Co-author of this article is Peter A. Morrison, Ph.D., RAND Corporation Senior Demographer (retired) and President, Peter A. Morrison and Associates, Inc. He is an applied demographer based on Nantucket Island, MA. He graduated from Dartmouth College and holds a Ph.D. from Brown University. Email: email@example.com; website:
- - -
Now that we are approximately two weeks past the 4th of July weekend, it is time to look at the COVID-19 early warning system we developed for San Juan County and described on June 2nd . Two weeks is an important marker because it is the average time it takes to develop COVID-19 after exposure, and July 4th would normally be a time that a lot of visitors are on the islands.
In our June 2nd report, we used San Juan County as an example of a COVID-19 warning system that could be employed by areas affected by seasonal populations. We developed a model that could derive useful “yellow” and “red” early warnings from the overall statistical trend in daily reported counts of local “confirmed coronavirus cases” that health departments collect. These “early warnings” were designed to help local communities affected by seasonal populations to detect and react to an impending risk, or to confirm the ongoing absence of risk if previously-imposed restrictions are gradually relaxed.
We selected San Juan County as an example because summer visitors increase the effective population by about 52 percent. So many newcomers almost certainly would amplify the risk of an outbreak, especially on Lopez, Orcas, and San Juan islands which are the destinations of most summer visitors.
For the early warning alert system, we adapted a traditional forecasting approach known as the “ARIMA” (Auto Regressive Integrated Moving Average) model. Details on the specific ARIMA model we developed for San Juan County are found in the original June 2nd article (via the link above).
The principal virtue of our model is its minimal appetite for data. A user need only assemble the ordered time-daily total cumulative COVID-19 case count. For this application, we acquired the original data and this update from the Hopkins University Coronavirus Resource Center.
The headline figure at the top of this article displays the actual 68-day time series of the cumulative case count data from March 22nd through May 28th, 2020 (daily red dots). The middle solid line extending to the right (beyond the red dots) traces projected values from May 29th onward through September 5th. This trend line represents a plausible “best estimate” of what lies ahead absent any departure from the prevailing quiescence before a summer influx of outsiders. The dashed lines above and below the solid trend line represent the 95% lower and upper prediction limits.
The early warning system is based on comparing reported data to the trend and watching to see if it starts veering toward that upper 95% prediction limit. An accelerated departure of reported data above the forecasted trend would warrant close observation, perhaps triggering a “yellow light” warning alert for preparatory actions. If reported data hit or exceeded the upper 95% prediction limit, then a “red light” would be triggered, which would activate pre-authorized “tightening” actions such as a ban on rentals, lock-downs, and so on.
Our forecast from the end of May shows that 25 cases were expected, which is only one case higher than the 24 cases reported for July 17th. Importantly, the reported number is well under the upper 95% prediction limit of 32 cases, and the reported cases are not going above, or moving away from, the forecasted trend. In fact, they tend to stay below the forecasted trend: On June 30th the forecast was for 22 cases and only 19 were reported; on July 10th, 23 were forecasted and 20 were reported. So far, so good for San Juan County.
We will continue to monitor San Juan County. If we see the reported cases going above the forecasted trend, we will provide a “yellow light” report as appropriate and keep a close eye on the need for a “red light” report.