4 ms·

The title "Starting a Company Can Turn You into a Psychopath" is complete nonsense when paired with the linked article. Correlation is not causation. With tha

The title "Starting a Company Can Turn You into a Psychopath" is complete nonsense when paired with the linked article. Correlation is not causation. With that being said, I've seen founders become lots of things both good and bad. The experience of doing something as challenging as creating a new business is certainly the type of endeavor which will likely forever alter your view of the reality in which we reside.

dnautics 13y ago
those two curves look the same to me. "Average" is a terrible metric, especially with the spread. My eye estimates the "p-value" for difference between those curves is about .8, depending on what distribution you think it is.
- glaugh 13y ago
  Well, p-value is <0.00001, but that's because it's a pretty big sample. You're right that the effect size is "small" (Cohen's d), which as you suggest you can basically see from the curves. "Advanced" tab here: https://www.statwing.com/open/datasets/88ed02bd6dda2f813ba98d5e01a6f239d7a5359c#workspaces/3859 https://www.statwing.com/open/datasets/88ed02bd6dda2f813ba98... Edit: I think I'm not allowed to respond to the reply to this comment for flamewar prevention reasons, so putting it here. You can also look at the nonparametric version of the test if you don't buy that parametric is good here (though I'd disagree because of the central limit theorem). p-value there is 0.00004
  - dnautics 13y ago
    I don't believe that p value. the non-founder curve looks fine, but the founder curve has bins with some pretty serious deviations from expected, so you are overestimating your confidence in your curve fit.
    - RyanZAG 13y ago
      The curves seem to look a bit deceiving, probably something in how our eyes work - switch to the 'count' tab instead of the 'percent' tab. It's much easier to see then just how much 'flatter' the founder curve is compared to the non-founder curve, and this flatness pushes up the deviation, but it pushes up the average far more and puts the confidence interval well out of range of the non-founder curve. Judging based off just the visual curve is not a great idea as our brains are wired up for matching patterns and not for working out these kind of differences properly. It's why math is such a good idea when making decisions.
      - dnautics 13y ago
        you're kidding right? Because I can make the non-founder curve just as "flat" by yanking up the scale on the percentage graph to 10,000%. The reason why it looks "flat" is because when you go to "count" the website scales both curves the same y-axis. I think you mean to say something like "kurtosis" - which is, if anything, obscured to the eye by the process of flattening by scaling. And math is not a good idea when you make assumptions like normality of curves which are absolutely not normal. In this case, using the t-test to calculate a p value. I think reification of poorly done math is a bigger problem than math by eye.
glaugh 13y ago
Agreed. Sort of a satire of how articles will take some research and draw wild conclusions about it for the headline. We also considered "Psychopaths start companies" and "Some third variable relates to both psychopathy and starting companies", all of which are equally valid hypotheses :)
diego 13y ago
You missed the [the title of this post is tongue in cheek] part.
- jessaustin 13y ago
  Humor is impermissible in HN titles!!1!
BMarkmann 13y ago
Of course, you could also assume that having an altered view of reality leads one to believe it's reasonable to strike out on their own (and in some cases start their own business). Again, doesn't imply causation, but it's possible way to flip the relationship on its head.
angrow 13y ago
Posting on Hacker News Can Turn You Humorless

dnautics 13y ago

those two curves look the same to me. "Average" is a terrible metric, especially with the spread. My eye estimates the "p-value" for difference between those curves is about .8, depending on what distribution you think it is.

glaugh 13y ago
Well, p-value is <0.00001, but that's because it's a pretty big sample. You're right that the effect size is "small" (Cohen's d), which as you suggest you can basically see from the curves. "Advanced" tab here: https://www.statwing.com/open/datasets/88ed02bd6dda2f813ba98d5e01a6f239d7a5359c#workspaces/3859 https://www.statwing.com/open/datasets/88ed02bd6dda2f813ba98... Edit: I think I'm not allowed to respond to the reply to this comment for flamewar prevention reasons, so putting it here. You can also look at the nonparametric version of the test if you don't buy that parametric is good here (though I'd disagree because of the central limit theorem). p-value there is 0.00004
- dnautics 13y ago
  I don't believe that p value. the non-founder curve looks fine, but the founder curve has bins with some pretty serious deviations from expected, so you are overestimating your confidence in your curve fit.
  - RyanZAG 13y ago
    The curves seem to look a bit deceiving, probably something in how our eyes work - switch to the 'count' tab instead of the 'percent' tab. It's much easier to see then just how much 'flatter' the founder curve is compared to the non-founder curve, and this flatness pushes up the deviation, but it pushes up the average far more and puts the confidence interval well out of range of the non-founder curve. Judging based off just the visual curve is not a great idea as our brains are wired up for matching patterns and not for working out these kind of differences properly. It's why math is such a good idea when making decisions.
    - dnautics 13y ago
      you're kidding right? Because I can make the non-founder curve just as "flat" by yanking up the scale on the percentage graph to 10,000%. The reason why it looks "flat" is because when you go to "count" the website scales both curves the same y-axis. I think you mean to say something like "kurtosis" - which is, if anything, obscured to the eye by the process of flattening by scaling. And math is not a good idea when you make assumptions like normality of curves which are absolutely not normal. In this case, using the t-test to calculate a p value. I think reification of poorly done math is a bigger problem than math by eye.

glaugh 13y ago

Agreed. Sort of a satire of how articles will take some research and draw wild conclusions about it for the headline. We also considered "Psychopaths start companies" and "Some third variable relates to both psychopathy and starting companies", all of which are equally valid hypotheses :)

diego 13y ago

You missed the [the title of this post is tongue in cheek] part.

jessaustin 13y ago
Humor is impermissible in HN titles!!1!

BMarkmann 13y ago

Of course, you could also assume that having an altered view of reality leads one to believe it's reasonable to strike out on their own (and in some cases start their own business). Again, doesn't imply causation, but it's possible way to flip the relationship on its head.

angrow 13y ago

Posting on Hacker News Can Turn You Humorless