Hacker News Reader

Comment by rodionos | Hacker News Reader

This discussion about coffee has become too scientific. When I was a child, my parents would not let us drink coffee, but tea was not only allowed it was part of daily routine as long as I remember. We didn't question. When I grew up someone told me the story about two life-term prisoners in medieval times, one of them allowed to drink only tea, the other coffee. The coffee-guy lived longer. That was enough for me to stake my rights and start drinking coffee in high school.

View on HN

If you're basing decisions off a single anecdotal* data point from "medieval times", I can see how anything else would seem "too scientific" in comparison.

* And fictitious

I made a lot of good decisions based on incomplete data, and I'm not sure why you labeled this story fictitious. Here's some background: https://en.wikipedia.org/wiki/Gustav_III_of_Sweden's_coffee_...

If you want to drink coffee, drink coffee. Using a goofball medieval study (with a sample size of literally 2 people) to justify your personal choice is pretty silly.

Your problem with sample size in this particular case is that you fail to account for its duration. Convert sample size to temporal units and you might draw different conclusions since p-values might not look as bad now. An average FDA trial is 3 years, so assuming the prisoners lived for 40 years on their prescribed caffeine diet, you get sample size of 2*40/3 = 26.7. Not bad for 17th century. Last but not least, turn off --verbose mode.

Are you seriously comparing an FDA trial to a 17th century trial? If you don't see the absurdity of trying to equate the quality of those two I don't know what to tell you, setting aside your idea that a "26.7" sample size would be anywhere near significant enough to draw any sort of meaningful conclusion.

That just made my day.

This fallacy is known as "pseudoreplication": the erroneous belief that multiple samples (or a long measurement) of dependent data (i.e. the same person year after year) is equivalent to getting many independent samples.

If you measure the same two people year after year, you just measure their genetic predispositions and personal habits more and more precisely. You cannot distinguish them from the results of your coffee/tea intervention.

http://www.statisticsdonewrong.com/pseudoreplication.html

In case of caffeine addiction tests, I would argue that a life-long trial of fewer people over a short-term trial of many people would produce more statistically significant results because the impact is accumulative over a long period of time. Consider asbestos or even smoking. Compare the results of 10 lifelong smokers vs 1000 smokers who tried it just once.

Anyways, did you read the article on wikipedia? The prisoners were twins and their rations was the same - 3 intakes per day.

The goal of a trial is to eliminate other possible explanations for the phenomenon in question. You take a large sample, and randomly assign people to groups, so all the idiosyncratic variations average out.

With only one person per group, you can't eliminate many possible reasons for their differing life spans: one may have caught a disease the other was not exposed to, they may have had different nutrition, different amounts of exercise, different habits, a zillion other things.

After all, do other twins die at exactly the same time? No. They die years apart for many other reasons. Unless this twin experiment could eliminate all those possible reasons, it shows us nothing about coffee specifically.

Calling this medieval is doing it a disservice. The study took place in the late 18th century, during the height of the enlightenment.

Interesting. How did you determine that there were not other factors at play that caused the outcome of the one prisoner who drank coffee to live longer?

Also, this is supposed to be a joke, right? :~)

Medieval tea-drinking and coffee-drinking in the same place? Where might this have been?

Sweden: https://en.wikipedia.org/wiki/Gustav_III_of_Sweden's_coffee_...

Hardly medieval!

Reply on news.ycombinator.com