Suggested citation:
Alvarado, J., Allen, A., Medina-Cetina, Z., Pompelli, G., Cochran, M. & Z-Losoya, E. (2021). Actionable Information - Research Briefs - 3 - Analysis on U.S. States COVID-19 dashboards. https://r7-cbts-sgl.engr.tamu.edu/2021/05/01/actionable-information-research-briefs-3-analysis-on-u-s-states-covid-19-dashboards/
@misc{Alvarado2021b,
author = {Alvarado, Juan Pablo and Allen, Alexi and Medina-Cetina, Zenon and Pompelli, Gregory and Cochran, Matt and Z-Losoya, Enrique},
title = {Actionable Information - Research Briefs - 3 - Analysis on U.S. States COVID-19 dashboards},
url={https://r7-cbts-sgl.engr.tamu.edu/2021/05/01/actionable-information-research-briefs-3-analysis-on-u-s-states-covid-19-dashboards/},
year={2021},
month={May}
}The Research Brief #2, titled U.S., and Mexico COVID-19 Dashboards, presents an exhaustive list of COVID-19 Dashboards developed by official entities of each state (Alvarado et al., 2021). The research brief also shows the number of different variables included for on platform, presenting the information in a spatial manner for the U.S. and Mexico.
This Research Brief #3, and #2 on COVID-19 Dashboards, is focused on the analysis of the variables of U.S. dashboards from different perspectives and classifications. In order to do this, the sequence of activities of Fig. 1 was followed.
Figure 1: Sequence of activities.
As such, the following sections describe the dashboards, and their variables in terms of:
An additional section in this Research Brief compares the amount of information presented in each dashboard with social States of Risk such as cases and deaths. The final section summarizes the results and includes the key takeaways and conclusions of the analysis.
Table 1 lists the different platforms used by the State entities to develop their dashboard(s). As it can be seen, a total of 29 out of 50 dashboards, were developed using either solutions by ESRI or Tableau. This is followed by a minority of dahsboards developed in Microsoft Power BI, and some other solutions such as Looker, DataViz, Google APIs, and Plotly.
| Platform | # of dahsboards |
|---|---|
| ESRI (ArcGIS) | 22 |
| Public Tableau | 17 |
| Microsoft Power BI | 8 |
| Looker | 1 |
| DataViz | 1 |
| Google APIs | 1 |
| Plotly | 1 |
The Fig. 2 shows the total number of variables included in the U.S. State COVID-19 Dashboards. The color scale reflects the quartiles of the distribution. As such, 25% of dashboards present between 9 and 43 variables, 25% between 43 and 57 variables, 25% between 57 and 72, and a final 25% between 72 and 172.
By hovering over the States, a pop-up window will display the total number of variables, and a set of links to visit the dashboards.
Figure 2: Number of variables presented in U.S. State Dashboards.
To take a better look at the distribution of the number of variables, a relative frequency histogram is included in Fig. 3. Below, some descriptive statistics are shown:
Figure 3: Relative frequency histogram for total number of variables.
The variables in the dashboard were further classified according to the type of evidence. In Bayesian inference evidence can come from observations, model predictions, expert beliefs, or a combination of them (Medina-Cetina et al., 2016).
After classifying the variables by type of evidence, it is possible to calculate the percentage for each one with respect to the total number of variables in each dashboard. These percentages are:
Most of the variables come from observations from evidence on cases, deaths, and the capacity of health system services. A percentage less than 10% are model predictions, with variables such as the effective reproduction rate or models to predict deaths, and ICU bed occupancy. The only State that present expert beliefs is California, that includes expert opinions to adjust the value of daily COVID-19 tests.
The next classification of variables is by type. The different types are:
Fig. 4 shows a boxplot for the 4 different types of variables. The distributions for the 4 types are similar, meaning that the variables in the dashboard are roughly equally divided in the 4 categories. Notably, given the nature of the pandemic and the development of the dashboards, there is a considerable number of variables that change in both time and space such as the time series of COVID-19 cases by each county in a state.
Figure 4: Type of variable distribution.
According to the Bayesian Risk definition, and framework from (Medina-Cetina & Nadim, 2008), Risk is defined as:
Risk = [Hazard] * [Vulnerability] * [Consequences]
Where:
Variables required to estimate the Hazard, Vulnerability, and Consequences, have to depict the threats, the systems, and the impacts respectively.
The Risk Assessment framework is expanded into the following for representing Risk Assessment & Management:
Risk = [H*V*C] + [Cost(AC+PC)−AC−PC]
Where:
In order to reduce the state of risk, mitigation strategies such as active countermeasures (AC), and passive countermeasures (PC) can be applied.
Considering Risk Assessment and Management, the 5 Risk Components are:
Fig. 5 shows a boxplot for the variables classified in the 5 Risk Components. The figure shows a higher number of variables for Systems and Mitigation across the 50 dashboards, with median values of 20 and 19 respectively. In contrast, Impacts and States of Risk have a median value of 7.
Figure 5: Classification by Risk Component.
Examples of variables in each Risk Component are:
Fig. 6 is a spatial representation of the number of variables on each Risk Component. By hovering over a state, a pop-up window will display the total number of variables for the corresponding Component.
Figure 6: Map of variables classified by Risk Component.
In order to assess the relationship between the amount of information presented in the dashboards with the management of the pandemic in the 50 states, Fig. 7, and Fig. 8 present a scatter plot of the total number of variables (and the number of variables for the 5 Risk Components) and the COVID-19 cases and deaths per 100k population.
Figure 7: Number of total variables, and variables per Risk Component compared to cases per 100k.
Figure 8: Number of total variables, and variables per risk components compared to deaths per 100k.
The previous figures show no apparent relationship between the amount of information (number of variables) and the social States of Risk represented by the normalized COVID-19 cases and deaths. This is, there is no tendency for the cases and deaths to decrease with an increase on the number of variables from the dashboards.
Bayesian inference, and Bayesian decision-making, suggests that by having information about the components of the Risk equation, the decision making improves. This improvement allows for a better Risk Assessment and Management. This is not what the previous figures suggest.
To explore further the relationship, and the previous affirmation, Fig. 9 presents a similar scatter plot, but this time using the variables needed for Risk Assessment only (Threats, System, Impacts, States of Risk). The color coding on the scatter plots differentiates states that are below the median of the combined number of variables (Threats + Systems + Impacts + States of Risk), between the median and the third quartile, and above the third quartile. Once again, no evident correlation is found in the figures.
Figure 9: Number of Risk Assessment variables compared to cases and deaths per 100k.
As a last exercise, states that include an aggregate index in their dashboards were identified. Table 2 includes the states, the name of the aggregated indexes, and the variables used to compute them. Finally, these states are marked with yellow in Fig. 10.
| State | Index | Variables considered |
|---|---|---|
| Alabama | COVID Risk Indicator | Declining new cases (2 weeks), Percent positive declining, Testing goals met, Visits for covid-like symptoms declining |
| California | County Risk Levels | New COVID cases per day per 100.000, Positivity rate (7 day rolling average), ICU availability |
| Colorado | COVID-19 dial | New cases (7 day incidence level), Percent positivity of COVID tests , Impact on hospitalizations (Stable or declining) |
| Connecticut | Leading and secondary indicators for schools | New covid 19 cases per 100.000 per day, Percent test positivity, New covid hospitalizations per 100.000 per day, Percent COVID-19 like illness hospital visits |
| Idaho | County Transmission Risk Levels | New daily cases 7 day rolling average per 100.000, Testing positivity rate, Regional hospital bed ocupancy |
| Illinois | COVID-19 Resurgence Criteria | Test Positivity 7-Day Rolling Average, Hospital Bed Availability 7-Day Rolling Average, COVID-19 Patients in the Hospital 7-Day Rolling Average |
| Indiana | Advisory level | 7-day positivity rate, Weekly cases per 100.000 residents |
| Louisiana | Community Risk Level | Cumulative 7 day incidence per 100.000, Cumulative 7 day percent positivity |
| North Dakota | County Risk Levels | Active cases per 100.000, Tests per 100.000, Test Positivity Rate |
| Ohio | Opportunity Index | Opportunity domains: Transportation, Education, Employment, Housing, Health, Access, Crime |
| Oregon | County Risk Levels | Confirmed and presumptive cases, Cases per 100.000 residents, Test Positivity rate |
| Virginia | Pandemic metrics - Composite score | Cases, Percent positivity, Outbreaks, HC workers, ED visits, ICU hospitalization, Hospital beds, PPE |
| Washington | Metrics for Risk Assessment | Rate per 100K newly diagnosed cases, Daily molecular testing rate, Percent of positive molecular tests, Percent of adult staffed adult care beds, Beds occupied by COVID patients, Percent of adult ICU staffed beds occupied, Percent of adult ICU staffed beds occupied by COVID-19 cases |
Figure 10: Number of Risk Assessment variables compared to cases and deaths per 100k.
From the figures included in this section, it can be concluded that no evident correlation exists between the amount of information presented and social States of Risk. This holds true for the total number of variables in the dashboards, the number of variables required for a Risk Assessment, and also for those states that presents aggregated indexes, some of which are called Risk Indexes. Potential reasons for these findings are: