2 thoughts on “Links For Today”

1. “Reinfected” seems like an exaggeration. Test-retest reliability for COVID-19 is fairly low, because there’s a lot of noise. The Lancet cites estimates of false negative rates between 2% and 33% in repeat sample testing, and false positive rates of 0.8%-4%, so it’s easy for people to fall under these cases:
   (1) Get a false positive, then a true negative, then another false positive (likelihood
   (2) Get a false positive, then a true negative, then get infected later and get a true positive.
   (3) Get infected and test positive, get a false negative, then get a true negative later.
   (4) Get infected and test positive, recover and get a true negative, then get a false positive later.

   Then there’s the scenario implied in the article:
   (5) Get infected and test positive, recover and get a true negative, then get reinfected

   A good estimate would simulate the likelihood of different observation patterns using probability distributions for the false positive and false negative rates, and some prior for infection and reinfection rates, to evaluate how strong an odds ratio this study gives us in favor of reinfection. But here’s a simplified back-of-the-envelope version.

   The linked article reports:

   All the Marines were beginning basic training and were initially held in Navy quarantine for two weeks, after two weeks of at-home quarantine, according to the study. Once training began, recruits were tested for COVID-19 every two weeks over a six-week period.

   The result: 19 of the 189 recruits who already had COVID tested positive for a second infection during the study.

   Assume everyone’s tested clear as of enrollment, and that that’s a true positive.

   Looks like recruits were tested three more times. At 0.8% false positives, that’s a 99.2%^3=97.6% probability per recruit of consistently testing negative, which naively predicts that we should expect 2.4% of recruits to test positive, or around 4.5. At 4% false positives, that’s a 96%^3=88.5% probability per recruit of consistently testing negative, so we should expect 11.5% of recruits to test positive, about 22. The observed number of 19 is nicely bracketed by those predictions.

   By contrast, the article reports:

   Of 2,247 recruits who had not previously had COVID, 1,079 (48%) became infected during the study.

   That’s well outside the range we should expect from false positives alone.

   Not only doesn’t this establish actual cases of reinfection, it barely provides any Bayesian evidence for reinfection, if we believe those false positive estimates!

2. Thanks for the comment. For the record, nothing linked here should be considered as an endorsement of the content. In fact some things are posted that are strongly against the managements beliefs and conceptions of what is true. The goal is to provided links of what people are thinking, saying, and reporting with the goal to inform and not to instruct.

   That said, feel free to criticize the contents of every and any link posted. Sometimes we can’t help but be sarcastic about we post ourselves.