No. As far as I can tell, approximately 100% of people in New York want to be tested, for many different reasons. You're assuming something and projecting it as truth.

"These studies would ideally be true random samples so that we can know what the true infection rate is."

But by your own logic, there can be no "random sample"...we can't tie people down and force them to be tested, so we have to ask them. This means we're back to the self-selection problem that apparently ruins the sample.

The truth is, the New York study did pretty much what we always do to get a random sample: asked a bunch of people at random to participate in the program.

In NY, the issue was not about wanting to be tested, it was about "are you there when they're picking people to test"; quoting from the article, which is very thin on info, people were selected "at 40 locations that included grocery and big-box stores". (In Santa Clara, with the Facebook ads, people "wanting to be tested" by going to a testing center is not anywhere close to 100%.)

The assumption that people who are at grocery and big-box stores are representative, infection-wise, of the population of a whole is not obviously true to me. Many other posters in this thread point out reasons why they might be more likely to be infected.

> But by your own logic, there can be no "random sample"...we can't tie people down and force them to be tested

This approach may go against American sensibilities, but other states (SK, DE, IT) have had better luck with the approach of sampling everyone in a town/region/area. Or at least pick people truly at random, not just a random sample of people who happen to show up in a public place in the middle of a pandemic with a shelter-in-place order in effect. (Also, I'm not sure how "my own logic" makes any claims about the impossibility of a random sample, could you elaborate?)

Even in America, you can at least try to account for the ways in which your sample is likely to be different from a truly random sample. The study by Stanford in Santa Clara county I know at least tried to do this, but of course you can't account for "thinks is infected" as a demographic attribute because that's the dependent variable; I don't know if the NYC study did this type of accounting, but I assume they tried to too.

> The truth is, the New York study did pretty much what we always do to get a random sample: asked a bunch of people at random to participate in the program.

You clearly know about sample bias, so I'm not sure why you think going to a single location (or even 40 single locations) and randomly asking people to be sampled is a good way to get a random sample. Think about going to every Costco and Whole Foods in the state and asking "people at random" for their political views. You're going to get a biased sample, even though you're nominally asking people at random. You can try to account for bias, based on demographic factors you observe in your sample that correlate with political views, right?

But we don't know enough about infection rate to be able to effectively account for infection rate given that people who are showing up at grocery and big-box stores on a random day are more likely to be exposed (they're at a store!) and thus more likely to have been infected than the population as a whole? Literally the only info you have is that they're more likely to be infected -- there aren't well-established demographic correlates with infection.

It's the same critique. If you pick the people anywhere other than from their home, the argument is that they're not at home, therefore they're more likely to have it.

OK, so how do you get them in their homes? If you solicit them on the internet, then there's sample bias because you're telling them they'll be tested. If you knocking door-to-door, it's the same thing: you have to tell them what you're going to do, so you're "selecting for people who want to be tested". You can't win.

"Or at least pick people truly at random,"

You can't force people to take blood tests against their will. You have to tell them what you're doing, and why you're doing it, and they must consent to participate. The same objection always applies to any "random" selection of humans: it's biased towards the people in the place at the time of selection, who agreed to be selected.

"you can at least try to account for the ways in which your sample is likely to be different from a truly random sample."

Right, as all legitimate researchers in this area do.

As I said before: this is all just a highbrow way of rejecting studies for lowbrow reasons. You will never find a survey without some form of sample bias. You control for it and move on.

We are now seeing multiple independent serological surveys with different methods pointing in the same direction. It isn't a methodological error.

Yeah, obviously we agree here. So: how do you effectively control for them in this case? That's my point. It's hard, and you haven't offered any mechanism for this particular case, just assurances that people who know what they're doing do in fact know what they're doing.

Are you one of those people? Please fill us in on what they're doing to address sample bias. They "try to control for it" -- how in this case? So far you've offered nothing specific, just that experts control for it, and you're implying that it's illegitimate to question whether there might be a systemic bias because all the cited studies seem to select populations more likely to be infected.

> OK, so how do you get them in their homes? If you solicit them on the internet, then there's sample bias because you're telling them they'll be tested. If you knocking door-to-door, it's the same thing: you have to tell them what you're going to do, so you're "selecting for people who want to be tested". You can't win.

Am I understanding correctly that you're saying that because you can't get a perfectly random sample you shouldn't try to minimize selection bias? You can't "win", but you can get closer than these studies did. There's a clear difference between a Facebook ad "Stanford seeks people for COVID-19 tests" and "Your number was chosen at random and we'd like to test you for COVID-19 in the interest of science."

One should absolutely try to minimize selection bias, in addition to controlling for its inevitability. In Germany, as you propose, they are selecting people at random from a central database of residents, which is not correlated with whether they have symptoms, feel comfortable shopping in public, etc. That is better than showing up at a grocery store, clearly, right?

As for people who agree to be selected, yes, there's nothing we can do about that in liberal societies. I'd like to know what this rate is. In Germany, I recall that it was low.

> this is all just a highbrow way of rejecting studies for lowbrow reasons

OK, this is the second time you've implied that I'm uninformed, or am acting in bad faith or with bad motives (or whatever "lowbrow reasons" means). None of these are true. I've ignored the personal nature of your vague dismissals up to now in the interest of conversation, but I'm done doing so.

No, you're not. I was explaining why your argument is a truism in disguise.

"OK, this is the second time you've implied that I'm uninformed, or am acting in bad faith or with bad motives (or whatever "lowbrow reasons" means). None of these are true. I've ignored the personal nature of your vague dismissals up to now in the interest of conversation, but I'm done doing so."

I don't know if you're doing it in bad faith or not, but you're definitely making a highbrow lowbrow dismissal. You've set up an argument that can never be refuted, for reasons I've explained.

I'm arguing that the selection mechanisms they're choosing to use are bad compared to what they could and should be using, namely, random sampling as was used in Germany. Yes, those still have problems as you've mentioned, but they are much better overall. I'm sure there are reasons these studies didn't use those mechanisms -- some bad, like that it's hard to get a good random sample, and it's easy to run facebook ads or camp out at a grocery store, and probably some good, like that it's easier enough that it makes results available sooner.

I'm also arguing that because of this bias, and because we are now seeing a few of these studies with similar biases despite different methodologies, and because, of the papers we could read (namely just Stanford's), the accounting for selection bias is weak, it's risky to make statements about trends indicated by these papers.

You could refute my argument by explaining how these not-very-random selection mechanisms could be effectively accounted for post-selection. I've asked you to do this several times, and you have not done so.

I am not making an argument that can never be refuted. I'm asking you to refute it, and I've given you one potential argument, and you have chosen not to do so, repeatedly.

I'm sorry, but in this exchange, the person putting forth arguments that amount to vague dismissals that can't be refuted was, um, not me.