Does Tyler Cowen Sleep?

Marginal Revolution by the Numbers

I was curious if I could measure the impact of daylight savings on Tyler Cowen’s posting habits on his world renowned blog, Marginal Revolution (MR). So, I threw together some python code, collected every post that has ever been made on MR, and started looking at the data. After doing a little digging, and seeing how often Cowen posts, I realized that a better question might be “Does Tyler Cowen sleep”?

Cowen, posting on MR since August 2003, has accumulated over 28,000 unique posts, more than 5 million words, and averages ~185 words per post. Alex Tabarrok is the next most regular author with ~4,400 posts, 1.1 million words, and ~260 words per post. Readers of MR might be surprised to know that in the early days there were other authors besides Cowen and Tabarrok. In total, 24 people have contributed at one time or another. However, the most recent post not by Tyler Cowen or Alex Tabarrok was ~7 years ago by E. Glen Weyl.

Overall, MR posts peaked in 2011 with 2,071 posts but the blog has consistently had 1,500+ posts every year since 2007. If you do some simple math, for 2021, the blog had an average of 4.77 posts every day, or 1 post ever 5 hours. Cowen by himself averaged 3.93 posts per day or 1 post every 6 hours! So…when is he blogging? And, how does he find the time to do that AND engage in public speaking events, host a podcast, write books, do research, teach, etc.? Perhaps only Tyler and God know the answer to that… but, let’s break it down by hour!

Cowen’s favorite time to post has shifted over the years. His main hours appear to be early morning, right around noon, and the hours following midnight (these might be prescheduled posts). The 7 hours from 5PM to 12AM sweep the bottom seven spots. Cowen is either an evening sleeper… or this is when he does other stuff besides write. There appears to be a large shift is behavior beginning in the end of 2014 towards posting in the very early hours of the day. The change seems too drastic to be a simple behavioral change but I could be wrong. Some possibilities are that Cowen

• now preschedules when posts will go live

• moved and became part of a new timezone… yeah I know that is a stretch

• hired someone to edit and post for him

• changed online habits due to other obligations or priorities

• what am I missing? Tell us Tyler Cowen!!

Notice the big shift in posting behavior in late 2014

The change becomes even more apparent when you just graph the first post of the day (see above graph). What used to be a 7AM post suddenly transitioned to a 1AM post in 2015 and doesn’t look to have wavered since then. Speaking of not wavering, the most recent date that Cowen went more than 24 hours between posts was on May 25th of 2009. Talk about consistency!

Red = more than 24 hours between posts

Assorted Links

If you are anything like me one thing that keeps you coming back to Marginal Revolution are the assorted links. Multiple times a week Cowen will post a list of the most interesting articles that he has come across, or that readers have sent him during the week. So, where is TC getting his news? New York Times comfortably comes in first with ~2,398 links, then Twitter with ~1,835, and in third, The Washington Post with ~1,120 links.

Books

And, lest you think Cowen is only reading the news…Tyler also regularly posts a list of “What I’ve been reading”. Looking only at the posts with “What I’ve been reading”, or variations of it, for the title, I aggregated the total number of books Tyler has listed. These total to about ~1,400 books. Assuming that he doesn’t post 20% of the books that he reads, Cowen is reading a book, and meaty books at that, every 3-5 days. So… if you want to be the author of one of the most famous economics blog in the world, READ!

Daylight Savings. Does it have an impact?

I thought I should at least finish what I set out to answer: Does daylight savings impact Cowen’s posting behavior? At this point you might be starting to think that Cowen is more of a machine than a human. Does he sleep in when we lose an hour in the spring? Does he get up a little earlier in the fall when we gain an hour back? Truth is, the data isn’t really clear on that point. After doing about as much P-hacking as I could stomach, I landed on the below model. I limited the data to the first post of the day after 5AM. My rational was that I really only wanted to look at the first post of the day and anything before 5AM felt like it could be part of the previous night (truth is I played around with a 4AM cut-off but the data looked cleaner at 5AM. Plus, who gets up at 4AM?) The results were directionally what I would expect, about 25 minutes later in the spring and 25 minutes earlier in the fall. However, both of these had large standard errors and changed drastically as I looked at different data sets or filters.

Maybe he is human after all? What does appear to be certain is that Cowen sleeps in a little longer on weekends (or at least doesn’t post as early). If you look at the below table, you’ll see that compared to Sunday, the holdout day, the weekday first posts are significantly earlier. Every weekday is around 30 minutes earlier with Friday being upwards of 50 minutes earlier than Sundays. Maybe weekends are when he gets up and runs a marathon before blogging?!

Final answer: Cowen is likely impacted by daylight savings. Maybe.

P.S.

Out of curiosity I also used a non-parametric model to predict daylight savings impact. Mean absolute error (MAE) for the linear model was 123 minutes. Using a tuned XGBoost model I was able to get the MAE down to 114 minutes. Using this more accurate model to estimate the counterfactual of what would have happened if it were not daylight savings, the model estimated that on average spring daylight savings pushed the average first post time back by -23 minutes. Fall daylight savings, on the other hand, only increased the time of the first post by +10 minutes. In this case I’d probably trust the non-parametric model more than the linear model but it is nice to be able to see under the hood a little with the OLS approach.