Iker Quesadillas
Full Member
- Joined
- Mar 12, 2021
- Messages
- 4,964
- Supports
- Real Madrid
I am curious what do you make of the lack of correlation between female and child deaths? I guess it could be a case of families being increasingly separated or a process issue (i.e. counting female bodies on day x and the "correlating" child bodies on day x+1)
The author explains that the number of women and number of children killed should be highly correlated. He finds no correlation and claims this means the numbers must be called into question.
But here's an issue. The author's hypothesis is that the health ministry "settled on a daily total arbitrarily. Then they assigned about 70% of the total to be women and children, splitting that amount randomly from day to day." If "about" 70% of the total is always set to be women and children, then the total should be highly correlated to women+children. But in the data set that the author uses, women+children is not highly correlated with total. The data set is here. You can plot the data on Excel. The R2 value is 0.228.
In other words, the author is claming that a relationship requires correlation, then proposing a theoretical relationship (70% generated from the total) that is not supported by a correlation. Can't do that!
The average W+C daily toll %, if you calculate it from the table, is 72 ± 21%. That is a large amount of variance, and there are massive outliers in either direction. The reason why there are massive outliers is because there are days when some numbers are very low. For example, on days 4 and 11, the death toll for women is zero. This raises two questions: 1) what are the odds that a 'random' split gives you zero twice in fifteen days, and 2) if you are generating numbers via some random generator, or generating them yourself, why would you even make zero a valid output?
This isn't meant to be a "debunking" or anything like that. It's more to show that, contrary to what Giggsy PO stated, it's not "just maths." There are choices made in what you present, ignore, highlight, downplay, etc.
Last edited: