How Does Anticipation of a Policy or Event Affect a Natural Experiment?

What Happens When People Respond to a Policy Before It Takes Effect?

A government announces that university scholarships will become available next academic year. Students who would otherwise enroll immediately may postpone enrollment to qualify. A school district announces an educational technology program, prompting teachers to begin training before the official rollout. A university announces new admission requirements, and prospective applicants change their behavior months before implementation.

In each situation, the policy may influence outcomes before the date researchers initially identify as the beginning of treatment.

This creates a problem for natural experiments. Researchers often distinguish observations before and after implementation, treating the earlier period as an untreated baseline. But what if the baseline already contains behavioral responses to the forthcoming intervention?

The resulting estimates may misrepresent the intervention's timing, magnitude, or causal effect. Understanding anticipation is therefore essential when policies are announced before implementation or when affected populations can reasonably predict what will happen.

Anticipation Can Make the Pre-Intervention Period Partially Treated

How Anticipation Changes the Logic of Natural-Experiment Identification

What Is an Anticipation Effect?

An anticipation effect occurs when expectations about a future intervention influence behavior or outcomes before the intervention's formal implementation.

Such effects may arise because people receive advance notice, infer that a policy is likely to occur, or prepare for a predictable institutional change.

For example, students expecting a new scholarship may delay enrollment. Teachers anticipating a digital assessment system may begin changing instructional practices. Universities expecting revised funding rules may alter expenditure decisions before the rules become effective.

Anticipation is not necessarily irrational or undesirable. It may represent a sensible response to information about future conditions. The methodological problem arises when researchers classify the affected period as untreated despite the policy already influencing behavior.

In a natural experiment, identifying the actual exposure-generating process therefore requires attention to expectations as well as formal implementation.

Why Is the Official Implementation Date Sometimes Misleading?

Policy evaluation often involves several dates that serve different purposes.

These milestones may coincide, but they often do not. Researchers should determine which date is causally relevant to their outcome and estimand.

If the research question concerns the effect of announcing a policy, the announcement itself may be the treatment. If the question concerns the effect of actual service provision, anticipatory behavior may represent a separate response that must be considered when defining the counterfactual.

What Is the No-Anticipation Assumption?

The no-anticipation assumption generally states that future treatment does not affect potential outcomes before the relevant treatment period.

Its precise formulation varies across causal designs, especially when treatment timing differs across units.

In a staggered-adoption difference-in-differences setting, researchers may assume that units scheduled to receive treatment at a later date have not yet experienced effects of that treatment during designated pre-treatment periods.

If those units begin responding before adoption, their supposedly untreated outcomes may already reflect the forthcoming intervention.

The assumption matters because the observed pre-treatment period is often used to estimate untreated trends, establish baseline relationships, or construct comparison groups.

Researchers should distinguish the absence of actual service delivery from the absence of any causal effect of future policy exposure.

How Can Anticipation Bias Difference-in-Differences Estimates?

Difference-in-differences compares changes in treated and comparison groups under assumptions about their untreated outcome trajectories.

Suppose a scholarship policy takes effect in August, but students in the affected region begin postponing enrollment in April.

If researchers use July as the pre-treatment baseline, enrollment may already be unusually low because of anticipation. Comparing July with September could then exaggerate the apparent increase associated with implementation.

Alternatively, if institutions begin improving services before the official launch, some benefits may already be incorporated into the baseline, reducing the estimated post-implementation change.

Thus, anticipation does not necessarily bias estimates in one direction. Its consequences depend on the behavior involved, the outcome, and the chosen comparison periods.

Researchers applying difference-in-differences should investigate whether future treatment affected outcomes during periods classified as untreated.

Can Anticipation Produce Apparent Pre-Trends in Event-Study Analyses?

Yes. Event-study models often estimate outcome differences at several periods before and after treatment.

Estimated effects in pre-treatment periods may indicate anticipatory responses, but they may also reflect differential untreated trends, confounding, model misspecification, or sampling variation.

Consequently, statistically significant pre-treatment coefficients should not automatically be interpreted as anticipation.

Conversely, nonsignificant pre-treatment coefficients do not establish that anticipation was absent. Pre-trend tests may have limited statistical power, and some estimators can be misleading when treatment timing and effects vary across groups.

Researchers should interpret event-study patterns alongside policy documentation, substantive knowledge, and the assumptions of the estimator.

The broader parallel-trends assumption remains distinct from no anticipation. Both may be required for a particular causal interpretation.

How Does Anticipation Affect Interrupted Time-Series Designs?

Interrupted time-series (ITS) designs examine whether an outcome's level or trajectory changes around an intervention.

When behavior changes before the formal intervention date, the observed series may begin departing from its historical trajectory during the announcement or preparation period.

A segmented regression using only the implementation date may then assign part of the response to the pre-intervention trend or estimate an incorrectly timed level change.

For example, a university announces a mandatory digital attendance system in January but activates it in March. Teachers may begin monitoring attendance more closely during February.

An ITS model that treats February as entirely untreated may mischaracterize the intervention's effect.

Researchers may consider an announcement interruption, a transition period, or a model allowing gradual effects when the data and intervention mechanism support these specifications.

However, selecting whichever interruption date produces the largest effect after inspecting the outcomes risks analytical selection bias.

Does Anticipation Threaten Regression Discontinuity?

Potentially, although the mechanism differs from time-based designs.

In regression discontinuity, treatment assignment changes at a threshold. If individuals know the rule and can influence their running-variable values, they may strategically position themselves on the preferred side of the cutoff.

For example, schools anticipating a grant eligibility threshold may alter reported infrastructure scores to qualify.

Such manipulation can undermine the continuity assumptions supporting the local comparison.

Anticipation can also change behavior before treatment receipt without necessarily changing assignment. Researchers should distinguish strategic sorting around the threshold from other responses to expected eligibility.

The relevant diagnostics depend on the integrity of the cutoff-based assignment mechanism.

How Can Researchers Establish When Anticipation Began?

The earliest plausible behavioral response may precede the official announcement, particularly when policy changes were publicly discussed or widely expected.

Researchers should reconstruct the information environment using policy proposals, official announcements, institutional communications, implementation memoranda, and other reliable records.

They should also consider whether the affected population had sufficient information and opportunity to respond.

For example, a scholarship announced two months before an application deadline may plausibly influence enrollment decisions. A policy announced after all relevant enrollment decisions were finalized may be less likely to affect that cohort's earlier behavior.

The timing argument should therefore connect information availability to a specific behavioral mechanism.

Researchers should avoid assuming that anticipation began on the earliest conceivable date merely because the policy had been discussed publicly.

Can Researchers Simply Remove Anticipation Periods?

Excluding periods immediately before implementation can sometimes reduce contamination of the untreated baseline, but it is not a universal solution.

Such an exclusion may be justified when a credible anticipation window can be identified and enough uncontaminated observations remain.

However, removing observations can reduce statistical power, weaken trend estimation, and alter the target estimand.

If anticipation began earlier than the excluded window, contamination may remain. If the window is selected after inspecting outcome changes, the exclusion may introduce additional analytical bias.

Researchers should justify the window using policy timing and behavioral evidence, then evaluate sensitivity to plausible alternatives.

Should Announcement and Implementation Effects Be Estimated Separately?

Sometimes. Announcement and implementation may represent distinct causal exposures.

A policy announcement can influence expectations and decisions even before resources or services become available. Implementation may then produce additional effects through actual access or enforcement.

Researchers may wish to estimate the total effect of the policy process, the effect of announcement, or the incremental effect of implementation after announcement.

These are different estimands. Separating them requires adequate variation, suitable observations, and assumptions that distinguish their effects.

If announcement and implementation occur close together or coincide with other changes, the data may not support separate identification.

Researchers should not interpret coefficients from a model containing two policy dates as distinct causal effects unless the necessary counterfactual assumptions are defensible.

How Can Researchers Evaluate Anticipation Empirically?

Empirical assessment may involve examining outcome trajectories around announcements, studying intermediate behaviors, comparing populations with different information exposure, and conducting sensitivity analyses.

For example, a scholarship evaluation might examine application submissions, enrollment deferrals, and inquiries about eligibility before implementation.

Such outcomes may provide evidence that students responded to the forthcoming policy.

Researchers may also compare estimates under alternative, substantively justified anticipation windows.

However, detecting a pre-implementation change does not by itself establish that anticipation caused it. Other events or trends may explain the same pattern.

Evidence is strongest when a documented information event, plausible behavioral mechanism, and compatible outcome changes support the same interpretation.

What If Anticipation Cannot Be Ruled Out?

Researchers should determine whether the intended estimand remains identifiable under plausible anticipation scenarios.

Some studies may estimate the effect of the entire policy process beginning with announcement rather than the isolated effect of formal implementation.

Others may restrict the analysis to periods before expectations plausibly formed, provided the resulting comparison remains informative.

Where anticipation timing is uncertain, sensitivity analyses can examine how conclusions change under alternative assumptions.

If the data cannot distinguish anticipation from preexisting trends or concurrent events, causal claims should be qualified accordingly.

The inability to exclude anticipation does not make the research worthless, but it may prevent the precise causal interpretation originally intended.

Evaluating a Scholarship Policy Announced Before Implementation

Suppose enrollment falls during the announcement period and rises sharply in August. A simple comparison of July and August could suggest a large positive policy effect.

However, some of the August increase may consist of students who would have enrolled earlier without the scholarship. The policy may have shifted enrollment timing rather than increasing the total number of students entering university.

A credible evaluation would therefore examine cumulative enrollment over an appropriate period, not merely the immediate post-implementation increase.

The example demonstrates how anticipation can alter both the counterfactual and the substantive meaning of the estimated effect.

Common Misconceptions About Policy Anticipation

How to Handle Anticipation When Designing a Natural-Experiment Evaluation

Begin by reconstructing the policy's information timeline. Determine when the intervention became sufficiently credible for affected individuals or institutions to respond.

Next, connect the expected behavioral response to the outcome being measured. A policy announcement may influence applications immediately but affect academic achievement only after students receive additional educational support.

For educational technology interventions, deployment dates deserve particular scrutiny. Software procurement, teacher training, student orientation, pilot access, and full operational use may occur over several months.

Researchers should not automatically classify all activities before full deployment as untreated. The appropriate classification depends on whether those activities can influence the outcome of interest.

Where possible, treatment timing and anticipation assumptions should be specified before examining post-intervention effect estimates.

Before Assuming No Anticipation, Verify These Conditions

Frequently Asked Questions About Anticipation Effects

Anticipation Can Shift the Effective Beginning of Treatment

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.