The study design is in question too, as it relates to the representativeness of the sample compared to the county’s populations to be able to generalize it.

Specifically, their sample:

1. Was recruited via Facebook ads.
2. The Facebook ads were targeted.
3. Their sample was self-selected from there.

So we know how targeted Facebook ads, to creepy if not disturbing levels that have caused people to advocate for legislation to prevent them from being able to use our private data.

Therefore, it’s most likely going to target people with an internet history related to COVID-19, specifically those who searched about symptoms and contact risk for transmission, and it was probably even more likely to target people who searched COVID-19 testing.

So those who got the ads are probably more likely to have had symptoms, believed they had a contact that put them at-risk, and wanted to get tested.

And then from there, people had to volunteer to participate and then actually participate, which includes driving to a location and getting a needle prick. As hard as it is to get people to participate in simple survey studies that take a few minutes from their own home even with an offer of rewards (or a chance at rewards), this study involved a lot more effort (plus people typically don’t like having to get pricked by a needle).

In addition, we know from surveys that there often a response bias based on how personal the topic is to them. I’m more likely to respond to a survey about sports and head injuries, because I got knocked out of a football game after getting earholed, than a survey about sports and ankle injuries because I never had one.

Therefore, the same is likely here. If a person didn’t have any symptoms, never had any known contacts with anyone who was positive and/or was displaying symptoms, and had been chilling at home in isolation the last few weeks, then that person is probably not as likely to volunteer as someone who had symptoms, had contact with someone who was positive, wanted to get tested, but wasn’t able to.

So if 20 their 50 positives (out of 3330; 1.5%), were true positives (more like only 5 were true positives though), but those 20 were 2.5 times more likely to be infected than the general population (only 8), than their original crude rate of is 6.25 times higher than the population (0.24%), which would only have been 4.8 times higher than their confirmed case rate.

But in reality, since 45 were likely false positives, and only 5 were true positives, then 2 were likely true positives and even if 1 was a false negative, then 3 out of 3330 would put their crude rate at only 1.8 times the confirmed case rate.