Game Time Distribution

I read an old paper the other day.

The paper tries to fit a distribution on session time of online video games. In particular, they measured game time for CounterStrike. They checked 16000 user sessions, spread over 300 maps. It's from 2002, which is pretty dated, so the findings may no longer hold. But it had some fun charts.

It seems like most sessions took up less than 15 minutes, but there seems to be a somewhat long tail to this distribution. They tried fitting a few distributions on it, and it turns out that the Weibull distribution had the best fit.

In particular:

As Figure 2 shows, a Weibull distribution with β = 0.5, η = 20, and γ = 0 closely fits the PDF of measured session times for the trace.

The paper also mentions that this is in contrast to another paper that measures HalfLife session time which supposedly follows an exponential distribution.

Reading the conclusion of the paper, I wonder how much has changed since 2002.

Unlike many source models used to predict Internet traffic, player session “ON” times follow a distribution that is not heavy-tailed. Session times, in fact, follow a simple Weibull distribution, a distribution that is commonly used to model lifetime distributions for reliability analysis. This work, along with previous studies showing a constant resource demand for active clients, allows for the construction of a partial source model for FPS-based on-line games.

I can imagine that it's much more normal to play video games these days. And I certainly don't think it's unheard of to play for four hours straight. Especially on a weekend. Just checking this World of Warcraft forum on the topic yields this as a top response:

In the old days I'd pull an eight hour Saturday.

But now? I average 2 or 3 like you. I've got kids and homeschooling and work. So I play at night after everyone's in bed. Occasionally I'll pull 5 hours on a Saturday night and immediately regret it the next day.

Reading it, I'm kind of happy that I've stopped playing video games.