COVID-19 Data Analysis

As this time of uncertainty continues, our approach to analyzing the current situation and projections for the future needs to evolve. The first version of our model was designed around two expectations:

1. that the number of cases would significantly trail off, and
2. that consumer behavior and government policies would return to normal as the number of local cases fell.

What time has shown is that consumer behavior and government policies are moving towards their pre-COVID-19 norms much more quickly than the total number of cases is dropping. There are also some states in which the total number of cases is either steady or increasing.  Several states are now projected to reach their peak case load in the coming months, and it is possible that others will eventually see a second peak.

To keep our projections relevant to you, we have stopped using the number of cases as an indicator of consumer behavior.  Instead will be using consumer mobility data, captured from anonymized cell phone location data, to predict current and future in-person activity at your business.

We anticipate that some states may experience a decline in mobility if the total number of cases increases substantially.  Our new approach will allow us to continue providing helpful predictions regardless of whether the situation is improving, static, or worsening.

What we’ve added to the dashboard:

Return to Normal Rating (100%): 71.3% (+0.9/week)

This field shows how close your state is to normal (pre-COVID-19) mobility. For instance, in the example above, the state has regained 71.3% of the mobility they had lost. In parenthesis you’ll see a comparison to the previous week. Last week’s mobility rating was 70.4%. Note you may see a negative number in the parenthesis if mobility is trending downward.

Projected Return to 90%: 7/26/2020

This field projects the date at which the state will have regained 90% of their lost mobility.

During these challenging and uncertain times, the search for the light at the end of the tunnel keeps us all asking the same question: when will life return to normal?

The truth, at least for the time being, is that there’s no easy nor definitive answer.

As a homegrown, independent agency ourselves, we wanted to lend a hand to businesses and consumers alike and to provide context as to when some degree of “business as usual” might come back into our lives. Within the context of advertising we aim to project when retail business might open and when to spend advertising dollars so business can start generating revenue.

IHME COVID-19 Projections - http://www.healthdata.org/

The IHME is projecting hospital resource availability on the national and state level. Since the goal of shelter-in-place is to alleviate the strain on the healthcare system, we consider hospital resource usage to be a critical factor in predictive modeling. Resource data is derived from local governments and hospital networks.

State Health Agencies

The local governments have the most accurate records of virus spread in each community, as they are the ones receiving this data directly from hospitals and labs. With our primary focus on local shelter-in-place lifts, we’re using data from health agencies across all 50 states. The individual, state-specific sources ensure that our local data is truthful and relevant.

Directly from Local Authorities

Changes to shelter-in-place rules, school closures, and travel bans originate from governors and public health officials. Press conferences, public statements, and even social media updates from authenticated profiles will show us the most up-to-date status of a community. The quicker we get new data into our platform, the more significant our data can be.

Three major components of evaluation working together to build our model:

1. Manual Labor – A dedicated team is checking each of our sources by hand each day to ensure all data is current. They’re continuously compiling changes in local regulations and resource data and ensuring that any deviations and oddities are vetted before entering the platform.
2. Technical Analysis – Were ingesting resource usage data into our platform and enriching with various data sets unique to each local government. Algorithms built specifically for this purpose are employed to create the basis of our predictive model.
3. Automated Operations – Behind the scenes processes are taking in new information as it enters the platform and refurbishing our models in real-time. The “always on” nature of our platform ensures that our projections evolve as quickly as they can during this transforming environment.

The end goal of our predictive modeling is to combine:

• When hospital resource strain will reach its peak
• How quickly this peak will decline
• How local regulations will affect the potential shelter-in-place lift

Key data points in hospital resource strain and duration include:

• Resource usage per day per state (reported & projected)
• Number of beds, ventilators etc. needed each day
• When resource need is expected to peak
• How quickly this peak was/will be reached
• How quickly this peak is projected to decline

These numbers create a map of how long the healthcare system in each state needs us to uphold strict social distancing practices. But they don’t tell the full story, as local ordinances play a considerable factor in shelter-in-place mandates.

As an answer we’ve built an in-depth look at social distancing within each state. Multiple algorithms assign value to certain elements such as:

• Start date, end date and duration of shelter-in-place mandates
  • Revisions to shelter-in-place mandates
• School district closures
• Travel bans & curfews
• What each state defines as “essential business”
  • And if/how this definition changes
• What we’re deeming “Regional Alliances”- which states have agreed to a coordinated plan to reopen

Other Modeling Insights

• Smoothing methods are also used since communities are not always consistent with the day and time they update data.
• The hospital resource data is enriched with the local ordinance data to create a comprehensive look at the factors affecting social distancing in each state.
• Our models will evolve, and we will be habitually updating our Ad Planner as we continue to receive novel data.

• Local data reported by each state is accurate
• There will be no unexpected fluctuations in hospital resources
• Social distancing will continue at least through May 31
• No massive advancement in treatment (e.g. vaccine) in the next three months
• If exposure is suspected, 14 days remains the barometer for recommended quarantine even after social distancing regulations are lifted
• Once shelter-in-place is lifted, there will need to be continual:
  • Widespread, accessible testing
  • Ability to quickly and effectively quarantine the infected
  • Continued diligence about mitigating transmission risk

Specific Factors that Could Change Our Projections

• How much local governments’ decisions are affected by neighboring counties, states, etc.
• How exactly local governments will choose to lift shelter-in-place and business closures
• If there will be future waves of the virus in the near future
• How quickly the curve can grow with varying degrees of social distancing lift
• The number of people who have/had the virus, are asymptomatic, and unknowingly transmit
• Timeline & efficacy of vaccine/treatment

The Wild West of Pandemics

This crisis has left the world with new and unprecedented challenges. With this uncharted territory comes the potential for factors beyond our current realm of comprehension, be it psychological factors or even survival instincts, that can dramatically change the outcome of these predictions.

Why might my projected date be so far out if my state passed its peak hospital resource usage?

Our model is considering local government policies including stay at home orders, school closures, as well as regional factors such as statements by governors about coordinated reopening plans.

The governor said businesses can open now, why is the projected opening date so far out?

Our model is trying to predict when customers will be in stores, not when it’s legal for them to be open. We will be taking updated policies like re-openings as an indicator of what’s happening on the ground and use that in the model.