Here's an even simpler way to think about it: it's the left point of the standard 95% confidence interval from the Central Limit Theorem plus a hack for small s

Here's an even simpler way to think about it: it's the left point of the standard 95% confidence interval from the Central Limit Theorem plus a hack for small sample sizes. The Wikipedia page says the hack is almost equivalent to estimating p = (X+2)/(n+4) i.e. assuming each item starts with two upvotes and two downvotes.