The Inspection Paradox is Everywhere: a surprising statistical illusion
towardsdatascience.com
The Inspection Paradox is Everywhere: a surprising statistical illusion
1–5 of 5 posts
Re: The Inspection Paradox is Everywhere: a surprising statistical illusion
#2This is somewhat eye opening to me, in the same way that Miller and Sunurjo's work was.
> In other words, selecting which part of the data to analyze based on information regarding where streaks are located within the data, restricts your choice, and changes the odds.[0]
I'm not particularly great at statistics, so I fall trap to a lot of these kinds of misconceptions. I appreciate articles like this getting shared, it's incredibly interesting.
[0]: https://jasoncollins.blog/2018/06/28/explaining-the-hoACt-ha...
Re: The Inspection Paradox is Everywhere: a surprising statistical illusion
#3This is somewhat eye opening to me, in the same way that Miller and Sunurjo's work was. > In other words, selecting which part of the data to analyze based on information regarding where streaks are located within the data, restricts your choice, and changes the odds. [0] I'm not particularly great at statistics, so I fall trap to a lot of these kinds of misconceptions. I appreciate articles like this getting shared,…
Re: The Inspection Paradox is Everywhere: a surprising statistical illusion
#4Missing the weirdest one, which I sadly can't find on the net right now because I lost the name:
What's the average population size of alien planets when you are one in 7 billion having this question.
I don't quite get it together but it had 2 interesting problems:
- It allows for a prediction with a simple size of just 1 which differed from said sample
- it's debatable if we should calculate with 7 billion humans or one globalized "intelligent" race
If anyone could point me to a source to this I'd be happy :)
also related: the german tank problem https://www.eadan.net/blog/german-tank-problem/
Re: The Inspection Paradox is Everywhere: a surprising statistical illusion
#5Interesting article, especially when you think about public polling (political or otherwise). Polling data is going to be heavily biased with people who are willing and have the time to be polled, which may be vastly different than the audience you are trying to model. Seems like the author is talking specifically about something a little more quantitative, but to the general point about these kinds of samples being biased in ways that aren't obvious at first, it has something in common.