Even if statistical analysis checks out, the premise seems to be all wrong. Speaking generally now. I rarely read HN comments as having predictive intentions, but rather constructive (or not) criticism on things as they stand. Listening to criticism (i.e. negative feedback) about your product can be the difference between an amazing product and a mediocre products with a slow, drawn out fading into mediocre land.

Well, they started out by picking successes. A more valid approach would be to list all startup ideas mentioned on HN and see how they turned out. That includes every acquishutdown mentioned.

But that's vulnerable in the other direction. If 90% of startups fail, it's very easy to predict failure.

An interesting non-statistical inquiry would be to look at the problems people are complaining about and seeing whether they're still present in the successful projects (e.g. dropbox privacy).