Structural Multicollinearity in Interrupted Time Series Designs

Interrupted Time Series (ITS) analysis is one of the most widely applied quasi-experimental designs in health, policy, and social research. ITS models include a structural dependency between time-related predictors that can give rise to substantial multicollinearity. This feature arises from the deterministic construction of the post-intervention time variable and intensifies mechanically as the post-intervention period lengthens. As a result, collinearity is not merely a sampling artifact but a structural feature of the ITS model. To illustrate this issue, we replicate a widely cited ITS application and assess collinearity diagnostics across SAS, Stata, and R. The findings confirm that variance inflation factors for time and post-intervention terms routinely exceed conventional thresholds, demonstrating that the problem is embedded in the model architecture itself. Although mean-centering does not eliminate this dependence, alternative parameterizations may alter how it appears in numerical diagnostics. Although ITS is frequently recommended as a rigorous design for policy evaluation, this often-overlooked structural limitation has not been explicitly addressed in methodological literature. We argue that greater awareness is needed, urge researchers to report collinearity diagnostics, interpret post-intervention slopes with caution, and recommend greater diagnostic transparency and model sensitivity checks to help applied researchers recognize and account for this structural dependency.