Dynamic Interrupted Time Series
Dynamic Interrupted Time Series Analysis · Also known as: Dynamic ITS, ITS with lagged effects, time-varying ITS, flexible ITS
Dynamic Interrupted Time Series (Dynamic ITS) extends the standard ITS design by allowing intervention effects to build up, decay, or shift over multiple time lags rather than assuming a single instantaneous level change. It estimates how an intervention's impact evolves across time periods, making it especially suited to public health, health services research, and policy evaluation where effects accumulate gradually or wear off after initial impact.
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 Dynamic ITS when you have a sufficiently long regularly spaced time series (typically at least 12 pre-intervention and 12 post-intervention observations), a clearly datable intervention, and reason to believe the effect builds or decays across multiple periods rather than being immediate and permanent. It is well suited to aggregate administrative data in health, education, and public policy where randomisation is impossible. Do not use it with very short series (fewer than 8–10 observations in either segment), when the intervention date is ambiguous or staggered across many units without control, or when a panel design with multiple control units would better support counterfactual construction.
Strengths & limitations
- Captures the full dynamic trajectory of an intervention effect, including delayed onset, peak impact, and gradual decay.
- Requires no untreated control group — the pre-intervention series serves as the counterfactual trend.
- Applicable to aggregate administrative time series commonly available in health surveillance, education, and policy databases.
- More informative than standard ITS when effect timing is uncertain or theoretically expected to be gradual.
- Distributed-lag specification is flexible: lag length and functional form can be adapted to substantive theory.
- Requires a long pre-intervention series to estimate the counterfactual trend reliably; short series produce wide confidence intervals.
- The choice of maximum lag K is judgmental; too few lags miss delayed effects, too many lags overfits and inflates variance.
- Cannot rule out concurrent events (co-interventions, secular trends) that coincide with the intervention — a control series is strongly recommended when available.
- Autocorrelation correction is essential but adds analytic complexity and additional modelling choices.
- Staggered implementation across units requires more complex multi-group extensions.
Frequently asked
How is Dynamic ITS different from standard ITS?
Standard ITS estimates two parameters at the breakpoint — an immediate level change and a slope change — assuming the intervention effect is instantaneous and permanent. Dynamic ITS adds lagged intervention indicators so the model can detect effects that emerge gradually, peak at some lag, and then diminish. It is a distributed-lag generalisation of the standard segmented regression.
How many pre- and post-intervention observations do I need?
A common rule of thumb is at least 12 data points in each segment for standard ITS; dynamic models that estimate K+1 lag coefficients additionally require enough post-intervention observations to support that many parameters. As a practical minimum, ensure post-intervention length exceeds K by a comfortable margin and pre-intervention length is long enough to estimate trend and seasonality.
How do I choose the maximum lag K?
Ideally, K should be set from substantive theory — for instance, if a drug is expected to reach full effect within three months, K = 3. When theory is unclear, information criteria (AIC, BIC) can guide selection, but the choice should be reported and sensitivity analyses over plausible values of K should be presented.
Do I need a control series?
A control series (a comparable outcome or region not exposed to the intervention) is strongly recommended whenever one is available. It allows you to subtract concurrent secular trends or external shocks from the estimated effect, strengthening causal inference. Without a control series, unmeasured co-interventions remain a threat to validity.
What software can I use?
Dynamic ITS can be implemented in R (packages itsa, lmtest, sandwich for HAC errors; or ARIMA-based approaches via forecast), Stata (newey, prais commands), or Python (statsmodels). StatWise provides a guided interface that handles autocorrelation correction and lag selection diagnostics automatically.
Sources
- Lopez Bernal, J., 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. DOI: 10.1093/ije/dyw098 ↗
- Wagner, A. K., Soumerai, S. B., Zhang, F., & Ross-Degnan, D. (2002). Segmented regression analysis of interrupted time series studies in medication use research. Journal of Clinical Pharmacy and Therapeutics, 27(4), 299-309. DOI: 10.1046/j.1365-2710.2002.00430.x ↗
How to cite this page
ScholarGate. (2026, June 3). Dynamic Interrupted Time Series Analysis. ScholarGate. https://scholargate.app/en/causal-inference/dynamic-interrupted-time-series
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
- Dynamic Difference-in-DifferencesCausal inference↔ compare
- Interrupted Time SeriesCausal inference↔ compare
- Panel Event StudyCausal inference↔ compare