The founders of Instagram have launched rt.live, a new site estimating Rt, the effective reproduction rate of the coronavirus, for each US state. Briefly, when Rt is above 1.0, the infection spreads at a rate faster than the rate at which already-infected people acquire immunity, so the rate of new infection accelerates. If Rt dips below 1.0, previously infected people are recovering faster than new cases crop up, so contagion decelerates. The new website shows its work — rationale, empirical evidence, algorithm — but it’s a tough technical slog. The rt.live algorithm uses time lags and trend smoothing and the Poisson distribution, but the gist is still this: if the Rt >1, then the daily count of new cases will go up; if Rt<1, then the daily case counts will go down.
As of today’s update, the website indicates that 30 of the 50 states have an Rt below 1. Factor in a time lag for testing, and it might be reasonable to infer that the Rt<1 states will start showing actual drops in daily diagnoses in a week. Helpfully, the website also shows Rt scores from a week ago. Back then there were 37 states under the Rt threshold, suggesting that the model might be overly optimistic. Looking two weeks back, only 15 states had sub-1 Rt scores, of which 10 still meet the criterion. Looking at those 10 states, how many of them show actual decreases in daily diagnostic tallies during the three weeks since April 1? All ten of them.
Just because a state’s new diagnoses are dropping doesn’t mean that it’s nearly out of the woods. Idaho, Maine, Oklahoma, Oregon, Tennessee — these five states with Rt scores consistently below 1 and decreasing average daily case counts since April 1 never did have many cases to contend with. But the other five that make the cut for reduced contagion — Michigan, New York, Florida, Louisiana, New Jersey — have been hot spots.
It’s become increasingly apparent that the official daily counts vastly underestimate, maybe by as much as fifty-fold, the actual incidence of covid infections in this country. Adding further confusion, the states differ wildly from each other in the number of tests per 100,000 population, so there’s an apples-to-oranges problem. Still, the Rt scores and daily case counts measure trends within each state rather than comparisons between states, eliminating some of the noise.
Death counts aren’t as contingent on measurement bias, so they would seem to offer a more tangible basis for evaluating state-specific trends. Looking at those same ten states again, has the three-week decline in newly diagnosed cases corresponded with a downward trend in daily deaths? Not consistently. Four of the ten experienced decreasing deaths in each of the last 3 weeks, while for four other states the weekly death counts went up. Perhaps that’s not too surprising: death is a trailing indicator, lagging behind diagnosis by a week or more. Maybe most notably among the ten states under scrutiny, the four suffering the highest overall death totals — New York, New Jersey, Michigan, Louisiana — experienced decreases over the past week.
Some basis for optimism here — will keep tabs on these trends going forward.