A large amount of Show HNs are not startups; they are personal projects.
Relatedly, using "the startup is not dead" as a metric for the startup alive is a bad idea. People do not shut down everything when a Hacker News submission does not get upvotes (especially in the case of Show HN, where many are hosted on GitHub pages and are free to host, although you account for that in the regression).
The regression has a few issues:
> It's a classification problem (alive or dead) so OLS doesn't make sense.
> Score and Comments are multicolinear and cannot both be in the same model.
> You don't answer or give statistics toward "how well" the model predicts.
> Related to the comment earlier, you don't comment on the magnitude of the on_GitHub coefficients, which are huge and skew the entire result of the regression!
While I always appreciate analyses of HN data, the conclusions raise more questions than answers.
Relatedly, using "the startup is not dead" as a metric for the startup alive is a bad idea. People do not shut down everything when a Hacker News submission does not get upvotes (especially in the case of Show HN, where many are hosted on GitHub pages and are free to host, although you account for that in the regression).
The regression has a few issues:
> It's a classification problem (alive or dead) so OLS doesn't make sense.
> Score and Comments are multicolinear and cannot both be in the same model.
> You don't answer or give statistics toward "how well" the model predicts.
> Related to the comment earlier, you don't comment on the magnitude of the on_GitHub coefficients, which are huge and skew the entire result of the regression!
While I always appreciate analyses of HN data, the conclusions raise more questions than answers.