Interrupted Series with Pretrend Check¶
Time series — instantiates Regression-to-the-Mean Guardrail
Fits the pre-event trend and seasonality of a single series, then tests whether the outcome shifts level or slope at the event beyond what the extrapolated pretrend and a transient spike predict.
Interrupted Series with Pretrend Check works within a single outcome series measured many times before and after the event. Its defining move is fitting the pre-event trajectory — trend, seasonality, cycles — and projecting it forward as the counterfactual, then asking whether the outcome jumps in level or bends in slope at the moment of intervention beyond what that projection predicts. Crucially, it treats the extreme observation that triggered the intervention as one point on the series, not as the baseline: if the trigger was a transient spike, the pre-trend already predicts a fall back toward the line. Only a shift that exceeds both the extrapolated trend and the expected spike-decay counts as a candidate effect. It needs no separate control group, buying its leverage from temporal structure instead.
Example¶
A road authority installs a speed camera at a junction after a year with an alarming spike in serious collisions. The next year, collisions drop sharply, and the camera is hailed as the cause. Interrupted Series with Pretrend Check pulls the full monthly collision history — several years before and after the install. It fits the pre-install trend and the strong seasonal pattern (collisions peak in winter), and it notes that the trigger year was an unusual spike well above the fitted line. Projecting the pre-install trajectory forward, it estimates where collisions would have landed with no camera — already lower, because the spike year was atypical. The test then asks whether the post-install series drops below that projection. Part of the observed decline is recovered by the pretrend and spike-decay; only the residual step-down at the install date, if any, is attributable to the camera.
How it works¶
The rigor is in modeling the counterfactual trajectory before crediting the interruption:
- Assemble a long series. Gather many equally-spaced observations before and after the event, enough to estimate trend and seasonality reliably.
- Fit the pre-event process. Model trend, seasonality, and cycles from the pre-period, and locate the trigger observation on that fitted curve to gauge how transient it was.
- Project and test the interruption. Extrapolate the pre-event trajectory as the counterfactual, then test for a level shift or slope change at the event beyond the projection.
- Decompose the movement. Partition the observed post-event change into extrapolated trend, spike-decay, seasonality, and any residual step attributable to the event — leaving unexplained variance visible.
Tuning parameters¶
- Pre-period length — how many pre-event observations anchor the trend. Longer pre-periods stabilize the projection but risk spanning a regime change.
- Trend/seasonality model — linear, spline, or seasonal decomposition; richer models fit better but can overfit and absorb a real effect.
- Level versus slope — whether the test looks for a one-time jump, a change in trajectory, or both; the choice must match the intervention's expected mechanism.
- Autocorrelation handling — how serial correlation in the residuals is modeled; ignoring it understates uncertainty and manufactures spurious significance.
When it helps, and when it misleads¶
Its strength is that it exploits the temporal structure a naïve two-point before-after throws away, so it can separate a genuine step-change from a spike that was always going to decay — the antidote to the archetype's timeless-mean fallacy, which asserts reversion toward a fixed average while ignoring trend and season.[n1]
Its failure mode is confounding by concurrent events and by the series' own structure. A co-occurring change — a new road layout installed the same month as the camera — is indistinguishable from the intervention within a single series, and short pre-periods or unmodeled autocorrelation produce false interruptions. The classic misuse is crediting the intervention for the natural fall of a spike, treating a two-point comparison as if it were the full series. The guarding discipline is to inspect the pretrend explicitly — confirming the trigger really was a transient extreme — and to hunt for concurrent events that could masquerade as the interruption before attributing the step to the cause.
How it implements the components¶
time_process_and_regime_map— its core act: modeling trend, seasonality, and cycles so temporal structure is separated from the intervention's effect.pre_intervention_repeated_baseline— the long pre-event series is its baseline, revealing whether the trigger observation was transient.observed_change_decomposition— it partitions the post-event movement into extrapolated trend, spike-decay, seasonality, and residual step.
It reasons within one series and so builds no parallel control group (concurrent_counterfactual_comparison — Matched Extreme-Case Comparator) and does not translate its result into a graded claim (attribution_language_and_evidence_grade — Attribution-Claim Review Gate).
Related¶
- Instantiates: Regression-to-the-Mean Guardrail — it supplies the temporal counterfactual that distinguishes a durable shift from a decaying spike.
- Sibling mechanisms: Extreme-Selection Risk Flag · Multi-Baseline Measurement Protocol · Matched Extreme-Case Comparator · Randomized or Staggered Assignment · Reliability-Based Reversion Simulation · Shrinkage-Aware Expectation · Controlled Before–After Contrast · Placebo Time, Outcome, or Threshold Check · Attribution-Claim Review Gate
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Interrupted Series with Pretrend Check operates as a computation, comparison, model, or analytic representation used to infer, estimate, or choose because it fits the pre-event trend and seasonality of a single series, then tests whether the outcome shifts level or slope at the event beyond what the extrapolated pretrend and a transient spike predict
Independent corroboration: The frozen evidence defines Interrupted Series with Pretrend Check as 'Fits the pre-event trend and seasonality of a single series, then tests whether the outcome shifts level or slope at the event beyond what the extrapolated pretrend and a transient spike predict', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Interrupted time-series estimation of level and slope changes against a projected pretrend is a canonical quasi-experimental design.
Related originating lineages:
- Public Administration & Policy — Program and policy evaluation materially established the design's broad applied use when randomization is unavailable.
Review resolution: Both independent reviews place the primary lineage in statistics_experimental_design. The queued differences (origin_mode_disagreement) concern secondary metadata rather than primary provenance. The final retains public_administration_policy only where a reviewer supplied a formative-lineage rationale; this does not convert downstream applicability into origin. origin_mode=convergent because the reviewers document independently established or materially co-developing traditions. domain_reach=multi_domain records application breadth separately from provenance.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] The interrupted time-series design uses repeated observations before and after an event to test for a change in level or slope against the projected pre-event trajectory. It is a standard corrective to two-point before-after evaluations, which cannot distinguish an intervention effect from an ongoing trend or the decay of a transient extreme. ↩