Accounting for Game Size

Game Size Analysis:

When accounting for game size (the quantity of players per game), we find an interesting difference from the overall picture. Small games tend to have more concentration among the top 1-20% than Big and Medium games. This would imply for smaller games that more of the game is experienced by the 20% player bracket than for larger games.

What’s interesting, however, is that the top 0.1% of players for small games has less share than their counterparts. Yet this is most likely due to the Top 0.1% for games with less than 1,000 players being 1 person. Due to this, I wouldn't read too much into the small-game 0.1% figure.

Game Size deep dive:

Small games again tend to have the most variance in Share as the 68% and 95% sample ranges show the most variation. This again could be due to the low sample of players, but it could also indicate that small games can have very concentrated communities or can be very egalitarian.

What Does This Mean?

A Case study with Portal 2

A great example of an egalitarian game is Portal 2. 99% of accounts who have played it have less than 20 hours in the game, implying that they beat the campaign once or twice and never played again.

In other words, 99% of people that played Portal 2 experienced the same content, implying that the Dev’s greatest impact on the quality (as measured poorly by reviews) of the game was in those first 20 hours of content.

Implications for Game Development:

Spend your effort on the parts that people are going to see and experience. If 99% of people experience something great, then that will produce a far better effect for your game than if you spent your time on something that only 1% of people will experience.

Individual Games

Explore more player data: View the full game analysis

Methodology

How Playtime Concentration Is Calculated

Say we have a Game Y with a distribution that looks like this:

Player 1-100: 2 Hours
Player 101: 30 Hours
Player 102: 35 Hours
Player 103: 100 Hours
Player 104: 1000 Hours

Total Playtime = 1365 Hours
Top 1% of the Game’s Players by Playtime (Player 104) = 73.2% of the Game’s Total Playtime
Top 4% (Player 104+103+102+101) = 85.3% of the Game’s Total Playtime

Then for Game Y, we can say that the Top 1% of players hold 73.2% of the Total Playtime, with the Top 4% holding 85.3%. Repeat this for as many Top X% brackets as you want and you can eventually create the Scatter Plot shown in the first picture.

I made up these numbers but, roughly speaking, that's what a raw distribution in absolute hours looks like for any game on Steam.

Definitions

Percentile: The Top X% of Each game’s playerbase when filtering by Playtime. Also referred to as the Bracket.

Average Share: For the number of games analyzed, the bracket has on average this amount of share of total playtime.

68% Sample Range: The range of values of the Share of playtime for 68% of the number of games in the study.

95% Sample Range: The same as the 68% sample range, but for 95% of the number of games in the study. For you stats nerds out there, this is the 1 STDev and 2 STDev ranges from the mean (Average Share).

Number of Games: The quantity of games that are included in the study. When breaking down by game size (number of players), the Number of Games is simply the amount of the overall study that fit the specification:

Big Games: 100,000 (10^5) Players and Up

Medium Games: (10^3) 1,001 - 99,999 (10^5) Players

Small Games: (10^3) 1000 Players and down

Players: The quantity of accounts that have played a game, measured as having more than 0.01 Hours of playtime in a game. All players are owners, but not all owners are players.

Owners: The quantity of accounts that own a game (whether they have paid or not). If it is in your library, you own it, regardless of how you got it. Free Weekends are inclusive in this statistic.

Error and Bias

Observant readers will notice that the Quantity of games in the study is 23,778, whereas Steam has roughly 275,000 games, meaning that my sample is less than 10% of the total quantity of games on Steam. This is because I measured playtime, which is only directly visible from public player accounts. The problem with using this method is that for any game with less than 100 players in the overall account sample it becomes impossible to extract any useful information and at that point you are extrapolating what might not actually exist.

My account sample at the moment stands at 5.8 Million accounts, of which 2.3M are public, meaning I have access to the playtime. Calculating the total playtime of Steam itself through the sum of all CCU-Hours leads me to an estimate that I have roughly 2% of the total playtime of Steam in my 2.3M sample.

As an example of a very small game which has less than 1000 owners total, in my sample I would be lucky to have 10 people who own it, fewer having played it, thus I wouldn’t be able to make conclusions on the behavior of the player.

As another source of Bias, I displayed the size of the game as the quantity of sample players. The two are certainly correlated, but they are not the same. Yet for the purpose of showing the relationship between relative size and playtime concentration I think it's acceptable for the lack of precision.

Summing up the error and bias, I would say to not treat this study as definitive of Steam as a whole, but a fairly decent representation of the playtime concentration of the top 10% of games on Steam by player count.

What Could Project Mimir Investigate Next?

1. The numbers for game size are in terms of Players, not owners. For games with a large percentage of sitting unplayed in people’s libraries (Activation rate), this makes for an interesting difference, as it concentrates the playtime among a smaller share of players.

2. These are for all genres and tags across Steam. It would be interesting to see if the Pareto Principle is broken for different genres. This post was just to establish a baseline for what Steam as a whole looks like, but it could be that MMO’s tend to be more casual and more egalitarian (less concentrated among the top 20%), whereas FPS games have more hardcore audiences and less egalitarian (more concentrated among the top 20%) as an example.