is typically described only in the context of a population model. We describe a computational method to implement sequential monitoring in a randomization-based context. In particular, we discuss a new technique for the computation of approximate conditional tests following restricted randomization procedures and then apply this technique to approximate the joint distribution of sequentially computed conditional randomization tests. We also describe the computation of a randomization-based analog of the information fraction. We apply these techniques to a restricted randomization procedure, Efron's [Biometrika 58 (1971) 403--417] biased coin design. These techniques require derivation of certain conditional probabilities and conditional covariances of the randomization procedure. We employ combinatoric techniques to derive these for the biased coin design.

Sequential Monitoring of Conditional Randomization Tests: Generalized Biased Coin Designs

For the generalized biased coin class of randomization procedures, Smythe (1988) proved asymptotic normality of the conditional linear rank test. Clinical trialists often undertake interim analysis to determine whether to stop the trial early for a substantial treatment effect. In this article, we will set up one interim analysis using a conditional randomization test. The joint asymptotic distribution of the interim test statistic and the final test statistic will be explored. We also define the concept of conditional information under a randomization model.