Shining a Flashlight on COVID Risk at Halloween
by Jason S. Brinkley, PhD, MS, MA
As the father of four, I’ve spent much of the pandemic glued to television and internet resources watching the reporting on the spread of the COVID pandemic. With the upcoming mix of holiday and flu seasons, it seems that most expert sources are reporting a coming surge of COVID cases. Questions about participating in social activities loom. A recent report by the National Retail Federation suggests that 58% of Americans still plan on celebrating Halloween, and 44% of parents plan on trick-or-treating. The risk of COVID exposure while trick-or-treating in a global pandemic is something that we have no actual data to rely on, so any measure of potential risk would have to rely on other data and a bit of forecasting. But how do scientists and data experts make those forecasts, and exactly what can we say about risk for something that has never happened?
A lot of prediction, forecasting, and statistical modeling rely on synthesizing different characteristics down into a single metric. We do that using mathematical projections. Think about a mobile or three dimensional diorama of the solar system set up in a child’s room. If you can walk around the room, then you can definitely see where the planets align along with their size and relative position to the sun. But if you can’t walk around the room and see the entire picture, you could also shine a flashlight on the model and look at the shadows along the wall. Depending on where I shine the flashlight, the shadows on the wall appear to be different. No viewpoint is a perfect representation of the 3D image, but if we shine the flashlight a certain way, then we might get a really good idea of what is going on. This is what we do with data when we have many characteristics and we can’t actually visualize the full picture of what’s happening. In essence, we shine a statistical flashlight on the data and look at projections on a wall. In the same way that we know not to put the flashlight too close to a particular planet (as it will make the shadow look bigger than those of larger planets), we have mathematical tricks to know good places to hold a flashlight.
So what does all this mean for Halloween and COVID risk? If we take a number of indicators that are related to both COVID risk and the potential for trick-or-treating, then a projection might give us some measure of how they all associate, and we can use that as our forecast. So I linked data between the John’s Hopkins COVID Tracker to County Health Rankings data from the Robert Wood Johnson Foundation. Specifically, I pulled data in three groups across all US counties:
- COVID Risk – Incidence Rate and Case Fatality Ratios
- Health Risk – Percent Poor/Fair Physical Health, Average Poor Mental Health Days, Flu Vaccination Rate, Diabetes Prevalence, Adult Obesity Rate
- Trick-or-Treating Specific – Food Environment Score, Ratio of Population to Dentists, Children in Poverty, Child Mortality, Percent Below 18, Percent Rural
The first question is whether these are all (or even the best) indicators for Halloween COVID risk? Some of them speak more toward regional poverty or regional health and less toward actual exposure probabilities. That is a fair question and such dialogue is common among scientists. Also, the specific projection techniques I use here don’t account for the fact that counties are “close” to one another and that COVID risk changes (dramatically) over time. See my note on methodology. Scientists routinely critique the models that they use for making estimates, and it as a normal and healthy thing to do. Just as meteorologists show different hurricane forecasts and talk about various models, so too should we keep putting other scientific models under the microscope.
Putting all this together in a projection of potential risk gives us a score for almost every county in the country (with the exception of much of Utah where limited data was available for analysis). Higher scores would suggest higher “risk.” Overall, this projection finds most of the country having about the same mid-range scores with some “hot spots” in the Dakotas, Arizona, Texas, and other parts of the US south (see picture below). You can see how your county scored by going here. There are some places where this model likely gets it very wrong (move the online version of the map to check out Alaska).
At this point, I would consider this analysis exploratory and I would say that there are well studied alternatives. Here is a great one that is COVID-specific and doesn’t consider regional factors that may be more Halloween- or trick-or-treating-specific. If you find that your area has high risk, then there are a multitude of steps you can take if you still want to celebrate. For me, one is in choice of Halloween costume, and I have decided to go with something with full coverage. Follow me on Twitter (@DrJasonBrinkley) to see some of my Halloween night adventures.
Of course, you may be looking for alternative ways to celebrate. Ice cream from McDonalds is a popular option when you can get it; and as further proof that there is a map or dashboard for everything, you can find out availability at www.mcbroken.com.
*Note on methodology: Data from both sources were linked at the county FIPS level. FIPS that had limited data or small geographic spaces (US Territories, tribal lands, etc.) were removed. Less than 2% of data were missing on non-COVID indicators and were imputed using EM algorithm techniques. The combined data were analyzed via principal components analysis with the first principal component serving as “COVID Halloween Score.” Overall, this component represented 40% of the variance in the overall data and had a Pearson correlation of 0.35 with the main outcome of county COVID Incidence Rate. The score itself had a slightly skewed bell-shaped distribution. Visual inspection of COVID Halloween Score and log transformed Incidence Rate showed a clear monotone curve-linear relationship. All analyses were performed in SAS® software with data visualization in Tableau® public.
Jason S. Brinkley, PhD, MS, MA is a Senior Researcher and Biostatistician at Abt Associates Inc. where he works on a wide variety of data for health services, policy, and disparities research. He maintains a research affiliation with the North Carolina Agromedicine Institute and serves on the executive committee for the NC Chapter of the American Statistical Association and the Southeast SAS Users Group. Follow him on Twitter. [Full Bio]
Previous posts in this series:
- The COVID Denominator
- Halloween by the Numbers 2019
- We R-Naught Ready for an Epidemic
- How Do Machines Learn? Part 3: They Recover
- How Do Machines Learn? Part 2: They Fight
- How Do Machines Learn? Part 1: They Learn
- Taste Testing Generic Drugs
- Halloween by the Numbers
- What Kills Kids?
- The Golden Age of Health Research Funding
- Does Living on a Prayer Work?
- The Opioid Data Crisis
- Income Lost from Snow Days*
- What the #$@&*! Is Blockchain?
- Opportunistic Research Opportunities
- Text Mining UFO Data: Little Green Aliens or Santa’s Elves?
- Should You Know Your Doctor’s Home Address?
- The Population Bullet
- The Unknown Unknowns of Missing Data
- Communicating Science–More Than Just Good Words?
- Counting Alabamas
- The Third World in Your Own Backyard
- The Unrealistic Gold Standard
- Does MACRA Signal the Beginning of the End for Medicare Claims Data?
- Think You Aren’t Extraordinary? Odds Are You’re Wrong
- Mapping by Words
- Are We Asking Too Much From Surveys?
- Making Better Comparisons
- What Kills Us?