I can understand why a “wait and see” approach sounds reasonable for moving from alert level 4 to level 3. However I think it is based on false reasoning. Let me explain why. #nlpoli 1/n

You’ve probably heard of the reproduction number R. This is the average number of secondary infections caused by each infected person over the course of their illness. 2/n

If R is greater than one, the disease tends to spread. If R is less than one, it tends to die out. So health officials are very interested in measuring R. 3/n

But R is just an average. In reality most infected people will infect no one, some will infect one, some will infect two, and so on.

To estimate R, you need to observe the disease spreading in the community, and analyze that spread statistically. 4/n

The problem is: we have no cases, so there is nothing to measure. Even if we have some new cases over the next two weeks, it will be very difficult to measure R reliably from them. This is due to “clustering” or “overdispersion”. 5/n

The number of secondary infections from an infected individual is highly variable. Epidemiologists believe the standard deviation is about three times R. This means you must study hundreds of cases to get a reasonable estimate of R (by e.g. the central limit theorem). 6/n

We will not have hundreds of cases in the next two weeks. If we have more than a handful, it will probably be due to a sporadic random event which will tell us little about R. 7/n

I do not think we can measure R to a useful degree of precision when there are so few cases. But I could be wrong, I’m not an expert. 8/n

However, I believe the concerns I raise above are credible, and I encourage @Johnrockdoc and @cmoh_nl to consult with the able statisticans that are advising them, and ensure that this issue has been considered and is clearly understood.

I meant to add a link about overdispersion for covid-19: https://www.sciencemag.org/news/2020/05/why-do-some-covid-19-patients-infect-many-others-whereas-most-don-t-spread-virus-all

Also, look at the confidence interval on this R estimate for NL. Waiting 2 weeks will not tell the difference between R=0.8 and R=1.2, when we have almost no cases. https://twitter.com/DFisman/status/1259094798088142848

An addendum to this thread.