Monthly emergency department visits were rising before an intervention. After the intervention, visits continue rising — but at a slower rate. What is the correct ITS interpretation?
AThe intervention had no effect because visits continued to increase
BThe intervention caused a negative level change at the breakpoint
CThe intervention caused a negative slope change — the trend still went up, but grew more slowly
DITS cannot be interpreted when the outcome moves in the same direction before and after
ITS distinguishes two types of intervention effects: an immediate level change (a jump or drop at the breakpoint) and a slope change (a change in the rate of trend). Visits continuing to rise but more slowly is a negative slope change — the intervention didn't reverse the trend, it decelerated it. Option A is the classic misconception: equating 'no reversal' with 'no effect,' which ignores slope-change detection entirely.
Question 2 Multiple Choice
What is the fundamental difference between ITS and a simple pre-post comparison of means?
AITS requires a comparison group; pre-post comparison does not
BITS explicitly models the pre-intervention trend and tests whether outcomes deviated from where that trend predicted they would be
CITS can only detect level changes, while pre-post comparisons measure both level and slope changes
DPre-post comparison controls for seasonality; ITS does not
The core ITS insight is that the pre-intervention trend itself serves as the counterfactual — what would have happened without the intervention. The model projects that trend forward and asks whether the post-intervention data deviates from the projection. A simple pre-post comparison of means ignores whether a trend was already present, making it unable to distinguish a genuine intervention effect from a continuation of an existing trajectory.
Question 3 True / False
A statistically significant level change at the ITS intervention point proves that the intervention caused that change.
TTrue
FFalse
Answer: False
A significant level change is consistent with causation but does not prove it. The key threats to ITS validity are co-interventions (other events occurring simultaneously at the breakpoint) and secular trends that coincide with the intervention. If a new hospital policy and a nationwide health campaign both launched in the same month, the level change could reflect either or both. Strong causal inference requires ruling out these competing explanations, ideally through a control series that shares secular trends but was not exposed to the intervention.
Question 4 True / False
Seasonality in the outcome variable can be addressed within an ITS regression model by including Fourier terms or monthly indicator variables.
TTrue
FFalse
Answer: True
Many health outcomes cycle predictably with the calendar — flu peaks in winter, drowning in summer. If an intervention coincides with a seasonal peak, a naive ITS model will misattribute the seasonal change to the intervention. Fourier terms (sine and cosine functions of time) or month indicators model the periodic variation explicitly, separating seasonal effects from the estimated intervention effect. This is why a long pre-intervention series is important: you need enough data to characterize the seasonal pattern before the breakpoint.
Question 5 Short Answer
Why is a long pre-intervention time series important for ITS validity, and which two specific threats does sufficient pre-intervention data most directly address?
Think about your answer, then reveal below.
Model answer: A long pre-intervention series is essential for two reasons: (1) it allows the model to reliably estimate the underlying secular trend and seasonal pattern, providing a credible counterfactual against which to measure post-intervention deviation; (2) it helps distinguish a genuine intervention effect from regression to the mean — if the intervention was triggered by an unusual spike, a long baseline shows whether the spike was anomalous or part of a real trend. Without sufficient pre-intervention data, the trend estimate is unreliable, and seasonal confounding cannot be adequately controlled.
The ITS design's strength is using the unit's own prior trajectory as the counterfactual. But this requires that trajectory to be well-characterized. Short pre-periods cannot reliably separate seasonal patterns from trend. They also cannot identify regression to the mean: an outcome that triggered intervention by spiking unusually high will naturally decline afterward, which can mimic a genuine intervention effect. A long pre-period makes these confounds visible and estimable.