We examined how policy effects changed in the second wave (late 2020), at a finer spatial level (units roughly 1/20 of a country rather than whole countries). Policies got a bit weaker overall (66% combined reduction vs 80% in spring). The best reading is that safety measures and individual protective behaviour persisted after the first wave, even when governments said it was OK to stop. School closure was notably weaker (10% vs 35%), probably because the safety measures enforced since spring really did make schools safer.
Converted from the open-access Nature Communications full text (Europe PMC, PMC8492703). Markup removed; equations summarised; figures/supplement and the reference list omitted.
Abstract. European governments use non-pharmaceutical interventions (NPIs) to control resurging waves of COVID-19. However, they only have outdated estimates for how effective individual NPIs were in the first wave. We estimate the effectiveness of 17 NPIs in Europe’s second wave from subnational case and death data by introducing a flexible hierarchical Bayesian transmission model and collecting the largest dataset of NPI implementation dates across Europe. Business closures, educational institution closures, and gathering bans reduced transmission, but reduced it less than they did in the first wave. This difference is likely due to organisational safety measures and individual protective behaviours — such as distancing — which made various areas of public life safer and thereby reduced the effect of closing them. Specifically, we find smaller effects for closing educational institutions, suggesting that stringent safety measures made schools safer compared to the first wave. Second-wave estimates outperform previous estimates at predicting transmission in Europe’s third wave.
A second wave followed the reopening of European societies (≈August 2020–January 2021). First-wave NPI effectiveness was measured relative to baseline contact patterns in the very early pandemic, when safety measures and protective behaviours were lacking — so first-wave estimates proxy how much transmission is associated with various areas of public life operated without safety measures, and are likely inadequate for an ongoing pandemic. After the first wave, contact patterns did not return to pre-pandemic normal (distancing, testing, ventilation), likely making areas of public life safer and reducing the additional effect of closures. Governments also need estimates for the finer-grained NPIs used later (specific business-sector closures, small gathering-size bans, nighttime curfews). Identifying individual effects requires a multinational, subnational dataset (national modelling would obscure local heterogeneity and risk ecological fallacies). Since existing trackers lack granular subnational data, we built a novel dataset across 114 regions in 7 countries (Austria, Czech Republic, England, Germany, Italy, Netherlands, Switzerland) and a semi-mechanistic hierarchical Bayesian model with a latent random walk and stochasticity for low case counts.
The combined effect of all NPIs was smaller in the second wave. All NPIs together reduced $R_t$ by 66% [61–69%], vs first-wave medians of 77–82%. The most stringent set actually implemented in each region reduced $R_t$ by 56% [40–64%] (vs 76–82%), even though second-wave NPIs were often similarly strict or stricter. These differences are likely explained by pre-intervention contact patterns, safety measures, and protective behaviours — population immunity, reduced adherence, and ascertainment changes seem less important.
A detailed assessment. Second-wave NPIs were more spaced out (only 23% of interventions starting in the same 10-day window vs 83% in the first wave), enabling finer-grained identification given the larger subnational dataset (9.2× more NPI implementations than the largest Europe-focused study). Key estimates (reduction in $R_t$):
Robustness. 17 sensitivity analyses across 86 experimental conditions (priors, structure, epidemiological delays, region resampling, unobserved confounders): some NPIs robustly outperform others, so high-level policy conclusions hold; but as an observational study, the true strength of unobserved confounding is unknown.
Generalisation across time. Second-wave estimates should predict current effects if safety measures/behaviours are similar (and they have been relatively stable through the second and third waves). Testing on third-wave national data (Jan–May 2021, influenced by B.1.1.7 and vaccination): first-wave estimates overestimate observed $R_t$ changes by ~18 percentage points on average, while second-wave estimates overestimate by only ~2 pp — though this experiment has limitations (confounding by VOC arrival or unrecorded NPIs).
Implications. Closures and bans still considerably reduced transmission in the second wave, but less than the first. First-wave estimates overestimate effectiveness in an ongoing pandemic. Educational institutions with appropriate safety measures can be made considerably safer; only the strictest gathering-size limits remain effective; face-to-face businesses still carry considerable transmission; and stricter mask policies and nighttime curfews can help. NPI effectiveness being dynamic in time is an important, under-discussed policy consideration; real-time modelling of evolving NPI effects should be a priority.
Data. A custom NPI dataset: 7 countries, 114 regions of analysis, 1 Aug 2020 – 9 Jan 2021, ~19,000 region-days, >5,500 NPI entries (each with start/end dates, quotes, and government/university/legal/media sources). Regions of analysis are the highest administrative division with uniform NPIs; whole countries for Austria/Czechia/Italy/ Netherlands, stratified random samples of 15 regions elsewhere (stratified by first-wave COVID deaths); regions with <2,000 cases excluded. Data collection used semi-independent double entry (each country by two authors; second-round entry with access to sources/quotes but not the first round’s NPI data), interviews with local epidemiologists, external-source validation, and automated checks — ~950 hours of manual collection. B.1.1.7-contaminated observations (>10% VOC) were excluded (mostly English regions).
Model. A semi-mechanistic hierarchical Bayesian model (adapting Brauner et al.): $R_{t,l} = \tilde{R}{0,l} \cdot \prod{i=1}^{I} \exp(-\beta_i x_{i,t,l}) \cdot \exp(z_{t,l})$, where $\tilde{R}{0,l}$ is the no-NPI reproduction number on 1 Aug 2020 (TruncatedNormal(1.35, 0.3²)), NPI effects $\beta_i$ are time-invariant and shared across locations with an Asymmetric Laplace shrinkage prior (80% mass on positive effects), and $z{t,l}$ is a per-location latent weekly random walk flexibly absorbing unobserved factors (e.g. unrecorded NPIs, low adherence). $R_{t,l}$ drives a discrete-renewal infection process with additive infection noise (important at low incidence); infections are convolved with infection-to-confirmation / infection-to-death delays (re-estimated from second-wave linelist data: onset-to-confirmation gamma mean 5.28 d; onset-to-death gamma mean 18.61 d; generation interval mean 4.83 d) and observed cases/deaths follow country-specific negative binomials. The model is invariant to time-constant IFR/IAR. Implemented in NumPyro with NUTS (4 chains, 250 warmup + 1250 draws = 5000 samples; no divergences). Data and code: github.com/MrinankSharma/COVID19NPISecondWave.