LessWrong analytics (February 2009 to January 2017)

Table of con­tents


In Jan­uary 2017, Vipul Naik ob­tained Google An­a­lyt­ics daily ses­sions and pageviews data for LessWrong from Kaj So­tala. Vipul asked me to write a short post giv­ing an overview of the data, so here it is.

This post cov­ers just the ba­sics. Vipul and I are ea­ger to hear thoughts on what sort of deeper anal­y­sis peo­ple are in­ter­ested in; we may in­cor­po­rate these ideas in fu­ture posts.

Pageviews and ses­sions

The data for both ses­sions and pageviews span from Fe­bru­ary 26, 2009 to Jan­uary 3, 2017. LessWrong seems to have launched in Fe­bru­ary 2009, so this is close to the full du­ra­tion for which LessWrong has ex­isted.

Pageviews plot:

30-day rolling sum of Pageviews

To­tal pageviews recorded by Google An­a­lyt­ics for this pe­riod is 52.2 mil­lion.

Ses­sions plot:

30-day rolling sum of Sessions

To­tal ses­sions recorded by Google An­a­lyt­ics for this pe­riod is 19.7 mil­lion.

Both plots end with an up­ward swing, co­in­cid­ing with the effort to re­vive LessWrong that be­gan in late Novem­ber 2016. How­ever, as of early Jan­uary 2017 (the lat­est pe­riod for which we have data) the scale of any re­cent in­crease in LessWrong us­age is small in the con­text of the gen­eral de­cline start­ing in early 2012.

Top posts

The top 20 posts of all time (by to­tal pageviews), with pageviews and unique pageviews rounded to the near­est thou­sand, are as fol­lows:

Ti­tle Pageviews (thou­sands) Unique Pageviews (thou­sands)
Don’t Get Offended 681 128
How to Be Happy 551 482
How to Beat Pro­cras­ti­na­tion 378 342
The Best Text­books on Every Sub­ject 266 233
Do you have High-Func­tion­ing Asperger’s Syn­drome? 188 168
Su­per­hero Bias 169 154
The Quan­tum Physics Se­quence 157 130
Bayesian Judo 140 126
An Alien God 125 113
An In­tu­itive Ex­pla­na­tion of Quan­tum Me­chan­ics 123 106
Three Wor­lds Col­lide (0/​8) 121 93
Bayes’ The­o­rem Illus­trated (My Way) 121 112
926 is Petrov Day 121 115
The Baby-Eat­ing Aliens (1/​8) 109 98
The non­cen­tral fal­lacy—the worst ar­gu­ment in the world? 107 99
Ad­vanced Place­ment exam cut­offs and su­perfi­cial knowl­edge over deep knowl­edge 107 94
Guess­ing the Teacher’s Pass­word 102 96
The Fun The­ory Se­quence 102 90
Op­ti­mal Em­ploy­ment 102 97
Ugh fields 95 86

Note that Google An­a­lyt­ics re­ports are sub­ject to sam­pling when the num­ber of ses­sions is large (as it is here) so the in­put num­bers are not ex­act. More de­tails can be found in a post at Lu­naMet­rics. This doesn’t af­fect the es­ti­mates for the top posts, but those wish­ing to work with the ex­ported data should be aware of this.

Each post on LessWrong can have nu­mer­ous URLs. In the case of posts that were re­named, a sig­nifi­cant num­ber of pageviews could be recorded at both the old and new URL. To take an ex­am­ple, the fol­low­ing URLs all point to luke­prog’s post “How to Be Happy”:

All that mat­ters for iden­ti­fy­ing this par­tic­u­lar post is that we have the sub­string “/​lw/​4su” in the URL. In the above table, I have grouped the URLs by this iden­ti­fy­ing sub­string and summed to get the pageview counts.

In ad­di­tion, each post has two “canon­i­cal” URLs that can be ob­tained by click­ing on the post ti­tles: one that be­gins with ei­ther “/​r/​less­wrong/​lw” or “/​r/​dis­cus­sion/​lw” and one that be­gins with just “/​lw”. I have used the lat­ter in link­ing to the posts from my table.

Source code

The data, source code used to gen­er­ate the plots, as well as the Mark­down source of this post are available in a GitHub Gist.

Clone the Git repos­i­tory with:

git clone https://​​gist.github.com/​​cbdd400180417c689b2befbfbe2158fc.git

Fur­ther read­ing

Here are a few re­lated Pre­dic­tionBook pre­dic­tions:


Thanks to Kaj for pro­vid­ing the data used in this post. Thanks to Vipul for ask­ing around for the data, for the idea of this post, and for spon­sor­ing my work on this post.