An ITS analysis of a hand hygiene intervention in a hospital models monthly infection rates for 24 months before and 24 months after the intervention. The model estimates two types of effects: a level change (immediate) and a trend change (gradual). Why is distinguishing between these two effects important?
AOnly the level change matters because it represents the true intervention effect
BAn intervention might produce an immediate drop in infections (level change) and also alter the ongoing trajectory (trend change) — capturing both provides a complete picture of the intervention's impact over time
CThe trend change is always larger than the level change
DThe distinction is only important for statistical reasons, not for clinical interpretation
Some interventions produce immediate effects (a new antibiotic formulary reducing infections immediately) while others change the rate of improvement (a quality improvement program that gradually reduces infections over months). Some produce both. An ITS model that only captures level change would miss a gradual effect that accumulates over time; one that only captures trend change would miss an immediate impact. The segmented regression framework estimates both, allowing researchers to distinguish between immediate and sustained impacts of the intervention.
Question 2 Multiple Choice
An ITS analysis uses OLS regression and finds a significant level change after a policy intervention. However, the Durbin-Watson statistic is 0.8. What is the concern?
AThe model has too many parameters
BThe Durbin-Watson statistic of 0.8 indicates strong positive autocorrelation — standard errors from OLS are too small because consecutive observations are not independent, inflating the significance of the intervention effect
CThe intervention effect is overestimated by 0.8
DThe pre-intervention trend was not linear
Time series data are typically autocorrelated — adjacent monthly observations are more similar than distant ones. OLS assumes independence of errors, producing standard errors that are too small when autocorrelation is present. A Durbin-Watson statistic well below 2 indicates positive autocorrelation. Solutions include Newey-West standard errors, generalized least squares (GLS) with an autoregressive error structure, or ARIMA-based ITS models. Ignoring autocorrelation leads to false significance — an apparent intervention effect that is really just serial correlation.
Question 3 Short Answer
ITS is often considered a strong quasi-experimental design even without a concurrent control group because the pre-intervention trend serves as the counterfactual. What is the main threat to this internal validity?
Think about your answer, then reveal below.
Model answer: A co-occurring event or change that happens at the same time as the intervention (a 'history' threat) could produce the observed change in level or trend. Without a control group, it is impossible to distinguish the intervention effect from other changes occurring simultaneously. For example, if a hospital implements hand hygiene protocols at the same time a new antibiotic is introduced, the ITS cannot separate their effects. Adding a control series (a similar hospital without the intervention) substantially strengthens the design by controlling for co-occurring temporal events.
This is the fundamental limitation of single-group ITS designs. The pre-intervention trend controls for existing trajectories but cannot account for new events coinciding with the intervention. The addition of a control series creates a controlled ITS (similar to DiD with time series data), which is considered one of the strongest quasi-experimental designs because it combines the trend-based counterfactual with a between-group comparison.