EvalCompassEvalCompass
Causal inference & experimentsCausal inferenceTime seriesPolicy evaluationAggregate dataSegmented regression

Interrupted Time Series Analysis

Interrupted Time Series Analysis
Also known as: ITS、Time-series intervention analysis、Segmented Regression Analysis

Uses continuous before/after series to estimate changes in outcome level or slope. Consider seasonality, autocorrelation and concurrent changes. [1]

Interrupted time-series design and diagnostics worksheet — openReferences and checked sections
Difficulty
Intermediate
Cost
Medium
Duration
Illustrative planning durations, not common standards: collection 1–4 weeks, analysis 1–2 weeks
Content updated: 2026-09-13Source checked: 2026-09-13 (includes AI cross-checking)No documented human reviewIncludes fictional examples

How it works at a glance

Specify intervention time and hypothesize immediate level versus gradual slope changes beforehand. Compare post-trends with counterfactual extrapolated pre-trends. [1]

This is not a two-point comparison. Check frequency, duration, missingness, seasonality and error dependence, adding suitable comparison series where possible. [1]

ResultsTime →InterventionCounterfactual trendChange in level and slope
A conceptual diagram of changes in a time series’ level and slope before and after an intervention. Examine concurrent policies, seasonality, autocorrelation, and other factors. A change in the line alone does not establish an intervention effect.

When to use it and when it is unsuitable

When to use it

  • Monthly/annual administrative, disease or crime series around law, policy or program changes
  • Nationwide or whole-area policies without easy controls but with long series
  • Health/public-health effects of drug approvals, infection measures or screening using aggregates
  • Series permits pre/post trends and seasonality at required precision

When it is unsuitable

  • Short series or changing definitions leave weak counterfactual-trend grounds
  • Concurrent external shocks, other policies or crises cannot be separated
  • Only aggregates available but individual changes/distributions needed

Participants, roles, and deliverables

Participants and roles

1–2 evaluators/statisticians using R or similar software, securing data with administrative statistics staff

Main deliverables

  • Time-series plot: observations, fitted and counterfactual lines
  • Segmented coefficients with level/slope changes and intervals
  • Autocorrelation diagnostics: ACF and DW
  • Comparison findings for controlled ITS

Implementation steps

  1. Check outcome and series
    Define outcomes such as monthly emergency transports/crimes and monthly, quarterly or annual units. Collect around intervention and check missing/outlying values.
  2. Identify intervention and code time
    Identify exact T and code 3 variables: time t, intervention dummy D=0 before/1 after, and time since intervention t*.
  3. Check autocorrelation
    Examine residual dependence, seasonality and overdispersion; select models/uncertainty suited to outcome distribution and series.
  4. Fit segmented regression
    Estimate pre-level/intercept, pre-slope, post-level change and post-slope change; interpret each coefficient substantively.
  5. Visualize findings
    Show observations, fitted post-line, extrapolated counterfactual and uncertainty. Differences are model estimates; a bend alone does not prove causality.
  6. Comparison-series checks and sensitivity
    Where available, examine selection rationale/common shocks for controlled designs. Vary seasonality, implementation ramp-up/delay and analysis periods.

Strengths and limitations

Strengths

  • Analyzes level and trend changes with clearly timed interventions
  • Separates immediate levels from slope changes to examine hypothesized effects
  • Suits whole-population policies or aggregate-only contexts such as health and urban policy
  • Valid comparison series help assess common external explanations

Limitations

  • Counterfactual continuation is more uncertain with short pre-series
  • Concurrent shocks prevent effect separation and threaten internal validity
  • Aggregates cannot directly establish individual causal effects; subgroup analysis needs suitable data/design
  • Unaddressed autocorrelation underestimates errors and can produce false significance

Common pitfalls

  • Do not use timepoint count alone as pass/fail. Length depends on variability, seasonality, dependence, effect shape and precision.
  • Ignoring dependence or distributions misestimates uncertainty. Use diagnostics and suitable models.
  • Distinguish seasonality from intervention changes; explain intervals, cycles, periods and adjustment.
  • Check concurrent policy changes, such as same-day regulations, to avoid attribution errors.

Conditions of use and evidence

  • Predefine effect shapes and examine autocorrelation, seasonality and time-varying confounding. [1]
Fictional example; not actual results

Track waiting after a booking system starts

A fictional model: pre-waiting is 100−t minutes; introduction at t=12 gives immediate −8 minutes and additional slope −0.5 minutes/month.

Month tEstimate without interventionEstimate with interventionDifference
1288 minutes80 minutes−8 minutes
1486 minutes77 minutes−9 minutes

Common errors in judgment

Common mistake

Compare only immediately adjacent months

How to revise

Examine trends/seasonality over sufficient before/after series

ReasonTemporary fluctuations are easily mistaken for effects.

Common mistake

Assume more timepoints resolve concurrent policies

How to revise

Record concurrent changes and consider comparison series/evidence

ReasonSimultaneous factors cannot be separated by count alone.

Organize your thinking by writing

Interrupted time-series design and diagnostics worksheet

A practical template is available to fill in and print on A4 paper.

Open worksheet

Related methods

Difference-in-DifferencesRegression Discontinuity DesignSynthetic Control MethodRandomized Controlled TrialContribution Analysis

References and checked sections

  1. Interrupted time series regression for the evaluation of public health interventions: a tutorial
    James Lopez Bernal, Steven Cummins, and Antonio Gasparrini · 2017
    Section consulted
    The interrupted time series design / Step 4: regression analysis
    How it relates to the text
    Predefine effect shapes and examine autocorrelation, seasonality and time-varying confounding.
    Source checked: (includes AI cross-checking)

Editorial and checking information

Content updated: Source checked: (includes AI cross-checking)No documented human reviewIncludes fictional examplesEditorial policy and scope of checkingCite this pageReport a correction

How to cite this page

EvalCompass Editorial Team. “Interrupted Time Series Analysis”. EvalCompass, updated 2026-09-13. https://evalcompass.jun-nakatani.com/en/methods/its/
Add the date you accessed the page. Text created by this site is available under CC BY 4.0. Quotations, external materials, and bundled fonts have their own terms of use. When reusing content, credit the source, link to the license, and indicate any changes.

Suggest a correction or improvement to this page

Change history
  • — Checked sources for definitions and conditions and added fictional examples, judgment rationales and completed entries for every field. Revised unsupported historical or uptake claims. Reviewed procedures and cautions, adjusting fixed counts, thresholds and institutional generalizations. Preserved field positions, types and counts while revising fixed conditions and thresholds.