Robust Interrupted Time Series Analysis
Also known as: robust ITS, outlier-robust ITS, robust segmented regression, robust ITSA
Robust Interrupted Time Series Analysis is a quasi-experimental method that estimates the causal effect of a policy or intervention on an aggregate outcome over time, using segmented regression fitted with outlier-resistant or heteroskedasticity-consistent standard errors. It is widely used in health services research and public-health evaluation when the time series contains influential observations, non-constant variance, or mild autocorrelation.
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 Robust ITS when you have an aggregate time series of at least 8-12 observations per segment (more is better; 50+ is ideal), a clearly defined intervention date, and reason to suspect outliers, non-constant variance, or mild serial correlation that would undermine standard ITS. It is the preferred formulation when data come from administrative systems prone to recording anomalies or when pilot periods cause early post-intervention volatility. Do not use it as a substitute for addressing severe autocorrelation — an ARIMA-corrected ITS is more appropriate there. Avoid it when the series is very short (fewer than 6 points per segment), as robust estimators are unreliable in small samples.
Strengths & limitations
- Retains the intuitive causal logic of standard ITS while producing valid inference under heteroskedasticity and mild autocorrelation.
- Protects against a small number of outlying time points distorting level-change and slope-change estimates.
- Sandwich (HC/HAC) standard errors require no distributional assumptions beyond a correctly specified mean function.
- Can be applied to both single-group and controlled ITS designs (with a concurrent comparison series).
- Widely available in standard statistical software (R itsadug, Stata itsa, Python statsmodels) with minimal additional effort over standard ITS.
- Robust methods correct inference but do not eliminate bias from model misspecification — if the segmented-linear form is wrong, neither standard nor robust ITS is valid.
- With very short segments (n < 6 per segment), robust sandwich estimators are unreliable and can produce worse coverage than OLS standard errors.
- Robust ITS does not fully address strong autocorrelation; Newey-West lags must be carefully chosen, and severe AR structure calls for explicit ARIMA modelling.
- The method still requires a single, known breakpoint; multiple or gradual interventions need extensions such as multiple-change-point or dynamic ITS.
Frequently asked
How is Robust ITS different from standard ITS?
The causal model and identification logic are identical. The difference is purely statistical: Robust ITS replaces OLS standard errors with heteroskedasticity-consistent (sandwich) or outlier-downweighting (M-estimation) alternatives, so inference remains valid when the error variance is non-constant or a few time points are anomalous.
When should I use Newey-West HAC versus M-estimation?
Use Newey-West (HAC) when you suspect heteroskedasticity and mild autocorrelation but the data have no severe outliers. Use M-estimation (or MM-estimation) when you have clearly anomalous time points that would exert undue leverage on OLS. You may apply both in a sensitivity analysis.
How many time points do I need?
A commonly cited minimum is 8 observations per segment (before and after), though 12-20 per segment is preferred for stable sandwich estimates. With fewer than 6 per segment, neither standard nor robust ITS is trustworthy — consider single-case design methods instead.
Does robust ITS require a comparison group?
No, a single-group design is valid if no major concurrent events coincide with the intervention. However, adding a concurrent control series (controlled ITS) greatly strengthens causal claims, especially for robust variants where residual plots may mask confounders.
Can I apply robust ITS to panel data?
Yes — panel (multi-unit) ITS with cluster-robust or unit-fixed-effects robust standard errors is a direct extension and is often preferable for policy evaluations that cover multiple regions or facilities.
Sources
- 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 ↗
- Linden, A. (2015). Conducting interrupted time-series analysis for single- and multiple-group comparisons. Stata Journal, 15(2), 480-500. link ↗
How to cite this page
ScholarGate. (2026, June 3). Robust Interrupted Time Series Analysis. ScholarGate. https://scholargate.app/en/causal-inference/robust-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
- Robust Difference-in-DifferencesCausal inference↔ compare