Case Study: Analysis of COVID-19 Policies

Embracing the challenge of discovering
effective policies for COVID-19

Use the Trial Tool to analyze the effectiveness of your country or region’s policies

This article describes how Wide Learning™ analytical functionality, which has been described in a number of other case studies to date, can be used to analyze the characteristics of countries and regions (generically referred to as “characteristics” in this article) along with a variety of implemented policies to discover combinations of policies that are effective at limiting growth in the number of COVID-19 infections when paired with those characteristics.

* This case study is based on data that was available as of August 2020.

Do efforts to stimulate economic activity cause the number of infections to rise?

Around the spring of 2020, the number of people infected with the novel coronavirus, COVID-19 began to surge worldwide. Infections subsequently continued to increase, spurring national and regional governments to undertake a variety of initiatives to bring the situation under control.

A global discussion followed as people questioned whether to impose restrictions on travel and economic activity in order to avoid the risk of infection or stimulate a stagnating economy. The backlash against a variety of restrictions adopted in response to a second wave of infections starting during the summer of 2020 was particularly intense in Europe, and the issue continues to be a major point of contention in European society as of this writing.

Policies promoting domestic travel in order to jump-start economic-activity have also generated significant interest in Japan, and most readers will be aware of the fact that these policies, while being welcomed by the tourism industry and related sectors, have prompted an outpouring of concern and news coverage. Artists in the entertainment industry, as well as their fans, may have wanted to see restrictions relaxed even earlier.

Are there regions where one policy has generated different results?

Examples of policies designed to limit growth in the number of infections through social distancing include closures of public transportation and requirements of staying at home.
How can the effectiveness of such policies be analyzed?
For example, some countries that imposed the same type of requirements of staying at home, for example Japan and New Zealand, saw little increase in infections, while others, for example the United States, India, and Brazil, saw the number of infections rise into the millions.

It's difficult to assess causal relationships.

Research to date has proposed factors that might be related to growth in the number of infections, including regional classification (Asia, Europe, etc.), BCG immunization rates, and per capita GDP.
In fact, the effectiveness of policies may depend on combinations of such characteristics, and the fact that various policies are implemented in combination raises the possibility of complex interrelationships and interactions.

Assessing how individual characteristics and policies are related to the limitation of infections is no simple task.

Discovering important combinations of characteristics and policies that are related to the limitation of infections

The number of possible combinations is given by 2n, where n is a number of foctors.
When dozens of factors are analyzed, the scales are difficult to process, even when computers are used.
(For more information, see Four Characteristics of Wide Learning™.)

Consequently, in most past research scientists have chosen a range of combinations to analyze in advance so that a smaller and more limited quantity of data can be analyzed.
However, such work requires a high level of specialized expertise, and past knowledge does not necessarily offer hints when addressing novel problems like COVID-19(*1).
As a result, important targets for investigation may be overlooked.

Wide Learning™ provides a method for quickly discovering clues from enormous sets of data.

Wide Learning™ makes it possible to investigate combinations of astronomy-scale numbers of factors such as these and discover important combinations quickly by streamlining search order and facilitating effective use of intermediate search results.

In this article, we will attempt to discover what combination of policies was associated with the increase in the number of infections in a country (region) with certain characteristics.
For this purpose, we will use the information about whether each region was able to limit growth in the number of infections during the identified period (1: POS [positive] or 0: NEG [negative]) as a solution label based on the rate of increase in the number of infections in each region, and we will make software learn the status of the above policies and characteristics.

*1 : The WHO has declared that “This virus is not SARS, it's not MERS, and it's not influenza. It is a unique virus with unique characteristics.”

Data used

The following data is necessary in order to use the Wide Learning™ Trial Tool. Please download the linked files.

Data for learning the software about policies for limiting growth in the number of infections:
Data for learning the Wide Learning™ system (COVID_train.csv)


Data for forecasting policies for limiting growth in the number of infections:
Data A for facilitating Wide Learning™ forecasts: When previously unimplemented policies are implemented (COVID_test a_increment.csv)


Data for forecasting policies for limiting growth in the number of infections:
Data B for facilitating Wide Learning™ forecasts: When previously implemented policies are moved forward (COVID_test_b_advance.csv)

Table 1 below summarizes characteristics and policy implementation status aggregated from publicly available data for the U.S. as of March 25, 2020. Before and after data for the surge in the number of infections in various countries and regions worldwide can be collected every 4 days as a base date and used to perform analyses with Wide Learning™.

Table 1. Example characteristics and policy implementation status
Data label Country Base Date日 United_States_Mar_25_2020
Predictor variable [Characteristic] Region North America
[Characteristic] Vaccinated against BCG ―
[Characteristic] GDP per capita >$19443
[Characteristic] Over 65 years old >8.5%
[Characteristic] Average life span ≤78.7 years old
[Characteristic] Smokers >14.3%
[Characteristic] Urban dwellers >55.3%
[Characteristic] Wage earners >76.8%
[Policy] School closing >3 weeks
[Policy] Restrictions on gatherings 2 to 3 weeks
[Policy] Workplace closing 2 to 3 weeks
[Policy] Stay at home requirements >3 weeks
[Policy] Restrictions on internal movement >3 weeks
[Policy] Close public transport >3 weeks
[Policy] Cancel public events >3 weeks
Response variable Solution label 1

Data details

Discovering important combinations of characteristics and policies

The predictor variable for policies (“2 to 3 weeks/>3 weeks”) was created to serve as an indicator for determining whether the policy in question impacted infections within one week of its implementation or after that period, based on the assumption that the incubation period that follows infection before the onset of symptoms means that about two weeks elapse before a given patient generates a positive test result.

Data about the policy implementation status for various countries and regions, the number of infections reported, and national/regional characteristics was drawn from the Oxford University COVID-19 Government Response Tracker（*2）.
Indicator data from the World Bank（*3）was also used to assess national and regional characteristics.

For more information about data and how to create response variables, see the Supplemental Explanation at the end of this article as well as information from the list of related sources.

*2 : Oxford Covid-19 Government Response Tracker (OxCGRT)
*3 : Indicators - World Bank Open Data

Step1: Create data describing policies implemented
by each country or region to limit the spread of the virus
as well as associated results

Uploading learning data

When you upload learning data, the contents of that data will be displayed as a table. Each row will display data about the characteristics for a specific region and time period along with the implementation status of policies.

For more information about how to review statistical information about learning data, see “1-3. Check the data for learning.” For more information, see “Viewing Statistical Information about the Mammal Classification Case Study.”

Results of uploading learning data

Step 2: Analyze policy combinations in light of national characteristics

Generating combinations of important parameters

Next, click the “Proceed to Step 2” button. (The generation process will take some time.)
Wide Learning™ will select and display combinations of parameters that it has determined are related to infection growth.
Each row displays information about one combination, and the leftmost column displays important parameter combinations.

Results of generating important parameter combinations

Since the focus of this case study is the discovery of more important combinations, it introduces results obtained after the limitations on the Wide Learning™ Trial Tool have been removed (i.e., results obtained using the normal version of the software). These results tag more combinations for analysis than results obtained from the Trial Tool (which is limited to a maximum of three combinations).
The number of types and quantity of input data have been increased, and detailed data for which output was omitted is also used in the analysis.

Analysis of characteristics and policy results

Let’s try analyzing some of the important parameter combinations discovered by Wide Learning™. The goal in this case study, which analyzes COVID-19 policies, is to discover policy combinations and regional characteristics that were related to growth in infections, so we’ll focus on combinations that include both characteristics and policies in our analysis. That analysis will target the example combinations listed in Table 2 below.

Table 2. Example combinations targeted by our analysis
Pattern Combination type Example combination output
1 1 characteristic and 1 policy Characteristic A ∧ Policy X
2 1 characteristic and 2 policies Characteristic A ∧ Policy X ∧ Policy Y
3 2 characteristics and 1 policy Characteristic A ∧ Characteristic B ∧ Policy X
4 1 characteristic and 3 policies Characteristic A ∧ Policy X ∧ Policy Y ∧ Policy Z
5 2 characteristics and 2 policies Characteristic A ∧ Characteristic B ∧ Policy X ∧ Policy Y

* The highlighted data cannot be obtained using the Trial Tool.

As an initial, simple analytical method, we’ll narrow Patterns 1, 2 and 4 in Table 2 to a single selected characteristic and analyze which policy combinations were related to growth in infections in a progressive manner for each policy. There are a total of 16 types of characteristics, but here we’ll introduce analysis results while paying attention to countries with a comparative urban population ratio, which includes Japan, as one of the most distinctive characteristics in the list we’re examining.

Analyze policy combinations for countries with a high percentage of urban-dwelling residents

Table 3. Example combinations for countries with a comparatively urban population ratio
Characteristic Policy combinations Length Number of
positive
occurrences		 Number of
negative
occurrences		 Confidence
Urban dwellers
> 55.3%		 2-3 weeks after Restrictions on internal movement *1 2 72 49 0.60
2-3 weeks after School closing 2 52 34 0.60
2-3 weeks after Close public transport
∧ 2-3 weeks after Restrictions on gatherings		 *2 3 25 12 0.68
2-3 weeks after Restrictions on gatherings
∧ 2-3 weeks after Stay at home requirements		 3 28 16 0.64
2-3 weeks after Close public transport
∧ 2-3 weeks after Workplace closing		 3 31 19 0.62
2-3 weeks after Workplace closing
∧ 2-3 weeks after Stay at home requirements
∧ 2-3 weeks after Cancel public events		 *3 4 16 2 0.89
2-3 weeks after Restrictions on gatherings
∧ 2-3 weeks after Workplace closing
∧ 2-3 weeks after Restrictions on internal movement		 4 21 5 0.81
2-3 weeks after Workplace closing
∧ 2-3 weeks after Cancel public events
∧ 2-3 weeks after Restrictions on internal movement		 4 20 5 0.80
2-3 weeks after Workplace closing
∧ 2-3 weeks after Stay at home requirements
∧ 2-3 weeks after Restrictions on internal movement		 4 29 11 0.73

* The highlighted data cannot be obtained using the Trial Tool.

Table 3 extracts combinations between the characteristic of an urban population ratio and policies from the combinations discovered by Wide Learning™ and lists the number of examples that either satisfied or did not satisfy the condition as positive (“POS”) and negative (“NEG”), respectively. The table includes confidence figures, a type of data that is not output by the Trial Tool.
Confidence is indicated as the percentage of examples that matched the characteristic and policy from the learning data and that were positive.
For example, the combination of “urban dwellers is over 55.3%” and “2 to 3 weeks after Restrict_on_internal_movement” is given as (72 / 121 = 0.60 [60%]) since 72 of 121 examples were positive (72 positive versus 49 negative).
Although the Trial Tool only outputs combinations up to a length of three, the table includes combinations with a length of four, which can be output when using the normal version of the software.

*1: Confidence of about 60% for a single policy

Let’s examine Table 3.
Looking at combinations with a length of two, we see that in regions with a high urban population ratio, growth in the number of COVID-19 infections was limited with confidence of about 60% in the following examples: “2 to 3 weeks after Restrictions on internal movement” and “2 to 3 weeks after School closing.”
Based on this fact, restricting domestic travel and closing schools may be effective in limiting growth in the number of COVID-19 infections in regions with a high urban population ratio.

*2: Confidence of somewhat greater than 60% when two policies are combined

Looking at combinations with a length of three, we can verify that in regions with a high urban population ratio, growth in the number of COVID-19 infections was limited with confidence of about 68% in the following examples: “2 to 3 weeks after Close public transport” and “2 to 3 weeks after Restrictions on gatherings.”
When imposing closures on public transportation in regions with a high urban population ratio, it may be even more effective when those policies are accompanied by restrictions on gatherings.

*3: Confidence of close to 90% when three or more policies are combined

Finally, let’s examine combinations with a length of four, which are output by the normal version of the software. When implementing at least three policies, we can see that the software output a number of combinations with a confidence value of at least 0.8. In particular, the results indicate that combining “2 to 3 weeks after Workplace closing,” “2 to 3 weeks after Stay at home requirements,” and “2 to 3 weeks after Cancel public events” limited growth in the number of COVID-19 infections with confidence of close to about 90%.
In regions with a high urban population ratio, combining at least three policies can be expected to effectively limit growth in the number of COVID-19 infections, and the three combinations listed above can be expected to prove particularly effective.

Impact analysis of multiple characteristics and policies

In this way, we were able to discover the effectiveness of policies on a region-by-region basis through an analysis that focused on a particular characteristic. Next, let’s undertake a more sophisticated analysis of characteristics that can be discovered by combining multiple characteristics. We've used the normal version of Wide Learning™ for this analysis, too, in order to increase the types and quantity of input data.

Here we’ll examine the combinations that correspond to Patterns 3 and 5 in Table 2 in order to group combinations that include pairs of the same characteristics so that we can compare numbers of solution labels, which indicate whether the combinations in each group were related to limiting growth in the number of infections (1: POS) or whether they were related to conditions in which such growth could not be limited (0: NEG).

Based on characteristics groups that most combinations were negative and that only yielded a small number of positive combinations, we can conclude that most policy combinations cannot limit growth in the number of infections in regions with the corresponding characteristics, although we can nonetheless identify certain policy combinations that were more effective than other combinations.

Important combinations identified by Wide Learning

Table 4. Combinations that were related to limiting growth in the number of infections
in regions with low population density and high incidence of diabetes
Characteristic Policy combinations Number
of
positive
occurrences		 Number
of
negative
occurrences		 Confidence Number
of
countries
Population density
<= 66.9/km2
∧ Diabetes incidence
> 7.1%		 No Restrictions on gatherings
∧ 2-3 weeks after Cancel public events		 15 2 0.882 10
No Stay at home requirements
∧ No Close public transport
∧ 2-3 weeks after Cancel public events		 14 2 0.875 10

Discovering six effective combinations from a total of 663 combinations

For example, although Wide Learning™ extracted 663 combinations for the group that includes “low population density” and “high incidence of diabetes” as characteristics, most of those combinations were negative, meaning that they were not useful in limiting growth in the number of COVID-19 infections. Just six of the combinations yielded positive results, allowing them to be classified as combinations that were related to limiting growth in the number of infections.
Table 4 offers two representative examples obtained by omitting factors with redundant meanings from these positive combinations (there were 10 corresponding countries (*4 )).
These combinations support the conclusion that in regions with low population density and a high incidence of diabetes, canceling public events without imposing restrictions on gatherings, requirements of staying at home, or public transportation closures can be expected to yield results.

In this way, Wide Learning™ can conduct a comprehensive review of combinations of an enormous number of factors and then refine the pool to those that could be useful in classification, making it possible to easily calculate positive/negative bias in specific groups and simplifying the process of discovering combinations of distinctive factors.

*4 : Nine countries met the conditions for a positive result (Brazil, Canada, Chile, Columbia, Iran, Panama, Saudi Arabia, South Africa, and the U.S.), while Mexico met the conditions for a negative result.

Summary of our factor analysis

In this way, Wide Learning™ can comprehensively and quickly identify combinations of factors that are useful in determining whether growth in the number of COVID-19 infections can be limited. As a result, we were able to investigate what sort of policies were useful in the context of a variety of national and regional characteristics by analyzing confidence values and positive/negative bias for extraction results.

Based on the simple analysis described in this profile, we can conclude that in regions like Japan with a high urban population ratio, effective combinations of policies that could be undertaken when the number of infections rises include combinations such as the following: (a) domestic travel restrictions, school closures, and restrictions on public transportation and gatherings; and (b) temporary business closures, domestic travel restrictions, and cancellations of public events.

Although in this instance we analyzed factors that are effective when the number of infections is growing by treating the period when the rate of growth in infections fell as positive, we could treat the period when the number of infections remained low as positive in order to discover factors that either helped or did not help keep numbers low during a period of limited infections.

In this way, Wide Learning™ offers a way to analyze a variety of factors comprehensively and quickly for seemingly opaque societal issues, making it possible to discover effective knowledge.

Step3 to 5: Let’s forecast conditions in different regions using the combinations of factors that were identified

Of the combinations of important parameters discovered using the Trial Tool, those with large impacts can be used to learn the AI. In this way, it’s possible to predict whether those combinations would be able to limit growth in the number of infections when applied to new national/regional characteristics and policy implementation status data.

We’ll use the prediction data as input for the Wide Learning™ Trial Tool in order to predict whether policy combinations would be able to limit growth in infections in different regions and time periods. Click the “Proceed to Step 3” button on the Step 2, “Let's check the combinations of important items!” screen.

Weight Important Combinations

Step3: Checking the importance of each combination

On the next screen, you can view the relative weights for the important parameter combinations reviewed above. In this context, “weight” quantifies the importance of each combination of parameters so that we can determine whether they are related to limiting growth in the number of infections. Weight values are the result of determining that positive combinations are related to limiting growth in infections, while negative combinations are related to conditions in which infections were not limited.

Step4, 5: Forecasting whether the number of infections will grow for new regions and period data

Using these results, let’s forecast whether growth in the number of infections will be limited in each case by providing new region and period data as input.
Here we’ll use two sets of predictive data.

  • COVID_test_a_increment.csv
  • COVID_test_b_advance.csv
Upload the forecast data to the Wide Learning™ Trial Tool by dragging and dropping the files.

Test data A: Estimation of results if previously unimplemented policies were implemented (India)

The following screenshot shows the interface after data estimating what would happen if previously unimplemented policies were implemented (COVID_test_a_increment.csv) has been uploaded.
This data describes the implementation status of the following policies in India:

  • Data 1: If policies remained unimplemented
  • Data 2: If public events were canceled
  • Data 3: If workplace closures were imposed
  • Data 4: If workplace closures were augmented by restrictions on gatherings and cancellations of public events
  • Data 5: If the workplace closures and cancellations of public events described in Data 4 were replaced by school closures and public transportation closures

Verify that the data has been uploaded and click the “Proceed to Step 5” button.
In “Step5: Let's check the results of prediction,” you can check the results and scores for the regions and periods in the forecast data.
Cases with a score that is greater than 0.5 were forecast as positive (indicating that they limited growth in the number of infections). When you select one of the forecast data sets, you can review the important parameter combinations that correspond to that data set along with their importance (weight).

The five data sets provided in Test A support the following results:

  • Data 1: Negative judgment if policies remain unimplemented
  • Data 2: No change in score even if cancellations of public events are implemented
  • Data 3: Increase in score if workplace closures are implemented
  • Data 4: Positive judgment if restrictions on gatherings and cancellations of public events are added to workplace closures
  • Data 5: Negative judgment if workplace closures and cancellations of public events are replaced by school closures and public transportation closures

Let’s review those judgment results along with the corresponding parameter combinations.

Data 1 (policies unimplemented) and Data 2 (public event cancellations only) include no important parameter combinations that qualify as positive.
The score, 0.394, is less than 0.5, yielding a negative judgment.

By contrast, Data 3 (workplace closures only) yields a number of positive combinations for which Indian characteristics and workplace closures are related, pushing the score up to 0.427.
There’s no change in the negative judgment, but the result indicates that Wide Learning™ has determined that the potential for limiting growth in the number of infections has risen compared to the previous two data sets.

Data 4 (in which workplace closures have been augmented by restrictions on gatherings and cancellations of public events) yields new positive combinations that are related to restrictions on gatherings, and the resulting score exceeds 0.5, flipping the judgment to the positive side.
That said, no combinations related to cancellations of public events have been output.

Finally, Data 5 (in which the workplace closures and cancellations of public events in Data 4 have been replaced by school closures and public transportation closures) yields a score that fails to exceed 0.5, even though the same three policies were implemented, and the result is a negative judgment.
It's clear that combinations including school closures in particular appear only on the negative side. In this way, it’s possible to search for policy combinations that suit specific countries and regions by using Wide Learning™ to predict whether those combinations will limit growth in the number of infections based on data that anticipates conditions when specific combinations of policies are implemented.

Test data B: Estimation of results if existing policies had been implemented sooner (U.S.)

Next, we’ll introduce an example analysis of data that hypothesizes what would happen if existing policies had been implemented at a different time (COVID_test_b_advance.csv).
Countries and regions choose policies to address the pandemic and determine when to implement them based on their respective conditions.
Let’s leverage the benefits of hindsight to predict how much the surge in infections could have been limited (i.e., whether the surge in infections could have been shortened, or whether a stable plateau could be achieved sooner) if some or all of those policies had been implemented sooner.

There are a number of techniques that could be used to test this hypothesis, but here we’ve created test data for all 127 combinations (27 - 1) based on the assumption that the cost of moving a policy forward a certain amount of time from the date it was implemented is the same for all policies and that any of seven policies could be either moved forward in that manner or not. We then compared the results of forecasting whether growth in the number of infections would have been limited based on that data with the results as of the date the policy was actually implemented.

The above figure illustrates the relationship among the following: a label indicating whether growth in the number of infections in the U.S. could be limited (solution label: solid black line and first row in table), forecast results for the data using the date the policy was actually implemented (score: light green line on graph; forecast results: second row in table), and the pattern that was forecast to most limit growth in the number of infections in the hypothetical data in which policies were moved up one week (score: dark green line on graph; forecast results: third row in table).
As with March 5 and 9 in the figure, we've chosen the best-case scenario with the longest period of time during which infection growth was forecast to be limited by moving up the policies (positive) for the periods in which infection growth was not in fact limited (negative) and in which it was forecast that such growth would not be limited for the actual policy implementation status (negative).

In particular, we’ve prepared data for March 5 and 9 for the scenario in which school closures alone were moved up one week (the scenario that required the fewest policies to be moved up) for use as forecast data for the Trial Tool (COVID_test_b_advance.csv).
Data 1 and 2 describe the results of leaving the actual implementation dates unchanged (for comparison purposes), while Data 3 and 4 describe the results of moving school closures up one week.

As before, click the “Proceed to Step 5” button to forecast whether growth in the number of infections could be limited for this data.

The left table indicates whether each data set was determined to be positive or negative and provides the score for each, allowing you to click each data set and review the corresponding important parameter combinations.

Would moving up policies have limited infection growth?

Let’s analyze why moving up school closures in the U.S. by one week causes the forecast of whether infection growth on March 5 could have been limited to change from negative to positive based on the disparity in the corresponding combinations.
The following tables indicate whether moving up school closures caused differences in important parameter combinations.

Newly qualified important parameter combination
Factor combination Weight
North America
∧ Large number of smokers
∧ 3 weeks or more after School closing		 0.4817
North America
∧ Low average live expectancy
∧ 3 weeks or more after School closing		 0.4817
No longer qualified important parameter combination
Factor combination Weight
High per capita GDP
∧ 2 to 3 weeks after School closing		 0.6076
High urban population ratio
∧ 2 to 3 weeks after School closing		 0.3683
Large population of people age 65 and older
∧ 2 to 3 weeks after School closing		 0.1785
Large population of high-income earners
∧ 2 to 3 weeks after School closing		 -0.0709
High per capita GDP
∧ Low average life expectancy
∧ 2 to 3 weeks after School closing 		-0.3871
North America
∧ Large number of smokers
∧ 2 to 3 weeks after School closing		 -0.5313
North America
∧ Low average life expectancy
∧ 2 to 3 weeks after School closing		 -0.5313

The tables extract combinations related to school closures. In particular, the evaluation of conditions more than three weeks after school closures has reversed from the corresponding evaluation two to three weeks after school closures in North American regions where there are large numbers of smokers and in North American regions where average life expectancy is low (from about -0.5 to about +0.5), and this change is greater than changes in other weights.

Consequently, the learning data we’ve prepared suggests that the effects more than three weeks after school closures in the U.S. would be greater than the effects two to three weeks after those closures, and that conclusion is based on scenarios in North American regions where there are large numbers of smokers and North American regions where average life expectancy is low.

In this way, the combinations of important characteristics and policies discovered by learning can be used to forecast growth in the number of infections in new regions or timeframes.

Reference

Research findings related to this analysis were presented at the following scientific conferences:

Title: Effectiveness Estimation of Interventions against COVID-19 Using AI
Effectiveness Estimation of Interventions against COVID-19 Using AI
Conference:The Japan Society for Management Information, National Convention of JASMIN 2020 Autumn (November 7)

Title: AI-based analysis of impact of non-pharmaceutical interventions against COVID-19 with respect to country/region features
Hypotheses for impact of non-pharmaceutical interventions against COVID-19 with respect to country/region features
Conference:“The Japan Society for Artificial Intelligence, 16th conference of Special Interest Group on Business Informatics (SIG-BI) (August 29)

Disclaimer

The analysis of government policy and its effects on limiting growth in the number of infections based on publicly available data described in this article was not undertaken with medical oversight.
Its results were intended to verify the effectiveness of the Wide Learning™ analytical tool, and they should be considered only as one example of how the relationships between societal issues and policy can be analyzed.

Data details

The following types of data were analyzed using the normal version of the software:

Characteristics
Regions (*5) BCG immunization rate Per capita GDP (USD)
Percentage of population age 65 and over Diabetes incidence Extreme poverty rate (*6)
Number of foreign travelers Average life expectancy Percentage of smokers
Percentage of urban dwellers Percentage of high-income earners Percentage of population living in hygienic environment
Percentage of population with access to clean water Percentage of population with access to hand-washing facilities Number of hospital beds per 1,000 people
Number of physicians per 1,000 people Population density  

* The highlighted data cannot be obtained using the Trial Tool.

Policies
School closing Workplace closing Restrictions on gatherings
Stay at home requirements
(restrictions on unnecessary movement) 		Cancel public events Restrictions on internal movement
Close public transport  

*5 : Following seven regions: East Asia/Pacific, Europe/Central Asia, North America, South America/Caribbean, Middle East/North Africa, South Asia, Sub-Saharan Africa (as classified by the World Bank).
*6 : Percentage of population with daily income of less than USD 1.9.

Try the Trial Tool

No membership registration is required, and the tool is free to use.
Experience Wide Learning™!

Try the Trial Tool 

Wide Learning™ Case Studies

Here we introduce some specific case studies of Wide Learning™.
These case studies have been prepared based on the level of difficulty.

Identifying
Mammals

For people aged 12 and over

Let's see if we can identify mammals
using information about animals.

Go to the Case Study

Discovery of
Electoral Factors

For people aged 16 and over

What are
the features
of people
who win elections?

Go to the Case Study

Analysis of
COVID-19 Policies

For people aged 16 and over

Let's try to discover effective ways to prevent the spread of infections.

Go to the Case Study

　Case 4　

---

Coming
Soon...

Go to the Case Study