I looked at the data from 1.3 million Hacker News stories and found that when a story gets submitted makes a big difference (up to 172% better chance of getting on the front page). This article shows the analysis and results.
I used the official Hacker News API to get the stories using Python, and used R and ggplot2 to do the exploratory data analysis and plots.
> I’ve commited a grave statistical faux-pas in this article. Did you spot it? I showed that the number of votes a Hacker News story gets is correlated with the time it is submitted. Then, I told you that submitting your story at a particular time will cause it to have a higher chance of success. Correlation does not imply causation, however, so this is unproven. It is entirely possible that stories submitted on the weekend have simply been better than those submitted on weekdays, or that weekend readers are “vote-happy”. However, my intuition suggests to me that it is most likely that there is true causation here.
You probably should have led with that caveat. Ignoring it completely for the sake of an argument is misleading.
Relatedly, "what if I told you I found an easy way to increase your story’s votes by 172%?" is flat-out wrong. That metric implies an average behavior, which is not the case here.
A neural network simulation in a browser is a great HN submission. You were right that it fell through the cracks undeservedly, and finding such stories and giving them a second chance is the purpose of our experiment in reposting.
Looking through HN data to analyze posting times? not so much. Meta posts are a low-quality genre to begin with, plus analyzing posting times to figure out how to time HN is something of a cliché, and even if the findings are meaningful, it's hard to see how HN improves as a result of them. Also, we'd not likely invite a repost when the original had a misleading, linkbait title like "How I hacked Hacker News".
11 comments
[ 2.6 ms ] story [ 35.7 ms ] threadI used the official Hacker News API to get the stories using Python, and used R and ggplot2 to do the exploratory data analysis and plots.
You probably should have led with that caveat. Ignoring it completely for the sake of an argument is misleading.
Relatedly, "what if I told you I found an easy way to increase your story’s votes by 172%?" is flat-out wrong. That metric implies an average behavior, which is not the case here.
Your related point is a good catch. That was a brain slip on my part. Thanks for pointing it out!
If you found the solution when/if a story gets picked up by time-of-day then why submit it again? (I'm making an ironic joke)
A neural network simulation in a browser is a great HN submission. You were right that it fell through the cracks undeservedly, and finding such stories and giving them a second chance is the purpose of our experiment in reposting.
Looking through HN data to analyze posting times? not so much. Meta posts are a low-quality genre to begin with, plus analyzing posting times to figure out how to time HN is something of a cliché, and even if the findings are meaningful, it's hard to see how HN improves as a result of them. Also, we'd not likely invite a repost when the original had a misleading, linkbait title like "How I hacked Hacker News".
Please do repost the neural network though!