4 ms·

That is just nonsense. You don’t adjust for demographic bias in a sample by conditioning on a second parameter. There’s no universe in which underweighting the

• cameldrv 7y ago
  Not in this case. It's basically Simpson's paradox. Honestly I think the study is basically useless, because instead of a random sample, it's a sample of people who, when they saw an ad on Facebook for a free Coronavirus test, decided to sign up, and then actually show up for the test. The point is that the higher your (voluntary) participation rate is within a group, the less chance you have of selection bias. This goes up to the limit of 100% participation, where you have no selection bias. What we see in this study is that groups with a lower participation rate (Men, nonwhites, people farther from a testing site) have a higher positivity rate. That is exactly the selection bias we would expect. They then proceed to "correct" the data by undercounting the group (white women that live near a testing site) that has the least selection bias.
  • SubiculumCode 7y ago
    3,300 people signed up for this study and completed this within weeks. Have you recruited for a study before? That kind of participation is hard to achieve...practically unheard of, unless you have widespread interest and time: Something which a whole lot of people have right now given layoffs, fear, and the bit of monetary compensation for participation, I bet sampling was much wider and less biased than you think.
    • cameldrv 7y ago
      Yeah but most studies aren't testing for a virus that has everyone locked in their houses causing a global pandemic.
      • SubiculumCode 7y ago
        But that is the point. Selection bias towards those that are actually infected is minimized when a large portion of the population have the time and desire to get tested.
• JamesBarney 7y ago
  No his argument is that there are two sampling biases. The demographics of a tested population do not match the demographics of the county. If someone believes they've been infected they are more likely to seek testing. (self-selection bias) The study adjusted for the first but not the second bias. He believes the second bias is more important, so when they adjusted for the first they moved the adjusted average farther away from the true average. I think he's right, self-selection can be a far more impactful bias than things like sex and race.
• mlyle 7y ago
  I do think it's likely the study does overreport the infection rate. People who are really curious about their past COVID-19 status are not representative of the population as a whole. And stratified sampling can help to try and correct this, but... we all know that you can't really "save" or "fully condition" a convenience sample. On the other hand, it's hard to reconcile their finding with anything less than 90% or more of infections being missing from the case count. And 10% of the infection fatality rate is really exciting news. It also means that it's exceptionally likely that jurisdictions like New York are a huge slice of the way (15% of population infected?) to herd immunity.
  • cameldrv 7y ago
    I agree that cases are probably giving maybe a 10-20x undercount. That means that NYC has maybe 3 million infections. That means NYC is maybe half way to herd immunity, assuming reinfection doesn't happen and that immunity is lasting. The only problem is that to get the other half way there, another 20,000 people have to die, just in New York City. That doesn't even count the already existing infections that haven't died yet.
    • mlyle 7y ago
      Maybe this time we can do a better job of concentrating the infections in the less-likely-to-die group, and cut down on the deaths a little that way.