Obligatory Plug (let’s get this out of the way): I’m the co-founder of RJMetrics. Our hosted software helps e-commerce and SaaS businesses make smarter decisions using their data. If you’d like to see what RJMetrics can do for your business, sign up for a free 30 day trial. OK, onto the post…
I recently conducted a comprehensive Hacker News Data Analysis to settle a bet with my co-founder Jake (he won, by the way). As a byproduct of the experiment, I was sitting on a wealth of data about Hacker News submissions and their performance.
One of our team members suggested that we could use this data to learn more about how to maximize our chances of our Hacker News submissions making it to the front page. Rather than hoard the data, we decided to share it with the world in a blog post. Here’s what we found.
When to Post
We set out identify the most attractive “day of week”/”time of day” combination for making a submission to Hacker News. We started by looking at a simple graph of the total submissions made by weekday/hour so far in 2012. (Note that you can mouse over this chart to see specific data points.)
This shows the number of posts that are vying for the top slot at any given week/hour combination. Since the front page algorithm is largely influenced by the number of upvotes and the time since submission, I think these hourly slots are pretty good proxies for competitive cohorts. If you’re one of the top performers in your hourly cohort, you’ve got a very good shot of some front page attention.
Unfortunately, the sheer volume of posts isn’t a great indicator of your likelihood of hitting the front page. Let’s dig deeper.
Let’s face it: there are a number of stories that get submitted to Hacker News that have no shot at succeeding.
When considering the competition, it’s probably best to exclude posts that don’t represent a real competitive threat. Below, we show the percent of posts for each weekday/hour combination that receive only one vote (presumed to be duds).
The percentage of posts that go nowhere is staggering — north of 50% at times. The proportion of these weak posts is highest in the early morning hours (4-8AM Eastern) and lowest at night (8-10PM Eastern).
As a next step, let’s factor in the total number of votes that get cast in any of these hours.
In any given hour, we assume that there are a fixed number of readers who are likely to dole out a somewhat predictable number of points to a finite number of articles. To quantify this, we looked at the total number of points awarded to articles from each day/hour combination. (Note: we subtracted one point from each article’s score since every article gets a point just for being submitted and we only wanted to consider votes from the community.)
We then divided this number of points by the number of “potentially competitive” posts that we calculated above. With this analysis, we can see the points in time during which there are “the most votes to go around” per legitimate submission. In other words, peaks on this chart are times when the number of upvotes being cast is disproportionately high compared to the number of posts being submitted.
Based on this, it looks like Sunday at 2AM is a great time to submit a post if you want to maximize for upvotes relative to the competitive set. However, keep in mind that your competition at that time is also benefiting from the attractive ratio of votes to submissions.
If you want to also optimize for the amount of potential exposure you’ll get if your post is successful, you may want to also look back at the chart of points earned at each hour (since this is a likely proxy for traffic). When you factor that in, Wednesday at 8AM seems like a great all-around time to submit.
Interestingly, the 7AM – 9AM Eastern timeframe each morning looks extremely attractive. At 9AM, however, the ratio of points to submissions consistently plummets. This might mean that HN readers on the East Coast are at their desks reading and voting long before 9AM, but most submitters are holding out until that magical 9AM hour to make a run at the front page.
What to Post
Time of day is nice to optimize around, but ultimately content is king. If you have a killer story, it will dominate the front page no matter when you submit it. So, what moves the needle for the Hacker News community?
In my previous post, I looked at some categories of content (based on words that appeared in submission titles) and found that things like profanity and hot startups were far more likely to get upvotes than stories with sensationalist buzzwords and references to big companies.
The efficacy of the TSA’s body scanners has been a hot topic this year that has frequently raked in the points. Not surprisingly, posts about Y Combinator start-ups also rule the roost (acronyms like “S12″ indicate the class season and year of a Y Combinator startup). Hot companies like SpaceX and Stripe have also been performing extremely well.
Since Hacker News is largely a community of developers, I was curious to explore the average scores of the most frequently mentioned programming languages. (Some of these are frameworks and development-related technologies, but you get the idea). All of the terms shown here had appeared in at least 80 submissions.
Finally, I did the same thing with the names of companies, big and small. Average scores for the most frequently mentioned companies are below. (All of these appeared over 100 times.)
Just to reiterate, if your content is bad there is nothing you can do to make it popular on a social news site like Hacker News. However, it helps tremendously to know your audience and work with topics that have a track record of success. Picking a good time slot for edging out the competition doesn’t hurt, either.
I hope some of you found this helpful. On a related note, keep an eye out for my upcoming post on how SpaceX acquired a W12 startup that uses Lisp to smuggle Tesla Roadsters past TSA scanners. I’ll be posting it around 8AM Eastern next Wednesday.
This analysis was powered by RJMetrics. If you’d like to see what RJMetrics can do for your business, sign up for a free 30 day trial.