Need Help With Upcoming Episode on Mask Junk Science

[Eric Newhill]

OK,
"Among the 335,382 participants who completed symptom sur- veys, 27,166 (8.1%) reported experiencing COVID-like illnesses during the study period. More participants in the control villages reported incident COVID-like illnesses (n=13,893, 8.6%) com- pared with participants in the intervention villages (n=13,273, 7.6%). Over one-third (40.3%) of symptomatic participants agreed to blood collection. Omitting symptomatic participants who did not consent to blood collection, symptomatic seroprevalence was 0.76% in control villages and 0.68% in the intervention villages. Because these numbers omit non-consenters, it is likely that the true rates of symptomatic seroprevalence are substantially higher (perhaps by 2.5 times, if non-consenters have similar seroprevalence to consenters)."

So they questioned the villagers - both intervention villages and control villages - about their symptoms, if any.

13,893 control villagers reported symptoms (8.6%) and 13,273 intervention villagers reported symptoms (7.6%). Total of 27,166 with symptoms in both mask and no mask groups.

Of those reporting symptoms (27,166) 40.3% agreed to blood tests - that's 40.3 % of 27,166 = 10,948 agreed to blood tests.

Here is where I run into a barrier. .76% in control villages and .68% of experimental villages who had their blood tested showed signs of covid infection. That's where Alex's figures are coming from. It's right there.

So now we have 10,948 blood tests, but it doesn't give us the break out. Of the 10,948, how many were control group (non intervention villages) and how many experimental group (intervention villages)?

If you do not have the breakout - the denominator - you can't do the chi-squared or other calculations for statistical significance. .76% of what portion of the 10,948? .68% of what portion of 10,948? What are they hiding?

They should be telling us that of the 10,948 blood tested people - making up numbers for illustration - 6,500 were control villagers wearing and 4,448 intervention villagers. Then give us the numbers of serum positives for each group (or we could just derive the number of serum positive using the .76% and .68% if we had the denominator, which we don't). Then we can run some basic statistics to test for significance.

The denominator is key to performing the test. The math doesn't work without it. Is it .76% of 7,000 or of 8 or of 91? Who the hell knows? And why don't we know? Sample size is critical to calculating margin of error, variance, etc. in the statistics test.

Though that statistical/design approach has issues - like the blood tested were not randomly sampled (they are the subset/people who consented).

It also does not take into account asymptomatic covid (you only got selected to be blood tested if you self-reported symptoms). There are other issues.

The researchers attempt to hide their problems by then extrapolating back from those %s (.68 and .76) in the blood tested group to the entire population using some murky regression analysis. Then they claim statistical significance. This would never pass peer review, IMO. It's pure crap.

There is no test of statistical significance that can be applied to the blood tested group (Alex's favorite figures) because they do not supply a key figure (see above). Which is something that I have been saying in a round about way since the beginning of my entry into this discussion. Ellis be damned.

And that's the name of that tune.

 

[EllisR]

Just for the edification of others who might be following the thread, you said this yesterday,
"Using this measure, masks could prevent 99% of COVID cases, and your measure would still show up as statistically insignificant. Try it for yourself and see. Make up some huge number of people who report respiratory symptoms, in the no-mask group - let's say 10,000. Applying our seropositivity rate of 22%, that gives us 2200 in the no-mask/proven to have COVID group and 7800 in the no-mask/proven to not have COVID group. Now let's say only 100 people reported respiratory symptoms in the mask group. Applying our seropositivity rate of 22% gives us 22 in the mask/proven to have COVID group and 78 in the mask/proven not to have COVID group. Run your Chi-square test and see what you get...p=1. So masks prevented 2178/2200 cases, but your "outcome" shows that "masks don't make a difference"? What if it was the other way around. What if masks caused COVID and there were 2178 more cases in the mask group. But the researchers went ahead and claimed that masks weren't harmful, because p=1. Would you really be prepared to buy that?"

Right. I said that your "masked/not masked and blood tested positve/negative" was "terrible methodology" (the part you quoted was one of the reasons) yesterday. And it's still terrible today. So where's the contradiction? And none of this has anything to do with the regression I referred to above, so why did you bring it up???

"assuming that they have the same COVID rate as the consenters" - big assumption.

Agreed. I have no idea why Alex brought it up. I was pointing out why it was correct to say, hypothetically "the symptomatic seroprevalence rates would be higher", not just the cases. I worked out the hypothetical for Alex, to show that the rates would change. But I wouldn't do it for real, and neither did the researchers.

I see now that this is the big flaw in the study. I wasn't getting it because it isn't something that can be done that way. There are all kinds of confidence intervals around that kind of extrapolation, again, especially when starting from such small numbers.

Dude, how can this be the big flaw in the study, when nobody did this in the study???

"New population numbers"? I don't like my results so let's create a new population and see if we get something I like.

Seriously?

 

[Eric Newhill]

Me:"assuming that they have the same COVID rate as the consenters" - big assumption.
You: Agreed. I have no idea why Alex brought it up

He brought it up because he is correct. See my comment #141 above. You yourself are talking about extrapolations being key.

I'm still seeing you as a troll- more than ever now - and you're breaking out all of the troll techniques.

I have stated exactly what they did (comment #141). You can troll away, but now that I have calmly read the damn study, again, I know what I'm talking about and I am once and for all done with your trolling ass.

 

[EllisR]

OK,
"Among the 335,382 participants who completed symptom sur- veys, 27,166 (8.1%) reported experiencing COVID-like illnesses during the study period. More participants in the control villages reported incident COVID-like illnesses (n=13,893, 8.6%) com- pared with participants in the intervention villages (n=13,273, 7.6%). Over one-third (40.3%) of symptomatic participants agreed to blood collection. Omitting symptomatic participants who did not consent to blood collection, symptomatic seroprevalence was 0.76% in control villages and 0.68% in the intervention villages. Because these numbers omit non-consenters, it is likely that the true rates of symptomatic seroprevalence are substantially higher (perhaps by 2.5 times, if non-consenters have similar seroprevalence to consenters)."

So they questioned the villagers - both mask wearing and non-mask-wearing - about their symptoms, if any.

13,893 non-mask wearers reported symptoms (8.6%) and 13,273 mask wearers reported symptoms (7.6%). Total of 27,166 with symptoms in both mask and no mask groups.

Of those reporting symptoms (27,166) 40.3% agreed to blood tests - that's 40.3 % of 27,166 = 10,948 agreed to blood tests.

Here is where I run into a barrier. .76% in control villages and .68% of experimental villages who had their blood tested showed signs of covid infection. That's where Alex's figures are coming from. It's right there.

So now we have 10,948 blood tests, but it doesn't give us the break out. Of the 10,948, how many were control group (not mask wearers) and how many experimental (mask wearers)?

If you do not have the breakout - the denominator - you can't do the chi-squared or other calculations for statistical significance. .76% of what portion of the 10,948? .68% of what portion of 10,948? What are they hiding?

They should be telling us that of the 10,948 blood tested people - making up numbers for illustration - 6,500 were not mask wearing and 4,448 wore masks. Then give us the numbers of serum positives for each group. Then we can run some basic statistics to test for significance.

Though that approach has issues - like the blood tested were not randomly sampled (they are the subset of people who consented).
It also does not take into account asymptomatic covid (you only got selected to be blood tested if you self-reported symptoms). There are other issues.

The researchers attempt to hide their problems by then extrapolating back from those %s (.68 and .76) in the blood tested group to the entire population using some murky regression analysis. Then they claim statistical significance. This would never pass peer review, IMO. It's pure crap.

There is no test of statistical significance that can be applied to the blood tested group (Alex's favorite figures) because they do not supply a key figure (see above).

And that's the name of that tune.

Dude, the numbers you are looking for are in Table A1. I told you that several days ago, and I gave you and Alex the numbers, too.

Also, there are no "mask wearers" and "non-mask wearers". That's not one of the variables.

 

[EllisR]

Me:"assuming that they have the same COVID rate as the consenters" - big assumption.
You: Agreed. I have no idea why Alex brought it up

He brought it up because he is correct.

Correct about what?

You yourself are talking about extrapolations being key.

No I didn't. I've never even mentioned extrapolation prior to this post.

 

[Eric Newhill]

Yeah, they mention that the true rates are likely higher, but they don't try to extrapolate or anything. They stick with using the measured rates. What's wrong with that?



There are no individual subjects who are "mask wearers" and "non-mask wearers". You can't "question the villagers - both mask wearing and non-mask-wearing" because there are only "villagers". You may as well say "question the villagers - both lipstick wearing and non-lipstick wearing", or "question the villagers - both with red shirts and without red shirts". I point this out because it's confusing. It makes it seem like that was one of the variables they measured on the villagers - "age", "sex", "mask-wearing"... It confused you mightily earlier in this thread, and it took quite a while to straighten that out.

If it makes you happy, I changed my language in #141.


Maybe you're getting hung up on the use of the "extrapolated". You seem to get trapped by minutia very easily. Maybe you prefer to use the term "regressed". Whatever. It is the same concept (I am concept guy. That's why I manage people who are obsessed with details). Same thing. They went from blood tests results to making statements about the entire population. That is BS. If you don't think so, then you best step up and finally explain why you think it is legit instead of spit balling from the peanut gallery.

But don't try to gaslight me. Alex's figures come from exactly where I said they did. From the study, "Symptomatic seroprevalence was 0.76% in control villages and 0.68% in the intervention villages" Those are the % Alex is interested in. Once again, there were 10,948 serum tests. You explain how they got to .76% and .68% of the entire population of control and intervention villages. Any normal person calls that "extrapolating". Extrapolation can be performed by various methodologies, but it's all under the heading of extrapolation.

 

[EllisR]

By the way, Eric, I'm just letting you know that I saw your post berating me for using Chi-square, and (once again) claiming that ANOVA was the way to analyze the main result, before you managed to delete it.

 

[Robbedigital]

I can settle this.

I will so up a diaper out of N-95 material and wear it as underwear.
Next, we go to breakfast at a construction site Mexican food truck.
Then we take a road trip non-stop to 14 hours to Denver Colorado, windows up.
If you complain about fart smell, then Jeffrey Epstein didn't hang himself.

 

[Eric Newhill]

The numbers in that graphic come from "a regression of symptomatic seroprevalence on a treatment indicator, clustering at the village level and controlling for fixed effects for each pair of control-treatment villages." You would have to run backwards through the regression to get to the raw numbers, which isn't possible for us.

Yep. That's known as extrapolating. The technique for extrapolating from the small set of people that consented to blood test may be this regression you refer to, but it is extrapolation all the same.

I agree. Which is why you have to start with "people who reported symptoms of COVID-19" and "consented to blood collection" (the numbers I gave earlier).

So you agree with me because that is exactly what I said they did (see comment 141).

So it boils down to whether or not their extrapolation was legit. I say "no".

Like I said, we're all on the same page now, per your own words. This is your big moment. Explain why that extrapolation ....er ah regression....is a legitimate thing to do. What do you see as the downside (everything has a downside).

 

[EllisR]

If it makes you happy, I changed my language in #141.

And yes they extrapolate the %s from the serum tested to the population (146,783 for the control villages and it's the only way to go from the 10, 948 that were blood serum tested.

Maybe you're getting hung up on the use of the "extrapolated". You seem to get trapped by minutia very easily. Maybe you prefer to use the term "regressed". Whatever. It is the same concept (I am concept guy. That's why I manage people who are obsessed with details). Same thing. They went from blood tests results to making statements about the entire population. That is BS. If you don't think so, then you best step up and finally explain why you think it is legit instead of spit balling from the peanut gallery.

But don't try to gaslight me. Alex's figures come from exactly where I said they did. From the study, "Symptomatic seroprevalence was 0.76% in control villages and 0.68% in the intervention villages" Those are the % Alex is interested in. Once again, there were 10,948 serum tests. You explain how they got to .76% and 6.8% of the entire population of control and intervention villages. Any normal person calls that "extrapolating". Extrapolation can be performed by various methodologies, but it's all under the heading of extrapolation.

That is not what is meant by "extrapolation". "Extrapolation" means you infer an unknown value by using trends from existing data. But there were no unknown values used in this result. Everybody in the population was asked whether they had symptoms, so the number of people who didn't have symptoms of COVID was a direct measure of the population. If it was extrapolation, then they would have asked a sample of people if they had COVID symptoms, and then applied that trend to the whole population ("regression" would also work). See the difference? Then they took the all the symptomatic people who consented to have blood tests and directly measured whether they also had COVID seropositivity or not. If they had extrapolated, they would have measured some of the tests and then applied the results to all of the people who consented. Or they would have applied the 22% positivity to the symptomatic people who did not consent to the blood tests.

The entire population in the denominator has every relevant variable measured directly - there were no unknowns used. Now I know you got hung up on the blood tests, as though it mattered whether asymptomatic people also got blood tests. But if you were already asymptomatic, then you couldn't be "symptomatic seropositive". A blood test would have been irrelevant.

They got to 0.76% symptomatic seroprevalence by directly identifying all the people who were asymptomatic, which left them with 8.6%. Then they directly identified who would consent to the blood tests which left them with 3.6%. Then they directly identified who was seropositive, which left them with 0.76%. None of that took extrapolation or regression. There was no "taking the results of the blood tests and extrapolating the results to the whole population". There was only "measuring directly whether each subject had the outcome or not".

 

[EllisR]

I can settle this.

I will so up a diaper out of N-95 material and wear it as underwear.
Next, we go to breakfast at a construction site Mexican food truck.
Then we take a road trip non-stop to 14 hours to Denver Colorado, windows up.
If you complain about fart smell, then Jeffrey Epstein didn't hang himself.

I have to ask...

Peeing in the hot tub, farting in the car, wearing underwear that doesn't breathe...do you have a girlfriend?

 

[Eric Newhill]

Ellis,
Once again I owe you an apology. You are correct that numbers I'm looking for are in the table. It's table A1. I was looking at something else (table 1) which you pointed to. That was an incident of you being imprecise in your language.

Anyhow. The numbers for the serum test (the denominator) are treatment villagers = 5,006 and 4,971 are control villagers.
Positive serum test breaks down to .76 * 5,006 for treatment villagers and .68 * 4,971 for control villagers.

That means out of 5,006 treatment villagers 3,804 were covid positive and out of 4,971 control villagers 3,380 were covid positive.

Applying a chi square test for significance chi square = 79.1202. and that means the results are significant at p-value is < 0.00001

I can also now Run ANOVA just for kicks, but need to grab dinner first. At any rate, based on those figures, the blood serum test had significant results. All good so far.

That is not what is meant by "extrapolation". Everybody in the population was asked whether they had symptoms, so the number of people who didn't have symptoms of COVID was a direct measure of the population. If it was extrapolation, then they would have asked a sample of people if they had COVID symptoms, and then applied that to the whole population ("regression" would also work). See the difference? Then they took the all the symptomatic people who consented to have blood tests and directly measured whether they also had COVID seropositivity or not. If they had extrapolated, they would have measured some of the tests and then applied the results to all of the people who consented. Or they would have applied the 22% positivity to the symptomatic people who did not consent to the blood tests.

I disagree. They are, in my opinion, extrapolating (or whatever you prefer to call it) from a subset of both the intervention villages and the control villages. That subset is those who a) reported symptoms and b) agreed to a blood test and c) followed through and got tested. While the results from the serum tested are significant, I truly do not see how you can move from there to making a statement about the entire population. How is that subset going to be representative of the entire population? How does the methodology account for asymptomatic infections. There are lots of confounds.

entire population in the denominator has every relevant variable measured directly. Now I know you got hung up on the blood tests, as though it mattered whether asymptomatic people also got blood tests. But if you were already asymptomatic, then you couldn't be "symptomatic seropositive". A blood test would have been irrelevant.

Obviously. That's circular logic. The definition begs the result. However, we know that some people, many actually, are infected and have no symptoms. This is a study to test mask effectiveness. Getting covid demonstrates a lack of the masks ability to filter out the virus. Symptoms don't matter. That is why I suggested, way back, that the right way to do this would be randomly sample villagers from the treatment villages and from the control villages. Give them a quick covid test (I note that the desire to sample blood indicates that maybe they know the nasal swabs are inaccurate).

got to 0.76% symptomatic seroprevalence by directly identifying all the people who were asymptomatic, which left them with 8.6%. Then they directly identified who would consent to the blood tests which left them with 3.6%. Then they directly identified who was seropositive, which left them with 0.76%. None of that took extrapolation or regression. There was no "taking the results of the blood tests and extrapolating the results to the whole population". There was only "measuring directly whether each subject had the outcome or not".

Not following you here. That is not how I read the study. Look at the quote I keep pasting today. Left them with 8.6% what? How did they test asymptomatic people? Maybe I missed that. Please, instead of saying "read the study again" or making some snarky remark, try to be helpful and show me a quote, as I have done for you.

Reconcile your statement about identifying asymptomatic villagers with what the report says here;
"Among the 335,382 participants who completed symptom sur- veys, 27,166 (8.1%) reported experiencing COVID-like illnesses during the study period. More participants in the control villages reported incident COVID-like illnesses (n=13,893, 8.6%) com- pared with participants in the intervention villages (n=13,273, 7.6%). Over one-third (40.3%) of symptomatic participants agreed to blood collection. Omitting symptomatic participants who did not consent to blood collection, symptomatic seroprevalence was 0.76% in control villages and 0.68% in the intervention villages. Because these numbers omit non-consenters, it is likely that the true rates of symptomatic seroprevalence are substantially higher (perhaps by 2.5 times, if non-consenters have similar seroprevalence to consenters)."

Are you now telling me that the study ends with what I wrote in my first paragraph? What about Figure 1. It indicates that they have taken the results on my first paragraph and applied it to 146K people in control ("comparison") villages and 160K in intervention villages. Are you saying now that is not true? It sure looks to me like that's what they are saying and that would involve what I call extrapolation. Table A9 also suggest extrapolation.

If they aren't extrapolating and they're just saying that those people didn't report covid symptoms so they don't have it, no need to test, then we have to recalculate everything.

A chi squared would be based on control villagers 146,783 - 3,380 serum positive and intervention villagers 160,323 - 3,804 serum positive.

That results in the chi-square statistic = 1.6431. The p-value is .199897. Not significant at p < .05.

 

[Eric Newhill]

So in conclusion can we say that out of the control villages (n = 146,783) there were 3,380 serum positive cases (distilled down from reported symptoms, agree to test, tested. tested positive) and out of the intervention villages (n= 160,323) same distillation leaving 3,804 serum positive cases.

If the villager didn't report symptoms s/he was deemed to not have covid.

Then a chi squared would be based on this:

----------- positive test ---- negative symptoms
treatment 3,380 --------- 156,519
control. 3,804------------ 143,403

........and the results are not significant. The study flops and the media reports are wrong.

 

[Robbedigital]

I have to ask...

Peeing in the hot tub, farting in the car, wearing underwear that doesn't breathe...do you have a girlfriend?

Yes I do. And even If I had N95 underwear, I would still go to the bathroom degas, because duh.

 

[EllisR]

Yes I do. And even If I had N95 underwear, I would still go to the bathroom degas, because duh.

And she doesn't mind? That one's a keeper!

Does she happen to have a sister?

Just asking for a friend...

 

[EllisR]

Ellis,
Once again I owe you an apology. You are correct that numbers I'm looking for are in the table. It's table A1. I was looking at something else (table 1) which you pointed to. That was an incident of you being imprecise in your language.

Nope.

Dude, the numbers you are looking for are in Table A1.

Anyhow. The numbers for the serum test (the denominator) are treatment villagers = 5,006 and 4,971 are control villagers.
Positive serum test breaks down to .76 * 5,006 for treatment villagers and .68 * 4,971 for control villagers.

Where did you get 76% and 68% from?

I disagree. They are, in my opinion, extrapolating (or whatever you prefer to call it) from a subset of both the intervention villages and the control villages. That subset is those who a) reported symptoms and b) agreed to a blood test and c) followed through and got tested. While the results from the serum tested are significant, I truly do not see how you can move from there to making a statement about the entire population. How is that subset going to be representative of the entire population?

They can't be extrapolating. Your claim is that they are taking information from a bunch of people who got sick. The only information we have from that subset is "what proportion of them were sick with COVID and what proportion of them were sick with something else". Then they applied that information to people who didn't get sick? In order for that to be extrapolation you would have to be saying this:

"308,216 people did not get sick. Of those people, 22% did not "get sick from COVID" and 78% did not "get sick from something else."

A) That would be a silly, stupid thing to say - what would that even mean?
B) The researchers didn't say anything like that for their primary outcome, nor did they say anything like that anywhere in the study.

How does the methodology account for asymptomatic infections.

It doesn't, because the pre-specified primary outcome was "symptomatic seroprevalence". When you pre-specify an outcome, you measure the pre-specified outcome. You don't measure something else instead.

Obviously. That's circular logic. The definition begs the result. However, we know that some people, many actually, are infected and have no symptoms. This is a study to test mask effectiveness. Getting covid demonstrates a lack of the masks ability to filter out the virus. Symptoms don't matter. That is why I suggested, way back, that the right way to do this would be randomly sample villagers from the treatment villages and from the control villages. Give them a quick covid test (I note that the desire to sample blood indicates that maybe they know the nasal swabs are inaccurate).

There are arguments for and against several different primary outcomes. The best choice for the researchers was "symptomatic seroprevalence". Does it really matter what choices "random dudes on the internet" make after the fact? No, it's irrelevant.

Not following you here.

Sorry, I was trying to explain it in a different way to make it clearer. You can ignore it if I was unsuccessful.

Reconcile your statement about identifying asymptomatic villagers with what the report says here;
"Among the 335,382 participants who completed symptom surveys, 27,166 (8.1%) reported experiencing COVID-like illnesses during the study period.

The number of participants who said they didn't have symptoms = 308,216
The number of participants who said they had symptoms = 27,166
Total participants who completed symptom surveys = 335,382

More participants in the control villages reported incident COVID-like illnesses (n=13,893, 8.6%) compared with participants in the intervention villages (n=13,273, 7.6%). Over one-third (40.3%) of symptomatic participants agreed to blood collection.

Number of symptomatic participants who agreed to blood collection = 10,952 (from Table A1)
Number of symptomatic participants who didn't agree to blood collection = 16,214
Total number of symptomatic participants = 27,166

Omitting symptomatic participants who did not consent to blood collection, symptomatic seroprevalence was 0.76% in control villages and 0.68% in the intervention villages.

Number of symptomatic participants who agree to blood collection testing positive for COVID = 2293 (from post #86)
Number of symptomatic participants who agree to blood collection testing negative for COVID = 8659
Total number of symptomatic participants who agree to blood collection testing = 10,952

Are you now telling me that the study ends with what I wrote in my first paragraph?

No, I'm telling you that they collected the data listed above. After they collected that data, they summarized the results and presented those results in the paragraph you quoted.

If they aren't extrapolating and they're just saying that those people didn't report covid symptoms so they don't have it, no need to test, then we have to recalculate everything.

You have to. I don't as I've been following what they said all along.

 

[Eric Newhill]

Where did you get 76% and 68% from?

Right here, for 50th time
"Among the 335,382 participants who completed symptom sur- veys, 27,166 (8.1%) reported experiencing COVID-like illnesses during the study period. More participants in the control villages reported incident COVID-like illnesses (n=13,893, 8.6%) com- pared with participants in the intervention villages (n=13,273, 7.6%). Over one-third (40.3%) of symptomatic participants agreed to blood collection. Omitting symptomatic participants who did not consent to blood collection, symptomatic seroprevalence was 0.76% in control villages and 0.68% in the intervention villages. Because these numbers omit non-consenters, it is likely that the true rates of symptomatic seroprevalence are substantially higher (perhaps by 2.5 times, if non-consenters have similar seroprevalence to consenters)."

can't be extrapolating. Your claim is that they are taking information from a bunch of people who got sick.

Yeah duh. Of course that's my claim. Because that is what the study claims it did (see quote from above - now 51st time I've pointed it out to you)

only information we have from that subset is "what proportion of them were sick with COVID and what proportion of them were sick with something else". Then they applied that information to people who didn't get sick? In order for that to be extrapolation you would have to be saying this:

"308,216 people did not get sick. Of those people, 22% did not "get sick from COVID" and 78% did not "get sick from something else."

Huh? Why would I say that? I didn't say that

The study either ends with the blood serum test or those results get extrapolated to a larger population or this study is a complete miss mash of disjointed concepts and objectives (I think the latter) that you are defending in troll like fashion.

doesn't, because the pre-specified primary outcome was "symptomatic seroprevalence". When you pre-specify an outcome, you measure the pre-specified outcome. You don't measure something else instead.

OK. So you say now that the study ends with the blood serum test treatment villagers = 5,006 and 4,971 control villagers.

That's it. Those results define the study. If that's the case, then the study is baloney. You cannot make statements about how the masks worked (or didn't) for hundreds of thousands of subjects from a small subset who got blood tested).

number of participants who said they didn't have symptoms = 308,216
The number of participants who said they had symptoms = 27,166
Total participants who completed symptom surveys = 335,382



Number of symptomatic participants who agreed to blood collection = 10,952 (from Table A1)
Number of symptomatic participants who didn't agree to blood collection = 16,214
Total number of symptomatic participants = 27,166



Number of symptomatic participants who agree to blood collection testing positive for COVID = 2293 (from post #86)
Number of symptomatic participants who agree to blood collection testing negative for COVID = 8659
Total number of symptomatic participants who agree to blood collection testing = 10,952

I agree with those numbers. But what do they mean? How do they use them in a calculation to show significance?

Again, are you now telling me they calculated from the subset of serum positive? Like I did in comment 155?

If not, show your calculation. Put up or shut up time for you. I don't think you got it, but prove me wrong.

 

[Michael Larkin]

I wish I were better than I am at maths and stats, but hey, I'm not...

Near as I can tell, what happened was that they got villagers in a number of villages to agree to take part in the study. They had 146,783 from villages that went maskless and 160,323 from villages where people were supplied with masks. Is that correct so far?

Now: from my (admittedly rather naive) viewpoint, unless they had the opportunity and/or resources to test absolutely everyone who took part, they'd need to select a number of people from both groups (masked and maskless) to test for seropositivity for Covid.

Assuming a) 100% accuracy of the tests either way, b) that their selections were random and representative of whole populations, and c) that selections were sufficiently large for purposes of detecting statistical significance, then in theory, all would seem to be well.

However, if I got this right, their selections weren't entirely random. They only tested those who self-reported Covid-like symptoms. But Covid is known to be asymptomatic in some people, and, correct me if I'm wrong, few if any self-reporters were medically qualified to make the judgement.

How was it decided whether what they reported as Covid symptoms were in fact such? Were they questioned what their symptoms were? Were any of their self-reports rejected because they didn't actually indicate Covid? Were any of those who didn't self-report questioned to try to confirm they didn't in fact have Covid?

Anyway, like I said -- and if I've got it right -- self-reporting seems to have served as the basis for seropositivity testing. If so, right from the get-go, the study is potentially biased. I myself don't think one should ask the question at all. One should simply have tested sufficient numbers from each group regardless of whether they did or didn't report any symptoms.

Assuming, again, 100% accuracy of the seropositivity tests either way, that would seem to me to be the way to do it (happy to be corrected if wrong). Ideally, you'd test everyone in both populations, but you'd never get everyone to agree to be tested, and so the sample sizes are necessarily restricted. That wouldn't matter if the sampling was random, representative, and large enough to lead to statistically significant results.

Hence I'm tending to be sceptical of the study. There are so many possible confounding factors -- Most of all, not making a truly random selection, but one based ultimately on the reliability of reports from villagers about whether or not they had had Covid symptoms. And even if they hadn't had symptoms, so what? Not all people who contract Covid do have symptoms. There were very possibly some who were asymptomatic but would have been seropositive if tested.

Also, what exactly was the seropositivity test? Was it PCR-based? If so, how many amplification cycles were applied? 20, 30, 40, or what? The higher the number, the more likely the false positives.

Not only that: the very idea of identifying sample groups based on their self-reports kind of begs the question. IMHO, it would have been better to ignore the reports of mostly if not entirely non-medically qualified people altogether. The results could then have been presented in terms of:

There was an x number of people randomly selected from the maskless group, and a y number of people randomly selected from the masked group. It was found that the percentage of seropositive tests had statistically higher significance in the unmasked group.

Then the main confounding factor, at least to me, would be how accurate the seropositivity tests were. If of a type that was virtually 100% accurate, fair enough. If not, it'd still be open to question.

Forgive me if I haven't understood the maths and stats and my understanding is incorrect. Please tell me if it is, and why. I'm completely open to being corrected.

 

[Eric Newhill]

I wish I were better than I am at maths and stats, but hey, I'm not...

Near as I can tell, what happened was that they got villagers in a number of villages to agree to take part in the study. They had 146,783 from villages that went maskless and 160,323 from villages where people were supplied with masks. Is that correct so far?

Now: from my (admittedly rather naive) viewpoint, unless they had the opportunity and/or resources to test absolutely everyone who took part, they'd need to select a number of people from both groups (masked and maskless) to test for seropositivity for Covid.

Assuming a) 100% accuracy of the tests either way, b) that their selections were random and representative of whole populations, and c) that selections were sufficiently large for purposes of detecting statistical significance, then in theory, all would seem to be well.

However, if I got this right, their selections weren't entirely random. They only tested those who self-reported Covid-like symptoms. But Covid is known to be asymptomatic in some people, and, correct me if I'm wrong, few if any self-reporters were medically qualified to make the judgement.

How was it decided whether what they reported as Covid symptoms were in fact such? Were they questioned what their symptoms were? Were any of their self-reports rejected because they didn't actually indicate Covid? Were any of those who didn't self-report questioned to try to confirm they didn't in fact have Covid?

Anyway, like I said -- and if I've got it right -- self-reporting aeems to have served as the basis for seropositivity testing. If so, right from the get-go, the study is potentially biased. I myself don't think one should ask the question at all. One should simply have tested sufficient numbers from each group regardless of whether they did or didn't report any symptoms.

Assuming, again, 100% accuracy of the seropositivity tests either way, that would seem to me to be the way to do it (happy to be corrected if wrong). Ideally, you'd test everyone in both populations, but you'd never get everyone to agree to be tested, and so the sample sizes are necessarily restricted. That wouldn't matter if the sampling was random, representative, and large enough to lead to statistically significant results.

Hence I'm tending to be sceptical of the study. There are so many possible confounding factors -- Most of all, not making a truly random selection, but one based ultimately on the reliability of reports from villagers about whether or not they had had Covid symptoms. And even if they hadn't had symptoms, so what? Not all people who contract Covid do have symptoms. There were very possibly some who were asymptomatic but would have been seropositive if tested.

Also, what exactly was the seropositivity test? Was it PCR-based? If so, how many amplification cycles were applied? 20, 30, 40, or what? The higher the number, the more likely the false positives.

Not only that: the very idea of identifying sample groups based on their self-reports kind of begs the question. IMHO, it would have been better to ignore the reports of mostly if not entirely non-medically qualified people altogether. The results could then have been presented in terms of:

There was an x number of people randomly selected from the maskless group, and a y number of people randomly selected from the masked group. It was found that the percentage of seropositive tests had statistically higher significance in the unmasked group.

Then the main confounding factor, at least to me, would be how accurate the seropositivity tests were. If of a type that was virtually 100% accurate, fair enough. If not, it'd still be open to question.

Forgive me if I haven't understood the maths and stats and my understanding is incorrect. Please tell me if it is, and why. I'm completely open to being corrected.

Yes. Agree. Ellis is trying to dodge around all of that. it seems that the conclusion is based off only the blood tested and then they are trying to make a statement about hundreds of thousands of subjects based on that. That's it.

IMO, the study write-up is very confusing b/c they are jamming all kinds of tangential analysis into it. It's isn't about only "do masks work".It's all kind of metrics about compliance and even the favorite color of mask. Lots of smoke and mirrors and fluff. There are red herrings about regression analysis, but at the EOD, it seems that it's all based off the tiny not random sample that was blood tested.

 

[Alex]

So in conclusion can we say that out of the control villages (n = 146,783) there were 3,380 serum positive cases (distilled down from reported symptoms, agree to test, tested. tested positive) and out of the intervention villages (n= 160,323) same distillation leaving 3,804 serum positive cases.

If the villager didn't report symptoms s/he was deemed to not have covid.

Then a chi squared would be based on this:

----------- positive test ---- negative symptoms
treatment 3,380 --------- 156,519
control. 3,804------------ 143,403

........and the results are not significant. The study flops and the media reports are wrong.

how do you get from 3,804 positive blood samples to .76%

how do you get to 9.3% reduction