The methodology is completely pulled out of his ass.
There's a lot wrong with it but the main thing is using linear regression for "do you know someone who has died of X?" This is cannot be a linear relationship. As the number of casualties goes up, the % of people who know a casualty logarithmically approaches 100%.
This means the % of people who know a casualty will rise dramatically at first, and taper off. It also means it's not a good indicator for actual deaths.
Disregard my previous post. You are completly rigth. I apologise for psting it. I just found an estimate that seemed plausible and had an explanation without cheking it properly. I have now read the rest of that guys posts and i realize i look like a cretin promoting him.
Thats a hood point. I did not think of that. Sorry. I guess there arent enough datapoints for a logarithmic regresion.
The question is if at 60% there is enough deviation from the linear function?
As more people die. Intervewed people would know more than 1 victim leading to undercounting. Is this efect enough to counteract the logarithmic trend?
The methodology is completely pulled out of his ass.
There's a lot wrong with it but the main thing is using linear regression for "do you know someone who has died of X?" This is cannot be a linear relationship. As the number of casualties goes up, the % of people who know a casualty logarithmically approaches 100%.
This means the % of people who know a casualty will rise dramatically at first, and taper off. It also means it's not a good indicator for actual deaths.
Disregard my previous post. You are completly rigth. I apologise for psting it. I just found an estimate that seemed plausible and had an explanation without cheking it properly. I have now read the rest of that guys posts and i realize i look like a cretin promoting him.
No need to apologize. Just remember to meet independent media with the same level of skepticism as main stream media.
Thats a hood point. I did not think of that. Sorry. I guess there arent enough datapoints for a logarithmic regresion.
The question is if at 60% there is enough deviation from the linear function? As more people die. Intervewed people would know more than 1 victim leading to undercounting. Is this efect enough to counteract the logarithmic trend?