Uncertainty and inconsistency of COVID-19 non-pharmaceutical intervention effects with multiple competitive statistical models

Quantifying the effect of non-pharmaceutical interventions (NPIs) is essential for formulating lessons from the COVID-19 pandemic. To enable a more reliable and rigorous evaluation of NPIs based on time series data, we reanalyse the official evaluation of NPIs in Germany. As the first part of a multi-step validation and verification project, we focus on properly analysing statistical uncertainties for time series data. Using a set of 9 competitive statistical methods for estimating the effects of NPIs and other determinants of disease spread on the effective reproduction number $$\mathcal {R}(t)$$, we find significantly wider confidence intervals than the official evaluation. In addition to vaccination and seasonality, only few NPIs – such as restrictions in public spaces – can be confidently associated with variations in $$\mathcal {R}(t)$$, but even then effect sizes have large uncertainties. Furthermore, due to multicollinearity in NPI activation patterns, it is difficult to distinguish potential effects of NPIs in public spaces from other interventions that came into force early, such as physical distancing. In future, NPIs should be more carefully designed and accompanied by plans for data collections to allow for a timely evaluation of benefits and harms as a basis for an effective and proportionate response.