When Is an Interrupted Time-Series Design Stronger Than a Simple Before-and-After Study?

When Does Measuring an Outcome Repeatedly Provide Stronger Evidence of an Intervention Effect?

A university introduces a new academic support program. Student performance improves afterward, and administrators want to know whether the program caused the improvement.

A simple before-and-after study could compare average performance immediately before and after implementation. But suppose the university has monthly performance records covering several years. Those records might reveal that achievement was already improving long before the program began.

Rather than comparing two observations, researchers could examine the trajectory before implementation and investigate whether its level or rate of change shifted afterward.

This is the central logic of an interrupted time-series design. It can strengthen an intervention evaluation when repeated observations are available and a clearly defined intervention occurs at an identifiable point in time.

Interrupted Time Series Is Stronger When the Pre-Intervention Trend Matters

How Interrupted Time-Series Designs Improve Intervention Evaluation

What Is an Interrupted Time-Series Design?

An interrupted time-series design, commonly abbreviated as ITS, is a longitudinal quasi-experimental approach that examines an outcome measured repeatedly before and after a defined intervention or event.

The intervention represents an interruption in the observed series. Researchers investigate whether the outcome's trajectory changes around that point in a manner consistent with the intervention's expected effect.

For example, a university might examine monthly student withdrawal rates before and after introducing a new academic advising policy.

Rather than asking only whether withdrawal rates were lower afterward, an ITS analysis asks whether the post-intervention pattern differs from the trajectory expected from the pre-intervention observations.

The design has been widely used in evaluations of health policies, organizational interventions, and population-level programs. It can also be useful in educational research when suitable repeated outcome data are available.

How Is ITS Different From a Simple Before-and-After Study?

A conventional before-and-after study often compares one pre-intervention measurement with one post-intervention measurement.

Such a comparison describes the observed difference but provides limited information about whether the outcome was already increasing, decreasing, or fluctuating.

An interrupted time-series design examines a sequence of observations. The pre-intervention series provides evidence about the existing outcome trajectory, while the post-intervention series helps determine whether that trajectory changed.

The advantage of ITS comes from the information contained in the repeated observations and the ability to evaluate departures from an existing trajectory. It does not arise simply because the design uses more data points.

What Is the Counterfactual in Interrupted Time-Series Analysis?

The counterfactual is the outcome trajectory that would have occurred without the intervention.

In a simple ITS analysis, researchers often estimate this trajectory by extending the modeled pre-intervention pattern into the post-intervention period.

Suppose monthly research consultation requests increased steadily before a new support policy. The counterfactual projection might indicate that requests would have continued increasing at approximately the same rate without the policy.

The estimated intervention effect is then based on the difference between the observed post-intervention outcome and that projected trajectory.

This counterfactual is not directly observed. It depends on assumptions about the continuation of pre-intervention patterns and the absence of other relevant changes that would have altered the outcome.

What Is an Immediate Level Change?

A level change is an abrupt shift in the outcome around the intervention point relative to the projected pre-intervention trajectory.

For example, an institution introduces a mandatory online submission system, and the number of paper-based submissions decreases sharply immediately afterward.

An ITS analysis may estimate the magnitude of that immediate departure.

However, the interpretation depends on the measurement interval and intervention mechanism. An effect that develops over several months may not be appropriately represented as an instantaneous level change.

Researchers should specify whether an immediate response is theoretically plausible before choosing the statistical model.

What Is a Change in Slope?

A slope change occurs when the rate of increase or decrease in the outcome differs after intervention implementation.

Suppose a university's monthly student withdrawal rate had been declining slowly before a new retention program. Following implementation, the rate begins declining more rapidly.

The intervention may be associated with a change in the trajectory's slope rather than an immediate shift in its level.

Level and slope changes can occur separately or together. A model may therefore include both parameters when supported by the research question and data.

How Does Segmented Regression Work?

Segmented regression is a commonly used analytical approach for interrupted time-series data. It models the outcome trajectory before and after a defined intervention point.

A basic model for a single interruption can be written as follows.

The example illustrates two different effects. The intervention is associated with an immediate increase in the outcome, but the subsequent rate of increase is slower than before.

The estimated difference from the projected counterfactual at a later time depends on both the level-change and slope-change coefficients.

These coefficients should not automatically be interpreted as causal effects. Their causal meaning depends on whether the model provides a credible representation of the outcome that would have occurred without the intervention.

How Many Observations Does an ITS Design Need?

There is no universally sufficient number of observations for every interrupted time-series study.

Some evidence-review criteria have used a minimum of three observations before and after the interruption to classify a study as an interrupted time series. This is an eligibility threshold, not a guarantee that a statistically reliable model can be estimated.

Methodological guidance for segmented regression has often recommended substantially longer series, with approximately 12 observations before and 12 afterward sometimes used as a practical planning reference.

However, these figures should not be treated as universal requirements. The necessary number depends on outcome variability, autocorrelation, seasonality, anticipated effect size, intervention timing, and model complexity.

Researchers should determine the number of pre-intervention observations in conjunction with the number of post-intervention observations and the intended analysis.

Simulation-based sample-size or power assessment may be useful when sufficient information about the outcome process is available.

Why Does Autocorrelation Matter?

Autocorrelation occurs when observations are statistically related to observations from nearby time points.

For example, a university's enrollment rate in one month may be associated with its enrollment rate in the preceding month.

Ordinary regression procedures that assume independent errors can produce misleading standard errors and confidence intervals when serial correlation is present.

Researchers should examine the residual structure and consider methods appropriate to the degree and form of autocorrelation.

Depending on the data, these may include generalized least squares, autoregressive error models, or other time-series approaches. The method should be justified rather than chosen solely because it is available in statistical software.

What Happens When the Outcome Is Seasonal?

Many educational and institutional outcomes follow recurring calendar patterns.

Learning management system activity may rise before examinations and fall during academic breaks. Enrollment inquiries may vary across admission cycles. Student attendance can be influenced by the academic calendar.

If the intervention begins near a predictable seasonal change, an apparent level or slope difference may partly reflect that recurring pattern.

Researchers should inspect the time series for seasonality and use appropriate modeling strategies when sufficient observations are available.

A short series that covers only part of a seasonal cycle may not provide enough information to distinguish seasonality from an intervention effect.

Why Is the Intervention Date So Important?

An ITS design requires a defensible definition of the interruption.

Some interventions begin on a specific date, such as the implementation of a new institutional policy. Others are introduced gradually through orientation, training, pilot implementation, and phased adoption.

When exposure develops gradually, treating the intervention as an instantaneous event may misrepresent its mechanism.

Researchers may need to model a transition period, delayed response, or gradual change in exposure, depending on the intervention and available data.

The intervention point should be specified using substantive information rather than selected retrospectively to maximize the apparent effect.

When Is ITS Particularly Vulnerable to History Effects?

An important limitation of an uncontrolled interrupted time series is that another event occurring near the intervention point may produce a similar change in the outcome.

Suppose a university introduces a student-support platform at the same time it revises its academic probation policy. A subsequent decline in student withdrawals could plausibly reflect either initiative or their combined influence.

The time series may reveal a change in trajectory, but it cannot automatically determine which concurrent event caused it.

This is one of the alternative explanations for intervention-associated changes that must be investigated.

Researchers should document relevant policy changes, organizational events, changes in population composition, and other exposures occurring around implementation.

What Is a Controlled Interrupted Time-Series Design?

A controlled interrupted time-series design includes an additional outcome series that provides a comparison for the intervention series.

For example, researchers might compare monthly outcomes at an institution implementing a new policy with outcomes at a comparable institution that does not implement the policy during the same period.

The comparison series may help identify changes arising from broader events affecting both institutions.

However, the control series must be substantively appropriate. Differences in baseline trends, population composition, exposure to other policies, and measurement procedures can undermine its usefulness.

As with other comparison-group decisions, similarity in a few observed characteristics is not sufficient to establish that the groups have comparable counterfactual trajectories.

Does ITS Require Individual Participants to Be Followed?

No. Interrupted time-series studies can use repeatedly measured aggregate outcomes, such as monthly hospital admissions, weekly school attendance rates, or quarterly institutional performance indicators.

The same individuals do not necessarily have to contribute data at every observation.

However, changes in the composition of the population can affect the outcome series. For example, a university's average achievement score may change because different student cohorts are represented across academic terms.

Researchers should examine whether changes in the measured population or outcome definition could explain an apparent interruption.

When individual-level repeated measurements are available, additional longitudinal models may be appropriate depending on the research question.

Can ITS Establish Causation Without Randomization?

Interrupted time-series analysis can support causal inference when the intervention timing is well defined, the counterfactual projection is credible, and competing explanations are adequately addressed.

Its strength lies partly in the ability to examine whether an outcome departs from an established trajectory rather than simply changing between two observations.

Nevertheless, ITS does not eliminate unmeasured confounding or concurrent events. The design remains observational unless an experimental allocation mechanism is incorporated.

Researchers should therefore distinguish evidence of a change associated with the interruption from evidence that the intervention caused that change.

A carefully designed controlled ITS may provide stronger causal evidence than an uncontrolled ITS, although its validity still depends on the appropriateness of the comparison and the assumptions of the analysis.

When Is a Simple Before-and-After Study More Appropriate?

ITS is not automatically the preferred design for every intervention.

A simple before-and-after approach may be proportionate when the objective is descriptive, the outcome is measured at only two meaningful occasions, or the study is intended primarily to assess feasibility.

For example, researchers evaluating whether participants demonstrate improved knowledge immediately after a brief workshop may not need a long time series if their primary objective is to describe immediate learning.

In contrast, an institution-wide policy intended to change monthly student withdrawal rates may be better evaluated using a time-series approach when adequate historical and follow-up records exist.

The appropriate design follows the outcome's temporal behavior, the intervention's mechanism, and the strength of inference required.

Did a New Academic Support Policy Change Student Withdrawal Rates?

A simple comparison of the final pre-intervention month with the final post-intervention month shows a decline from 9.5% to 5.5%, equivalent to 4 percentage points.

However, the withdrawal rate was already declining by 0.5 percentage points per month before the policy began.

If that perfectly linear pre-intervention trend had continued without the policy, the projected withdrawal rate for Month 7 would have been 9.0%, and the projected rate for Month 12 would have been 6.5%.

The difference between the observed Month 12 rate and the projected pre-intervention trajectory is smaller than the raw 4-percentage-point decline between Months 6 and 12.

The data also show an immediate departure at Month 7: the observed rate of 8.0% is 1.0 percentage point below the projected 9.0%. Thereafter, the observed series continues declining at the same monthly rate as before implementation.

Under this simplified deterministic example, the pattern is consistent with an immediate downward level shift of 1.0 percentage point and no change in slope.

This is an illustration of the design's logic, not a formal statistical analysis. With only six observations per segment, no stochastic variation in the constructed series, and no comparison group, the example cannot establish the reliability or causal interpretation of the apparent interruption.

A real evaluation would need to investigate seasonality, autocorrelation, changes in student composition, relevant concurrent policies, and uncertainty in the estimated trajectory.

Nevertheless, the example demonstrates why examining the existing trend can materially change the interpretation of a before-and-after difference.

Common Misunderstandings About Interrupted Time-Series Designs

How to Decide Whether Interrupted Time Series Is Appropriate

Begin by examining the outcome and the available data. Determine whether measurements exist across enough periods to characterize the pre-intervention trajectory and evaluate what happens after implementation.

Next, assess whether the intervention has a clearly identifiable starting point and whether its expected effect is immediate, gradual, or delayed.

Finally, consider whether the data and design can support a credible counterfactual projection.

Before finalizing the design, inspect the time series graphically and examine whether its properties support the intended model. Consider temporal dependence, seasonal patterns, unusual observations, and changes in outcome definitions.

The choice of post-intervention assessment timing is also consequential. An intervention with delayed effects may require a sufficiently long follow-up period before its expected trajectory becomes observable.

Researchers should prespecify the intervention point, primary outcome, model structure, relevant covariates, and planned sensitivity analyses whenever feasible.

When a well-supported ITS design is feasible, it may provide substantially more informative evidence than a two-time-point comparison. When the necessary observations or assumptions are unavailable, presenting a simple before-and-after analysis with appropriate limitations is preferable to labeling an inadequately supported analysis as an interrupted time series.

Before Choosing an Interrupted Time-Series Design, Check:

Frequently Asked Questions About Interrupted Time-Series Studies

Interrupted Time Series Is Stronger When It Reveals What Was Already Happening

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.