Ask PG: Is d(users)/dt constant on HN? | Hacker News Reader

Ask PG: Is d(users)/dt constant on HN?

19 pointsfrisco15 comments

Eternal September wasn't so much an issue of "too many users" as "too many new users that couldn't be taught local customs before they overwhelmed said local customs." If it isn't roughly constant, especially over longer periods (i.e., quarter-over-quarter growth), you could work out when either you'd expect HN to fall victim to an Eternal March (demo day + applications = lots of press and spike in users?), or when you'd need to work out a different new user welcoming process?

View on HN

It looks like it's slightly steeper than flat. At least so far. Growth should flatten out at some point.

It's hard to predict social phenomena. I just try to react quickly when problems arise.

Wait - how do you know if something is a problem if you can't predict it? Problems are defined by their effects, right?

You look at the effect.

If frisco is correct, that sites degrade because of a too high rate of growth, this presents an extremely simple way to prevent that: rate limit new account creation.

You could cap it right now, and see what happens to the quality of the site.

This seems like a very good idea in terms of controlling spikes in comments - but you have to question whether this is the (main) cause of the degradation.

If limiting new accounts seems like the best solution, it should be done clearly and transparently - as the users you are trying to weed out are generally fickle and will not stay, so won't bother to check / sign up later. Let users know the reason why they can't sign up now, and when they should be able to.

However, users that have a legitimate need to comment (eg site / app owners) need to be allowed access - solving this could be trickier.

Let people input an email to be pinged when a spot in the queue opens up. Most people won't.

Or have glass walls. Facebook's growth rate is enormous, but it still retains its character because your personal Facebook's growth rate is actually rather small.

Not that this has any obvious parallels to HN.

Making it easy to track users I know would make me read and upvote their comments more. Letting me ignore someone (and making them invisible to the public if enough people ignore them) is another wall.

I think that this would just skew the leaderboard. Everyone would be following the famous members of HN (pg, bucheit, etc), and locally popular users would be distant seconds, while the normal user would make up a long tail of comparatively low karma.

I'm pretty high on the leaderboard (~30) and don't think the order of the top 100 matter. I wouldn't mind being removed from the list. Submissions matter more than comments. Also, the age of your account will matter more as the site grows.

PG - any chance that you guys are planning on releasing quarterly high level reports from HN? I did some searching for it, but couldn't find anything in the past (I'm a fairly recent user, but couldn't find any references). It'd be interesting in a navel-gazing sort of way to see what the HN crowd looks like.

Also, a lot of HN discussions are self-referential (esp. when discussing user testing, etc.). It would be really helpful to have some data to add to these discussions.

It looks like some people were simultaneously banned following that thar' flamewar (a bunch of posters disappeared at the same time, and there's chatter) so d(users)/dt can be negative.

leaker, d(users)/dt can be negative without bannings.

d(users)/dt is growth in users, not their number. To illustrate: if 600 users registered in February, then only 400 in March HN's growth would be negative; d(users)/dt would be negative.

Erm, wouldn't that be the second derivative?

d(users)/dt would be the rate of change of users. That can only be negative if the amount of users decreases, ie through bannings.

You're thinking of the second derivative, d^2(users)/dt^2, which is the rate of change of the rate of increase of users. If that is negative, it means that the growth has slowed down, as in your example with February and March.

I was asking about the second derivative; if d(users)/dt wasn't constant, then d^2(users)/dt2 is nonzero and eternal september risk is fairly high (assuming HN isn't shrinking). If you assume idealized user growth (which doesn't happen -- there are so many people in the end that would care to either join HN or troll HN), then eternal september is actually guaranteed if d^2(users)/dt2 > 0, since there will come a time t when (new users) > (old users).

Reply on news.ycombinator.com