Rerandomization to improve covariate balance in experiments

By Kari Lock Morgan, Donald B. Rubin. 2012

374 citations recorded by the source corpus.

Randomized experiments are the "gold standard" for estimating causal effects, yet often in practice, chance imbalances exist in covariate distributions between treatment groups. If covariate data are available before units are exposed to treatments, these chance imbalances can be mitigated by first checking covariate balance before the physical experiment takes place. Provided a precise definition of imbalance has been specified in advance, unbalanced randomizations can be discarded, followed by a rerandomization, and this process can continue until a randomization yielding balance according to the definition is achieved. By improving covariate balance, rerandomization provides more precise and trustworthy estimates of treatment effects.

Subfields: Statistics and Probability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Statistics and Probability

Citations in the corpus

The Econometrics of Randomized Experiments (2016, background): “ion in the empirical development literature. One paper that raised the question forcefully in this literature is Bruhn and McKenzie (2007). Theoretical paper discussing some of the formal aspects are Morgan and Rubin (2012) and Banerjee, Snowberg, and Chassang (2016). Here we oﬀer some comments. Fir”
Covariate-adjusted Fisher randomization tests for the average treatment effect (2020, methodology): “Negi and Wooldridge (2020) reviewed the asymptotic distributions of the estimators: √ N(τ̂∗ − τ) N (0, v∗) (∗ = n,r, f, l), with vn = p −1 1 σ 2 1 + p −1 0 σ 2 0, vr = vf = p −1 1 σ 2 a(1) + p −1 0 σ 2 a(0), vl = p −1 1 σ 2 b(1) + p −1 0 σ 2 b(0) + ∆x̄, (5) where ∆x̄ = (γ1−γ0)σ(2) x(γ1−γ0).”
Optimal a priori balance in the design of controlled experiments (2013, methodology): “6) then fixing t and re-randomizing until M2 [7](W ) ≤ t yields a reduction in variance relative to complete randomization that is constant over n: 1−Var(τ̂)/Var(τ̂) = η(1−Var ( i) /Var (Yi1)), η ∈ (0, 1) constant over n.”
Why Experimenters Might Not Always Want to Randomize, and What They Could Do Instead (2016, background): “Researchers conducting such field experiments in political science (as well as in fields such as economics, medicine, and other social and biomedical sciences) are often confronted with variants of the following situation (cf. Morgan and Rubin 2012).”
The Impact of Cash and Food Transfers: Evidence from a Randomized Intervention in Niger (2018, neutral): “As Morgan and Rubin (2012) point out, with a modest number of covariates, say 10, and a set equal to 0.”
Optimality of Matched-Pair Designs in Randomized Controlled Trials (2022, methodology): “…commonly-used tests including the two-sample t-test (Riach and Rich, 2002; Gelman and Hill, 2006; Duflo et al., 2007) and the “matched pairs” t-test (Moses, 2006; Hsu and Lachenbruch, 2007; Armitage et al., 2008; Imbens and Rubin, 2015; Athey and Imbens, 2017), the test statistics are studentized…”
Regression-adjusted average treatment effect estimates in stratified randomized experiments (2019, background): “Even when carefully designed, the probability of some baseline covariates exhibiting imbalance between treatment and control groups is high [8, 25].”