I recently meta'd some interesting research I'd done relating to the correlation (or lack of) between the time a question is asked and its view count. I got some comments suggesting I look into other interesting things, so I dusted off my text editor (lots of dust builds up in 23 hours, I guess) and did some analysis.

I first caused my computer to hate me eternally by repeatedly analyzing 30.7MB of raw post contents, parsing the HTML to get the plain text version. I created this graph (badly labeled and with a log10 scale for both axes), showing the relationship between length in characters (X) and view count (Y):

a graph showing a slight increase in post view count with longer body

The red line is the line of best fit, which looks a little odd on a graph with both axes logarithmically scaled (not sure if I'm even allowed to do that). The estimate it gives is that a post will get 6.3 more views for each 100 characters added to the body. You'd get better results just by posting 15 minutes earlier, according to my last graph.

With all of this nice data available, I decided to do something more interesting: look at the relationship between view count and votes. With a log10 X axis and a log5 Y axis (I don't think I'm supposed to do that, but oh well), here's a histogram (I think that's what it's called) sort of graph:

a graph showing a fairly predictable trend for views/votes

There's very little to no data for smaller numbers of views, and also very little for very large ones. It's a fairly predictable trend, with vote score (y axis) pretty much proportional (or exponential?) to view count (x axis). Feel free to comment about any issues with my math or phrasing, as the extent of my knowledge as of now is Algebra II and some googling.

I also did a linear regression on the views/votes, and got a surprisingly high (100s of times higher than any others so far in these graphs) R squared value of 0.52. It seems that about 1 out of every 500 views equals an upvote. And to anyone who's wondering, the best views-to-score ratio is 10.5:1, on this question, which currently has 14 upvotes and 147 views.

Use this information however you please, if you ever find a reason :)

  • \$\begingroup\$ The number of characters in a post isn't a good measure of post length, at least I think. Actual height of the post (incl. images and code blocks) vs. views might be more interesting, if possible at all. Or number of images or number of code blocks. Nice post btw. \$\endgroup\$
    – Bubbler
    Commented Nov 11, 2019 at 7:48
  • \$\begingroup\$ @Bubbler Measuring the height might be difficult, since I can't actually load a lot of the images for CORS reasons \$\endgroup\$ Commented Nov 11, 2019 at 13:00
  • 7
    \$\begingroup\$ I'm pretty sure there's a special hell, or at least an XKCD strip, for people who don't label their axes :P \$\endgroup\$ Commented Nov 11, 2019 at 17:57
  • 3
    \$\begingroup\$ I don't think there's anything wrong with transforming by different logs; the logged scales will just be different by a constant factor of \$\log_5(10)\$. \$\endgroup\$
    – Giuseppe
    Commented Nov 11, 2019 at 18:27
  • 1
    \$\begingroup\$ @DewiMorgan It would have taken a few extra minutes to label them with my JS canvas. I'm evil, I know. \$\endgroup\$ Commented Nov 11, 2019 at 20:08
  • 4
    \$\begingroup\$ Forgiven, because it was super interesting :). Despite xkcd.com/833. \$\endgroup\$ Commented Nov 11, 2019 at 20:42


You must log in to answer this question.

Browse other questions tagged .