Imagine threatening to put people in jail for criticizing your paper.
So anyway here are my critiques. ๐งต
So anyway here are my critiques. ๐งต
Chi-square test results here. Image from paper. A few words on the chi-square test and the sample used.
Large sample sizes with chi-square tests will often be statistically significant but meaningless.
Even randomly generated samples will usually be statistically significant from one another in a chi-square test of independence at around 500 or so.
Even randomly generated samples will usually be statistically significant from one another in a chi-square test of independence at around 500 or so.
These can tell you if a difference exists between groups, but not that the difference is due to your treatment. You can't distinguish effects from noise at that point.
The way the author uses statistical significance also seems to confuse it with effect size (eg phrasing like "highly significant"). A lot of focus on small p-values.
Size of p-values doesn't quantify the size of an effect or the strength of the evidence.
Size of p-values doesn't quantify the size of an effect or the strength of the evidence.
What you need is a measure of effect. Since statistical significance tells you very little, you want to know the strength of the association. For a chi-square test you could use Cramer's V.
Here's the data plugged into a calc: effect size is 0.004. Or basically nothing. .1 is when an effect is considered "weak."
This is because Cramer's V is also sensitive to sample size and 24 vs 300,000 is small.
This is because Cramer's V is also sensitive to sample size and 24 vs 300,000 is small.
Relevant paper; large samples will often be statistically significant but meaningless.
researchgate.net
researchgate.net
What's interesting is that if you add just one additional death in the treatment group it wouldn't be a significant relationship.
Basically the power in unbalanced chi-square tests rely almost entirely on the small sample.
When you have a sample of 24 and the difference between significance or not is one person, you can't rule out that you're on the cusp of an error regardless of what your p value (very small due to the 300,000 comparison group) tells you.
It's funny because it looks like at least one patient may have begun treatment, died, and was marked as excluded. Pretty convenient they weren't included in the analysis.
I have seen some discussion on if having a contingency table with zero values (eg: 24 alive, 0 dead) is a problem or not. I'm inclined to think it isn't inherently, but worth noting that SPSS will give you an error warning for this.
On method design in general:
Full third of the 24 patients in the treatment condition were also on a cocktail of other drugs:
Full third of the 24 patients in the treatment condition were also on a cocktail of other drugs:
Given that, how can we attribute the relationship to ivermectin and not any of the other drugs?
And given that all patients were not only on ivermectin, but on additional supplements, how do we know ivermectin had an effect and not, say, vitamin D alone?
And given that all patients were not only on ivermectin, but on additional supplements, how do we know ivermectin had an effect and not, say, vitamin D alone?
Or the difference between the treatment group being non-hospitalized and chilling at home, while the comparison group was people who were hospitalized?
This guy checking the references.
Perhaps unsurprising that many of the references used in the paper either don't say what she claims, or have been retracted.
Perhaps unsurprising that many of the references used in the paper either don't say what she claims, or have been retracted.
Incidentally I don't especially care about this topic or the treatments. You all know I never write about this.
I only had a look because "if you criticize my paper I will put you in jail."
Unsurprising there was much to criticize.
I only had a look because "if you criticize my paper I will put you in jail."
Unsurprising there was much to criticize.
"FBI please look into this website where people are allowed to openly peer review and criticize my paper."
Loading suggestions...