Citat:
Ursprungligen postat av
B.R.Donut
Snarare så faller dina beräkningar nu platt på marken. Tyvärr.
Blodgivarna är ett väldigt litet material att dra några vettiga slutsatser ifrån. Bara ca 40 prover per region och vecka i genomsnitt. Därav att man inte redovisar några siffror per region.
Bara det att andelen med antikroppar har "minskat" från v20-22 gör ju att man drar öronen åt sig. Om du inte redan har noterat det så studera konfidensintervallen i diagrammet.
Man säger heller inget om att resultatet (5%) är vägt för hela populationen. Vilket jag är rätt säker på att det inte är.
Jag skulle inte dra några som helst växlar på den undersökningen vad gäller absoluta tal, den enda slutsaten man (knappt) kan göra är att trenden över några veckor tycks vara ökande. Dessvärre är det snarare så att hade man gjort en parallell liknande undersökning så skvallrar konfidensintervallen om att man lika gärna kunde ha fått motsatt resultat...
Blodgivare kan möjligen vara något mer representativa än sökande på vårdcentraler men det räcker inte i det här fallet. Materialet är för litet och osäkerheten för stor.
Not really, they claimed to have 400 tests a week coming in and that definitely provides a large enough dataset to extrapolate for the country at large, which is why they did it. They claim to have taken 400 tests over a period of 5 weeks, which gives us 2,000 samples. This is definitely large enough to give us a dataset within a 95% CI.
http://www.worldvaluessurvey.org/WVSContents.jsp?CMSID=WVS7MET
For example, 1500+ is the minimum data sample size for extrapolating for countries with greater populations like America or Russia. 1,200 to 2,400 samples are generally used when extrapolating for countries. We’d receive a +-% of around 2% when extrapolating a dataset of this size. Anywhere from 3% to 7% if we assume the test is already weighted for false-positive prevalence, which I don’t believe they have weighted for previously (although they have released papers explaining how to weight for this recently) — or 2% to 6% assuming their current-best testing parameters. We’d receive 1% to 5%~ if we assumed they used the test parameters for the other dataset in the report.
Antibody rates falling and rising in between different sets isn’t strange. Especially considering the false-positive prevalence inherent in the tests. We’ve seen this happen multiple times in the other dataset from Vårdcentralen. We’ve seen it happen in other studies. It’s simply not enough to eliminate datasets, which are not definitive, they provide ranges in a population. It wouldn’t surprise me if the next set of results from the Vårdcentralen dataset provide a lower antibody rate again both nationally or regionally. Having such a low prevalence rate coupled with these testing parameters will do this.
But it’s nice that you agree that random-sample blood donors are likely to be more unbiased than primary care samples. I wish they would just release a random-sampled data-set so this debate wasn’t necessary. I still argue that 2,000 random-samples are more than enough to extrapolate results for the country, and that your attempts at throwing out the results over something that will occur within all low-prevalence antibody studies is unnecessary. I still stand behind calculations based on this dataset and posit the IFR to be anywhere from 1% to 1.3%, matching similar results across Europe.
som borde vara utanför