REVIEW 6 cited by
Data-Driven Switchback Experiments: Theoretical Tradeoffs and Empirical Bayes Designs
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
We study the design and analysis of switchback experiments conducted on a single aggregate unit. The design problem is to partition the continuous time space into intervals and switch treatments between intervals, in order to minimize the estimation error of the treatment effect. We show that the estimation error depends on four factors: carryover effects, periodicity, serially correlated outcomes, and impacts from simultaneous experiments. We derive a rigorous bias-variance decomposition and show the tradeoffs of the estimation error from these factors. The decomposition provides three new insights in choosing a design: First, balancing the periodicity between treated and control intervals reduces the variance; second, switching less frequently reduces the bias from carryover effects while increasing the variance from correlated outcomes, and vice versa; third, randomizing interval start and end points reduces both bias and variance from simultaneous experiments. Combining these insights, we propose a new empirical Bayes design approach. This approach uses prior data and experiments for designing future experiments. We illustrate this approach using real data from a ride-sharing platform, yielding a design that reduces MSE by 33% compared to the status quo design used on the platform.
Forward citations
Cited by 6 Pith papers
-
Balancing Interference and Correlation in Spatial Experimental Designs: A Causal Graph Cut Approach
A surrogate for the ATE estimator's MSE is optimized with spectral graph cuts to produce cluster-randomized designs that adapt to the spatial covariance and accommodate moderate-to-large interference.
-
Experimental Design When N Equals One
Under a finite-lag impulse-response model, optimal Markov treatment designs are characterized asymptotically: Bernoulli(1/2) for lag-specific and robust targets, and cycle length near K−1+√(K−1) for cumulative effects.
-
Power-Optimal Covariate Adjustment for Switchback Experiments
Training the CUPAC covariate and its regression coefficient with a between-cell-weighted loss improves switchback estimator power, with gains concentrated in within-noise-dominated regimes.
-
Calibrated Horizon-Weighted Local Projection Designs for Markov Switchbacks
Optimal Markov switchback persistence for a horizon-weighted local-projection target has a closed form under a balanced homoskedastic AR(1) assignment benchmark, and field designs should replace that formula with cali...
-
Robust and efficient multiple-unit switchback experimentation
Regular Balanced Switchback Designs combine item- and time-randomization with balanced treatment counts, yielding unbiased, lower-variance estimates of average treatment effects that are robust to carryover effects.
-
A Two-armed Bandit Framework for A/B Testing
A two-armed bandit based test statistic with permutation aggregation improves power for A/B testing in both i.i.d. and dynamic settings.
Discussion (0). Continue with ORCID to comment.