Policy Evaluation Interrupted Time Series
Also known as: ITS for policy evaluation, policy ITS, segmented regression for policy, policy impact ITS
Interrupted Time Series (ITS) for policy evaluation uses routinely collected aggregate time-series data to estimate the causal impact of a policy change. A segmented regression model splits the series at a known intervention date, estimating both an immediate level shift and a change in trend attributable to the policy — without requiring a randomised control group.
Key highlights
- Does not require a randomised control group — works with routine administrative, surveillance, or registry data.
- Estimates both the immediate impact (level change) and the sustained impact (slope change) of a policy, providing a nuanced picture of effect dynamics.
- Transparent counterfactual: the pre-intervention trend is the explicit baseline against which the post-intervention period is compared.
- Widely applicable across public health, economics, education, and criminology where aggregate time series are routinely available.
- Can be extended to a controlled ITS design by adding a parallel non-exposed series, strengthening causal credibility.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use policy evaluation ITS when a policy, regulation, or programme was introduced at a known point in time and you have a sufficiently long series of aggregate outcome measurements before and after. It is ideal when randomisation is ethically or practically impossible and only administrative or surveillance data exist — common in public health, economics, and social policy research. Do not use it when the series is too short (fewer than 8 pre-intervention points), when multiple concurrent policies confound the change point, when the exact implementation date is unknown or staggered across units, or when the outcome is not measured at regular intervals.
Strengths & limitations
- Does not require a randomised control group — works with routine administrative, surveillance, or registry data.
- Estimates both the immediate impact (level change) and the sustained impact (slope change) of a policy, providing a nuanced picture of effect dynamics.
- Transparent counterfactual: the pre-intervention trend is the explicit baseline against which the post-intervention period is compared.
- Widely applicable across public health, economics, education, and criminology where aggregate time series are routinely available.
- Can be extended to a controlled ITS design by adding a parallel non-exposed series, strengthening causal credibility.
- Requires a known, single, clean interruption date — staggered or diffuse policy rollouts undermine the design.
- Cannot control for concurrent events (other policies, economic shocks) that coincide with the policy change.
- Short series (fewer than ~8 points per segment) yield imprecise estimates and make trend modelling unreliable.
- Aggregate-level inference may not translate to individual-level causal effects (ecological fallacy).
- Results are sensitive to the choice of ARIMA autocorrelation structure; misspecification biases standard errors.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How many time points do I need before and after the intervention?
A common rule of thumb is at least 8 observations in each segment to estimate a stable trend. More is better: 12–24 points per segment substantially improve precision and the ability to detect autocorrelation. With fewer than 5 points in any segment, trend estimates are unreliable.
What is the difference between a level change and a slope change?
A level change (β₂) is an abrupt, immediate shift in the outcome at the policy date — like a sudden drop in emergency visits when a new drug policy takes effect. A slope change (β₃) is a gradual divergence in trajectory over time — like a steadily accelerating decline in smoking rates. Real-world policies often produce both.
How do I handle autocorrelation in the residuals?
First plot the ACF and PACF of residuals and run a Durbin-Watson test. If autocorrelation is present, fit an ARIMA error structure, use Newey-West (HAC) standard errors, or apply Prais-Winsten GLS correction. Ignoring autocorrelation inflates false-positive rates.
Can I add a control series to strengthen causal claims?
Yes. A controlled ITS design includes a comparison series from a group unaffected by the policy. If the intervention series shows a change at T* but the control series does not, the causal attribution is much stronger — this closely parallels the difference-in-differences logic.
What if the policy was rolled out at different times in different regions?
Staggered rollouts make standard ITS unreliable because the interruption date differs across units. In this case, a staggered difference-in-differences or a multi-site ITS model with region-specific break points is more appropriate.
Sources
- 1.Bernal, J. L., Cummins, S., & Gasparrini, A. (2017). Interrupted time series regression for the evaluation of public health interventions: a tutorial. International Journal of Epidemiology, 46(1), 348-355.
- 2.Box, G. E. P., & Tiao, G. C. (1975). Intervention Analysis with Applications to Economic and Environmental Problems. Journal of the American Statistical Association, 70(349), 70-79.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Policy Evaluation Interrupted Time Series. ScholarGate. https://scholargate.app/causal-inference/policy-evaluation-interrupted-time-series