Posted by bratao 1 day ago
https://www.chicagomag.com/chicago-magazine/september-2018/c...
I’m not disputing that there are literally toxic workplaces or environments that do correlate with cancer clusters.
when there are billions of people in the world it is expected that some where several get cancer at the same time.
So, you're probably not far off
That would be the answer if we asked whether such a coincidence ever happens in the world. In this specific case, the question is, 'what are the most likely causes?'
To calculate a p-value (roughly spoken), you need to start with a single hypothesis. Then you gather data and the p-value gives you the probability that your data occurs while your hypothesis is false. When you start with a finite set of multiple hypotheses, you need to take that in to account when calculating your p-value.
When you start with data and come up with a hypothesis afterwards, you would have to find the whole potential space of all hypotheses. So, for example, how many hospitals are there? Do you only consider US? Do you only consider nurses or other employees as well? What about only four nurses would that have made it to the news? What about other forms of cancer? What about time? Do you consider the time period of the last 50 years? As you think about what might have made the news, the set of hypotheses grows bigger and bigger and as it approaches infinity, the p-value for any data would approach one. Because when you have a very large set of unlikely hypotheses, the probability that your data accidentally supports one of them is quite large.
That's what parent was talking about.
The American Cancer Society says that in order to meet the definition of a cancer cluster, occurrences must be the same type, in the same area, with the same cause, and affecting a number of people that's "greater than expected" when a baseline for occurrences is established.
“Nearly 4 out of 10 people in the United States will develop cancer during their lifetimes," the society said on its cancer clusters webpage. "So, it’s not uncommon for several people in a relatively small area to develop cancer around the same time."
If (for example) twenty percent of the time 500 people were scanned, five at least had benign brain cancers, would this report be unusual or suspicious in itself?
That seems circular, like it's not a cancer cluster until we find out that it's a cancer cluster, or it's not a cancer cluster because we didn't determine a common cause, so don't worry too much about there maybe being a common cause that would make it count as a cancer cluster.
This underlines how stats are no substitute for reasoning about mechanisms.
https://futurism.com/neoscope/high-school-classmates-rare-ca...
[1] https://en.wikipedia.org/wiki/Devil_facial_tumour_disease?wp... (content warning)
[edit] I just rolled 60 dice and got a string of 7 2’s in a row. Can someone calculate the odds of that happening for me?
run an RNG that doesn't quantize to an integer and see how many repeats you get then:)
I'd guess it was their lawyers...