Policy Evaluation Placebo Test
Also known as: placebo test, falsification test, fake treatment test, placebo regression
A policy evaluation placebo test is a falsification check used in quasi-experimental research to validate a causal identification strategy. The researcher applies the same estimation method to a pseudo-treatment — a time period, group, or outcome where the real policy could not have had an effect — and checks that no spurious effect is detected. A null placebo result builds confidence that the main estimate reflects a genuine causal impact rather than bias or confounding.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use a placebo test whenever you report a causal estimate from quasi-experimental data and you want to validate your identification assumptions. It is particularly important for difference-in-differences (pre-trend placebo), regression discontinuity (outcome or bandwidth placebo), and synthetic control designs. Do not use placebo tests as a substitute for credible design — they are a complement, not a guarantee. Avoid placing excessive weight on a single placebo test, since any single check has limited power, and a passing placebo does not prove causality but merely fails to reject the design.
Strengths & limitations
- Provides a falsifiable, data-driven check on identification assumptions without requiring additional assumptions beyond those already in the main model.
- Can detect common threats — such as differential pre-trends, manipulation near a cutoff, or correlated shocks — that are otherwise invisible in the main estimate.
- Increases credibility and transparency of the causal claim; journals and peer reviewers routinely expect placebo evidence in applied microeconomics and policy research.
- Multiple placebo variants (temporal, geographic, outcome) can triangulate validity from different angles.
- Straightforward to implement: the researcher reruns the same code with a modified treatment indicator or sample restriction.
- A passing placebo test does not guarantee the causal estimate is unbiased; it only rules out specific, testable forms of confounding.
- Some placebo designs have low statistical power, especially in small samples, meaning a failure to reject zero could reflect insufficient power rather than a valid design.
- Outcome placebos require strong a priori knowledge that the chosen outcome is truly unaffected, which is not always available.
- In staggered or heterogeneous treatment designs, constructing a well-defined placebo period can be complicated.
Frequently asked
What is the difference between a placebo test and a robustness check?
A robustness check re-estimates the main effect under alternative model specifications or sample restrictions to see if the result changes. A placebo test reassigns treatment to a context where the true effect should be zero, to check whether the estimator produces a spurious effect. Both are specification checks, but placebo tests are specifically designed as falsification exercises.
What does a significant placebo result imply?
It implies that the identification strategy is detecting something other than the policy effect — possibly a violation of parallel trends, sorting around a cutoff, or a correlated shock. A significant placebo result is a warning sign that the main causal estimate may be biased and the design should be revisited.
How many placebo tests should I run?
At least one pre-treatment temporal placebo (if panel data are available) and, where feasible, one outcome or geographic placebo. Running many placebo regressions simultaneously raises the risk of false positives by chance; if you run multiple tests, adjust for multiple comparisons or report all of them transparently.
Can a placebo test be applied to randomised controlled trials?
Placebo tests are primarily used in observational and quasi-experimental settings where identification assumptions cannot be verified by design. In a well-randomised RCT, the need for a placebo test is much reduced because random assignment already controls for confounding. They are occasionally used in RCTs to check balance on pre-treatment outcomes.
What sample size is needed for a placebo test to be informative?
The placebo test needs enough statistical power to detect a non-trivial spurious effect if one exists. As a rough guideline, you need at least as many observations as the main analysis. Very small samples — fewer than 30 to 40 units — may produce uninformative (low-power) placebo tests even when the design is flawed.
Sources
- Imbens, G. W., & Wooldridge, J. M. (2009). Recent Developments in the Econometrics of Program Evaluation. Journal of Economic Literature, 47(1), 5-86. DOI: 10.1257/jel.47.1.5 ↗
- Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How Much Should We Trust Differences-in-Differences Estimates? Quarterly Journal of Economics, 119(1), 249-275. DOI: 10.1162/003355304772839588 ↗
How to cite this page
ScholarGate. (2026, June 3). Policy Evaluation Placebo Test. ScholarGate. https://scholargate.app/en/causal-inference/policy-evaluation-placebo-test
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Difference-in-DifferencesEconometrics↔ compare
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Permutation TestStatistics↔ compare
- Synthetic Control MethodCausal inference↔ compare