Home | Papers | Blog

Best time to post on HN & other time-effects in the HN community

Trinity term (the 3rd) has just started in Oxford, and I (Wybo) have just handed in two 5000-word papers.

The first paper is about circadian (24-hour scale) time-effects on Hacker News. Its main hypothesis is that the time at which people arrive on the site (greatly) impacts whom they are most likely to interact with. This because replies that make very similar points as previous replies to the same post, are generally not appreciated very much, and visitors thus can only (productively) reply to relatively new posts (not older than 2 hours).

Some time-pressure effects were indeed found, both on the micro-scale of interactions between prolific users (more than 25 posts in 40 days), and on the aggregate level, in 3-hour time-windows, where there was greater reciprocity and transitivity (friends of friends becoming friends). Even the phased introduction of daylight saving time between the US and UK (UK is 2 weeks later) was found to have a measurable impact on the community-structure.

Circadian rhythms thus do have an impact, even if a small one, on the reply-relationships that form (see the paper for limits and possible alternative explanations).

A very short video of the network over a day (aggregate for 40 days), and its fragmentation by time, can be seen below:

In addition – specifically for this blog-post – I graphed the best time to reply to comments, and the best time to create threads, both in terms of their expected ratings (the average ratings). This data was from just before karma-ratings for comments were hidden. See the graphs below:

Average ratings for replies (posts that are comments at any level in the thread) created at each hour of day (in UTC). Notice how little variation there is.

Average ratings for threads created at each hour of day (in UTC). Again notice the very modest variation.

As can be seen there is not really a best time to post, because ratings are roughly constant during the day (though 19:00 UTC is not the worst time and there are some more variations during the week (especially Saturday evenings in US timezones are lower rated)). It should be noted that only posts that made it to the frontpage were included, see here for a study of the new-page.

This relative lack of variation is (some) evidence against concerns about overnight threads standing less of a chance, that were expressed in this HN thread.

A possible reason for there being so little variation, is that when more people visit and create threads, threads are on the homepage for shorter periods (turnover is higher), and thus about the same number of people see, and rate them at any time. See how the gap in thread-longevity precedes the peak creation-period in the graphs below.:

Threads created at each hour.

Time on the frontpage for threads created at each hour (and on the frontpage for at least one hour).

The paper can be downloaded here (30mb, pdf, lot of HQ images). The tools (in Ruby 1.8) used in the research are available here (though they are a bit rough around the edges). And the full anonymized dataset (to the extent that removing nicknames is sufficient, yaml file format) can be downloaded here.

The second paper was about how an online global advisory parliament employing proxy-voting as its voting-mechanism, could improve political decision-making. And secondly it is about what methods might be employed to make it more likely to attain critical mass, such as integrating it with Facebook, and making it follow the agenda of real parliaments & summits rather than setting its own.

Critical mass in thread-based platforms/forums is my thesis-topic. More on this later, but a taste of what I am working on can already be had from this JS + HTML 5 canvas, agent-based simulation of users posting on a forum (thesis-proposal is here).