3 ms·
> The researchers enrolled nearly 350,000 people from 600 villages in rural Bangladesh Wow. I was about to complain about "small n", but... an experiment of th
by dragontamer 5y ago
> The researchers enrolled nearly 350,000 people from 600 villages in rural Bangladesh
Wow. I was about to complain about "small n", but... an experiment of this size is not small at all. Congrats on the researchers on doing such a big study!
- cfn 5y agoI think it is a bit cheeky of them. They promoted the usage of masks and distancing in 600 villages totalling 350,000 people and observed the results (mask wearing plus distancing) in mosques (by sampling, not by checking each individual).
- jerf 5y agoI know it's easy to get cynical about "sampling" when you see a stream of studies sampling populations of millions with n = 12 and getting p values just barely considered "significant", but that's not the fault of "sampling". That's just... gathering no data, basically, or if you prefer, going through the motions of gathering data without actually doing it.
- cfn 5y agoI am not arguing with the method, I only think the language of the paper is misleading to the point that the parent commenter aparently thinks they actually analysed hundred of thousands of subjects when they just extrapolated their observations to that number. It may even be right and significant but that is not my point.
- dragontamer 5y agoI've reread the article + paper with your discussion in mind, and I agree with you. It seems misleading, and I failed to understand the study from the headline / abstract. Its not a "bad" study per se, but its definitely entering clickbait. I'm still happy they conducted the study and presented their research.
- kragen 5y agoMaybe the relevant point is that it needed to be this big to get a P-value of less than .05, since they only got .04 even with 350,000 people. This is dramatically less effectiveness than I was expecting when I was pushing for mask mandates last year, and within a few months it was obvious that masks alone weren't having the large effect I had hoped for. Now we have quantitative estimates of how small the effect was (disappointingly small) and when it worked (cloth masks had almost no effect).
- dragontamer 5y agoWhen 70% of people weren't affected by the experiment (13% were wearing masks before the experiment started. 60% of people didn't wear a mask after the experiment), then of course the numbers are going to be small. Seeing a measurable difference when the population only changed from 13% mask to 40%ish mask usage is pretty outstanding IMO.
- kragen 5y agoI don't think that's correct; see https://news.ycombinator.com/item?id=28407881 https://news.ycombinator.com/item?id=28407881 for the argument in more detail.
- dragontamer 5y agoMy argument and your argument are seemingly unrelated. That means that the full matrix of situations are available: I'm correct and you're correct. I'm correct and you're wrong. I'm wrong and you're correct, and finally I'm wrong and you're wrong. ---------- So lets just try again from the top. The study shows that an 8-week mask-awareness campaign in Bangladesh across 300 villages (+300 "control" villages 2+km away where mask-awareness did NOT happen) leads to a 29% increase in mask usage and an 11% decrease in COVID19 cases. Is there anything wrong with this new, revised statement I made? It seems to be the big "A / B" test that everyone's wanted for mask studies.
- kragen 5y ago
- otabdeveloper4 5y agoSize of N isn't the issue, controlling for confounding variables is.