Table 1.
Summary of related works.
Fig 1.
Scheme of the SEIRD model.
Fig 2.
Visualization of the unnecessary SEC.
The green points represent the NPI plans with the optimal trade-off between the infection rate and the SEC, whereas the blue point represents the country’s actual implemented plan.
Table 2.
Uniform social, economic and aggregated cost for the 12 OxNPIs, i.e., uniform stringency multiplied by maximal strictness of each NPI. Economic cost is shown as a relative share of GDP loss in the period the NPI was implemented. The social cost is based on domain knowledge and expressed on a 1–12 scale. The aggregated column is the average of the two costs, when both are normalized to be between 0 and 1.
Table 3.
Uniform aggregated SEC for the 12 OxNPIs in the 17 studied countries over the study period. Italics denote the smallest and bold the largest number in a column. Abbreviations: CR, Czech Republic; UAE, United Arab Emirates; UK, United Kingdom; US, United States.
Table 4.
Uniform social, economic, and aggregated SEC for the 17 studied countries over the studied period. Italics denote the smallest and bold the largest number in a column. Economic cost is shown as a relative share of GDP loss, whereas the social cost is based on domain knowledge. The aggregated cost is the average of the two costs after normalization. Abbreviations: CR, Czech Republic; UAE, United Arab Emirates; UK, United Kingdom; US, United States.
Fig 3.
Comparison of different OxNPI configurations used in France over the studied period.
The SECs were computed using uniform (first row) and country-specific (second row) stringency that was fixed over the whole studied period. In all cases, the green line depicts Pareto-optimal OxNPI configurations, while the red line represents the configurations with the worst trade-offs between the infection rate and SEC. The black dots denote NPIs implemented by decision-makers. The results are averaged over five trials. Note that the scale of the y-axis (infection rate) is the same for all the panels, while the x-axis scales vary due to the different cost units. Economic cost is shown as a relative share of GDP loss, whereas the social cost is based on domain knowledge. The aggregated cost is the average of the two costs after normalization.
Fig 4.
Comparison of different OxNPI configurations used by six selected countries over the studied period.
Their SECs were computed using country-specific stringency that was fixed over the whole studied period. The green line depicts the Pareto front, while the red line represents the worst OxNPI configurations. The results are averaged over five trials. Economic cost is shown as a relative share of GDP loss, whereas the social cost is based on domain knowledge. The aggregated cost is the average of the two costs after normalization.
Fig 5.
The influence of NPIs on the unnecessary costs.
Blue bars (left y-axis) correspond to the NPIs’ impact to the unnecessary costs based on the linear regression coefficients, whereas the red bars (right y-axis) correspond to the mean difference between the strictness of an OxNPI implemented in real life and the strictness of the Pareto-optimal country-specific OxNPI configuration with the closest infection rate. The values were computed based on the data from all investigated countries during the study period.
Fig 6.
Daily country-specific time-varying stringency for seven selected countries and three OxNPIs.
The first, the second, and the third column represents “C1 School closing”, “C2 Workplace closing”, and “H6 Facial coverings”, respectively. The results are averaged over five trials. Note that the y-axis limits are different for all panels.
Fig 7.
Comparison of the 17 selected countries by the number of deaths per 100k citizens vs. (top row) the mean uniform unnecessary cost and (bottom row) the mean country-specific unnecessary cost.
Economic cost is shown as a relative share of GDP loss, whereas the social cost is based on domain knowledge. The results are averaged over five trials. The aggregated cost is the average of the two costs after normalization. Note that the x-axis limits are different for all panels.
Table 5.
Mean uniform and country-specific unnecessary costs for the 17 studied countries, i.e., costs that could be avoided with different OxNPI configurations while still having the same infection rate. The country-specific stringency was fixed over the whole studied period. The county-specific costs are averaged over five trials. Italics denote the smallest and bold the largest number in a column. Economic costs are shown as a relative share of GDP loss. Abbreviations: CR, Czech Republic; UAE, United Arab Emirates; UK, United Kingdom; US, United States.