Multi-period Interrupted Time Series
Multi-period Interrupted Time Series Analysis · Also known as: multi-period ITS, multiple-interruption ITS, segmented time series with multiple breakpoints, MITS
Multi-period Interrupted Time Series (MITS) extends the classic ITS framework to settings where two or more interventions occur at known time points within the same series. By fitting a segmented regression with multiple breakpoints, MITS estimates the level change and slope change attributable to each intervention while controlling for the underlying secular trend and for the effects of earlier interruptions.
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 MITS when you have a single aggregate or individual-level time series covering at least 10-12 observations per segment and two or more interventions occurred at known, externally defined time points. It suits sequential policy changes, clinical guideline updates, or regulatory amendments where no concurrent control group is available. Do not use it when breakpoints are chosen post hoc by scanning the data, when fewer than 8-10 observations exist per segment, when confounding events within segments cannot be ruled out, or when a control comparison series is available and a controlled ITS would be feasible.
Strengths & limitations
- Captures both immediate level shifts and gradual trend changes from each of multiple interventions in a single coherent model.
- Makes full use of the time-series data without discarding observations between intervention periods.
- Provides a separate counterfactual extrapolation and effect estimate for each intervention, enabling direct comparison of policy phases.
- Can incorporate control series, seasonal terms, or autocorrelation corrections within the same regression framework.
- Applicable to routinely collected administrative data where randomisation or matched controls are unavailable.
- Requires a priori knowledge of all breakpoint dates; data-driven breakpoint search invalidates the inferential framework.
- The parallel-segments assumption may fail if unobserved shocks occurred between interventions.
- Degrees of freedom are consumed rapidly: K breakpoints add 2K parameters, so short series with many interventions become under-powered.
- Autocorrelation corrections add complexity and require careful model selection; misspecification of the error structure biases standard errors.
- Effect estimates for later interventions can be sensitive to how earlier breakpoints are modelled.
Frequently asked
How many observations per segment do I need?
At least 10-12 observations per segment to estimate both a level and a slope change reliably. Fewer points produce very wide confidence intervals and make the level-slope decomposition unreliable. If segments are short, consider collapsing adjacent interventions or using a simpler model with only level changes.
Can I choose breakpoints by looking at where the data changes most?
No. Breakpoints must be set a priori from external information such as policy dates or guideline release dates. Selecting them by scanning the series for large shifts is equivalent to testing hypotheses generated by the data, which inflates the Type I error rate and makes the estimates uninterpretable as causal quantities.
How does MITS differ from running separate ITS analyses for each intervention?
Separate ITS models ignore that later segments inherit the trajectory from earlier ones. MITS models all interruptions jointly, correctly propagating the pre-existing trend into each successive segment's counterfactual. Separate analyses conflate later interventions with residual effects or trend changes from earlier ones.
What if my time series shows seasonal patterns?
Add harmonic terms or monthly dummy variables to the regression to absorb seasonality before estimating breakpoint effects. Unmodelled seasonality can produce spurious level or slope estimates, especially when interventions coincide with seasonal troughs or peaks.
Is a control group required?
Not strictly, but a concurrent control series greatly strengthens causal interpretation by ruling out confounding events that affect the outcome regardless of the interventions. When a plausible control series exists, a controlled MITS is strongly preferred over an uncontrolled single-series analysis.
Sources
- Kontopantelis, E., Doran, T., Springate, D. A., Buchan, I., & Reeves, D. (2015). Regression based quasi-experimental approach when randomisation is not an option: interrupted time series analysis. BMJ, 350, h2750. DOI: 10.1136/bmj.h2750 ↗
- 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. DOI: 10.1093/ije/dyw098 ↗
How to cite this page
ScholarGate. (2026, June 3). Multi-period Interrupted Time Series Analysis. ScholarGate. https://scholargate.app/en/causal-inference/multi-period-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 Interrupted Time SeriesCausal inference↔ compare
- Interrupted Time SeriesCausal inference↔ compare
- Panel Data Interrupted Time SeriesCausal inference↔ compare
- Panel Event StudyCausal inference↔ compare