The COVID-19 Global Policy Simulator analyzes the intersection between COVID-19 related policies, contagion rate, and government responses’ medium term effects on society in order to assess the full range of effects that may arise from implementing various policies in response to the global pandemic.
The model contains three main components:
- Intensity and Influence: How intensely have different countries implemented different policies? Which policies have been more influential are reducing contagion rate? Which indicators best predict whether a certain policy may be more or less effective at reducing contagion?
- Contagion Forecast: How will different policies impact contagion?
- Effects on Society: How will different policies impact society in the medium term (4 years)?
Our policy data comes from reliable and open sources such as major international organizations, NGOs, and official government statistics. All sources for the Effects on Society portion of the dashboard are noted on the dashboard itself. The Sustain Models team strives to validate data and routinely updates data as new releases become available. However, please note that all data faces limitations such as under/over-reporting and lack of investment into statistical capacity by some countries. Sustain Models does not collect any independent data for this model.
Different methodologies were used for different parts of the dashboard, hence the COVID-19 Sustain Model is actually made up of several models. The Contagion Forecast portion utilizes time series probability forecasting using three types of data to predict the COVID-19 series, including indicator data, policy data and COVID-related data. Indicator data reflects the country context (e.g. poverty rate, inequality, education, health); Policy data contains the policies the country have taken to fight the disease epidemic (e.g. quarantine, closure of schools, social distancing, travel bans); COVID data are the COVID-19 confirmed, cases. The basic model adapted is the improved multilayer perceptron. This model was selected given its superior performance in comparison to other options such as DeepAR and Gaussian distribution in probability forecasting.
An adapted Pearson’s correlation coefficient was used to measure impact and a Decision Tree Regression model was used to build the relationship between all the policies (independent variables) and the contagion rate (target variable) of all the countries. After that, Gini Importance was used to calculate the importance of each policy in the model.
The PIS categorizes each country based on how intensely each policy was implemented. There is the overall score and the score for each month since the first COVID-19 case was detected in a given country. The score ranges from 0 to 100, the closer to 100 the more intensely it was implemented.
The data for the Policy Intensity Score is comes from the Oxford Covid-19 Government Response Tracker (OxCGRT) and scores are computed by the Sustain Models team. The scores for Vaccination Policy also take into account percentage of people vaccinated.
Two sets of calculations were implemented to measure the policy intensity scores. One set measures the average policy intensity across the whole sample period. The other set measures policy intensity by month, wherein the first month corresponds to the first 30 days since the date of the first confirmed case. Each month is equivalent to 30 days.
Note that the date of first confirmed case is used as the reference period for all policies, except for vaccination policy. Given that vaccine development and production takes time, the starting period used to calculate the score for vaccination policy is set as the date of first registered vaccination.
These scores are updated at the end of every month.
The Country-Specific Influence Ranking ranks policies from 1 to 15 based on how well their implementation was able to “predict” a decrease in COVID-19 cases. The closer to 1 the greater the influence, the closer to 15, the weakest. A policies’ rank does not speak to its implementation nor does a lower indicate that said policy is necessarily less useful. This score is a snapshot of that policy’s influence on contagion in that specific country till date, changes in government strategies, in implementation, and in execution may produce a different result in the future. This ranking was obtained through a mix of deep learning methods and also takes into account the intensity of policy implementation.
No, the Country-specific Influence Ranking is useful only in gauging the influence of any given policy at reducing contagion in that given country. This ranking should not be used to compare across countries.
The Predicting Indicators are the indicators that best “predicted” whether the Country-Specific Influence Ranking would be high or low. These indicators aim to provide insight into why a certain policy might have ranked higher or lower in one country versus another.
Example: One of the top 50 indicators for the policy “School Closures” is “Individuals using the Internet”. Here, you will find the reported % of the population in that country with access to the internet according to the latest available data.
Within the Contagion Forecast subsection, Overall Cases refers to all COVID-19 cases reported from the beginning of the pandemic till present day, plus the projected cases. It includes both active and inactive cases. Hence, you may observe that the Overall Cases will increase regardless of the policy mix selection, the difference lies in how much they will increase by.
Conversely, Daily Cases refers only to the new cases reported each day, without the historical cases. Differently from the Overall Cases, Projected Daily Cases may be lower or higher than the current Daily Cases (observed).
The Pre-set Policy Mixes allow you to apply the reported policy intensity that each of the 10 included countries reportedly implemented during the peak of the COVID-19 pandemic in their country (greatest number of new cases) or during the two weeks directly afterwards. This pre-sets allow you to explore what the effect would be in the country of your choice, if that same policy intensity mix that was applied in one of the example countries were to be applied in your selected country. The slider will automatically adjust to the position that corresponds to the intensity with which the pre-set country implemented all 15 policies and the calculation will be made automatically.
NOTE: Some countries in the pre-sets had their “peak” prior to the existence of a COVID-19 vaccine. As a result that slider will be set to zero, this will impact the forecast. Feel free to adjust the Vaccination Policy slider to the appropriate level in order to get a more current result.
Given that the Contagion Rate forecast is only one month into the future, no duration slider is available. The forecast duration was selected for maximum accuracy and longer lasting effects are subject to maintaining policy measures and quality controls in place, as well as subject to new waves of COVID-19 and new variants. Given that the Effects on Society portion forecasts 4 years into the future, the option to select duration is included. This is also due to the fact that contagion rate data is released daily or near daily, while most of the data for the indicators included in Effects on Society is released annually.
If a policy slider is disabled, it means that a relationship between the selected indicator and the policy in question could not be ascertained through regression. As a result, that policy has not been assigned any value in terms of its effect on the forecast (moving the policy slider will produce no change).
Many indicators could not be included in the model for the following reasons:
1. Lack of 2020/2021 data
2. Insufficient evidence of relationship between COVID-19 policies and the indicator at hand.
3. Source lacked credibility
It’s also possible for some indicators within a “set” (i.e. stemming from the same source/study) to have passed our internal tests while others did not.) If there is a certain indicator that you would like to see included and for which 2020 and/or 2021 data is available, please e-mail us and we will review it: covid19model@sustainmodels.com
For media use, you may simply cite the COVID-19 Global Policy Simulator, Sustain Models.
Full recommended Citation:
Ayala, Garcia, Naniong, Tirado, Tolin, Zhou. (2021). “COVID-19 Global Policy Simulator.” Sustain Models. Accessed on:______. www.sustainmodels.com
