Thread by @connolly_s

Latest Israeli case data... @RanIsraeli @MLevitt_NP2013 it's complicated, but I'm going to try and explain what it says and what we can conclude. First the background 1/n

So there have been these spreadsheets popping up on social media reporting the age breakdown of cases and the percentage of the population fully vaccinated and highlighting the similarity of those numbers, citing datadashboard.health.gov.il/COVID-19/gener… as the source 2/n

Using that data I was able to compute statistical Z-scores, but as I do not speak Hebrew, my ability to verify the numbers was limited. Those Z-scores implied a significant difference in the case rate between vaccinated and unvaccinated with unvaccinated lower! 3/n

Now I must caution, the Z-score does not tell us why there is a difference between the two case rates, it only says if the difference is statistically significant or if it is more likely just pure chance. A Z-score between -1.67 and +1.67 is not considered significant 4/n

Some of the Z-scores computed using that data were as high as 7, which is a 1 in 16,000 chance of being a fluke. What could be the reason, well I hypothesized at least two explanations:
1. vaccination makes you more susceptible
2. unvaccinated less likely to get tested
5/n

Now this was also data that I couldn't verify easily, but it would be good to verify, so I used some tricks of the software engineer to dig and I was able to find data.gov.il/dataset/covid-… which happens to be the page with the age break down of cases by vaccination status 6/n

The screenshots of spreadsheets had the exact same number. Case closed, the data was verified... well not quite... there is the question of the % of the population vaccinated... those numbers are a little harder... but google translate shows them on the dashboard 7/n

And guess what, the vaccination rates also are the same. Case closed both data sets were verified... well we'll get to the question of whether the case is closed or not later. There were further rumours that the spreadsheets had stopped being published 8/n

However, now that I had the source data set I was able to see that it takes until Wednesday to collate the data set and update it, so here I was waiting for last weeks data set to show up today... and that it did... I don't know if the screenshots have yet, I have raw data 9/n

So I put in the numbers and got this. What appears to be three significantly lower rates for the unvaccinated (20-29, 40-49 and 60-69) and one significantly lower rate for vaccinated (90+) similar to the other weeks... why this long thread Stephen? 10/n

Well there was something niggling me... namely the dark green portion of the bars on this graph in the dashboard... after a bit more bashing you can see that we have three groups not two. I had been subtracting the vaccinated from 100 to get % unvaccinated 11/n

But what we really need to do is analyze all three groups: fully vaccinated, partially vaccinated and unvaccinated... so I added some columns, but when we remove the 6.9% of 20-29yo that are partially vaccinated the base rate for unvaccinated changes from 2703 to 3604 12/n

So this gives us a different table. We now have three groups to compare so we have three sets of Z-scores. I also give the P-values which are the probability that the observed difference could occur by chance, 0 is never, 1 is always 13/n

So what can we see:
1. The partially vaccinated have the lowest case rate. Significantly better than the fully vaccinated Z-scores of 10+!!! despite the low population %)
2. The unvaccinated have the highest case rate and it is significantly higher than the other two
14/n

But remember, the statistics do not tell us WHY there is a difference, only if the difference is significant. So let's start with the elephant in the room... 15/n

Why are the partially vaccinated having fewer cases than fully vaccinated?
1. The naïve assumption: "it's the second shot what done it".
2. My hypothesis is that side-effects overlap symptoms, so partially vaccinated don't get tested as "it's just a side-effect"
16/n

Essentially all explanations boil down to one of two theories:
1. Both groups report for testing at the same rate and the difference is because of the vaccination status
2. The groups report for testing at different rates.
17/n

Given the clinical trials showed an increase in effect between one shot and two shots, when comparing partially vaccinated with fully vaccinated I find it hard to believe the "it's the second shot what done it" theory. On balance I believe the reporting not case rates differ 18/n

IF I accept that it's reporting rates not case rates for partially vs fully THEN I have to assume that the reporting rates are different for each group. That becomes my base assumption as the simplest theory to explain all the data 19/n

Occam's razor's current working theory:
* unvaccinated are most likely to get tested => unvaccinated cases most likely to get detected
* partial vaccination status dismiss symptoms as side-effects => least likely to get detected
* fully vaccinated believe not at risk 20/n

Next Steps: If there is a dataset showing testing numbers (not case numbers) broken down by vaccination status, that data set would allow us to challenge this current working theory as we could work out the testing rate Z-scores 21/n

Notes: The most common objection to my earlier (incorrect) analysis was that people believed at this stage the unvaccinated were refuseniks and avoiding getting tested. I think we can dismiss that theory *for now*. Unvaccinated are most likely to get detected from this data 22/n

Notes: My personal bias is that I believe the vaccinated, by the choice of associating the novel mRNA treatments with the word vaccine, have a false understanding of the protections they are afforded, and they do not get tested as much. 23/n

Notes: Also if the vaccine does what it claims to do, i.e. reduce symptoms, then -as those with more symptoms are more likely to get tested- we would expect to detect fewer cases in the vaccinated group. Random testing the population could test that question 24/n

Conclusion: this data is just showing that the different groups get tested at different rates and therefore we detect cases at different rates. Partially vaccinated dismiss symptoms as side-effects. We known Fully vaccinated have symptoms suppressed. So no suprises here 25/25

@threadreaderapp unroll