For all it’s worth, though, I’m making the data I’ve gathered for the past 4 years or so available for anyone. They’re dumps of the complete tables, in CSV format. You can open it on any tool that supports CSV files, such as Microsoft Excel or OpenOffice Calc. You will also need 7-Zip to uncompress the .7z files.

This data is not for the faint of heart. Do not expect to open this on a spreadsheet and make a beautiful chart, it won’t work. For one thing, because the data is pretty massive; but also because there’s no time consistency to the data rows, since I had to resort to occasionally averaging data to make the table smaller. Read below for more information.

• games: the list of games as they appear on Gamespy’s site, plus their id (the same as gamespy’s list id), its developer name, and its short title (used on the mods graph). All games found on GameSpy’s top 40 list were automatically added to this list (and subsequentially to the stats lists). If a game had the read_mods field on, it was also read from GameSpy’s website even if it didn’t make to the top 40 games.
• games_stats: the game stats. Contains number of players and servers. Fields year/month/day/hour contain the time that data was gathered; however, if, when adding the last hour (23) it already has 23 other stats for that game on that same year/month/day (that is, the game had statistics for all hours of a given day), it would average them all (set the statistics as “day”) and set the hour as “0″. It’s a bit dubious (hour “0″ can either mean the averaged stats, or stats for hour 0), but it worked. In theory, if the data was 100% consistent, you’d never find any value in the “hour” field, only 0s, since it’d all be averaged eventually. But since it does not average the data when less than 24 samples are available for that day (like, when the server was down, or when gamespy was down, or when the script timed out before adding) it would sometimes leave the data alone. They’re still averaged by the graph generation script; this is mainly a database thing, meant to make it smaller and faster to read.
• mods: the list of mods, similar to the games one. Their id is internal. Also added automatically as new mods were found.
• mods_stats: same as games_stats, but for mods.

On the CSV file, field values are separated by semicolons and enclosed by quotes. The first line contains the field names.

Problematic ranges

Due to the problems that occured a few times when reading data, some of the data is ignored when making the graph, and is instead simulated: when creating the graphs, the graph-making script simply ignore these days and interpolates values from the nearest available data instead (previously it’d simply plot the data as available, making it look weird, but I changed it later and made it bypass these days). The list of “problematic days” (hard-coded into the graphic plotting script) are as follow:

• 20/june/2006 to 28/june/2006 (all data): on these days, the script didn’t really work because (apparently) it was blocked by the gamespy server. Data for these days isn’t available.
• 12/october/2006 to 13/october/2006 (mods): I introduced a feature that allowed the mods of the 20 top games to be read automatically (previously it would only read the mods of a few selected games). However, this created a bug in which the games that were set to always have its mod data read (ie, HL1, HL2) would simply ignore the mod data. That is to say, the mod data on these couple of days is lost for many games, so the data is instead simulated for all mods.
• 6/february/2007 to 13/february/2007 (all data): GameSpy had some kind of crash on their stats and stopped working on these days, giving either cached results (on the games pages) or no result at all (on the mods pages). Data for these days are more or less available but are corrupted and should not be considered.
• 7/april/2007 to 11/april/2007 (all data): GameSpy went crazy and a few games where getting ignored (including Half-life 1 and 2, Enemy Territory, Call of Duty 1 and 2, Quake 3, and others), listing little to no players and servers.
• 13/november/2007 to 14/april/2007 (all data): The update script was broken for a few hours (server was down?) so data isn’t available.
• 28/june/2008 to 4/july/2008 (Unreal Tournament): there was some kind of problem reading data for those days (it generated some super high numbers), so the game and mod data was deleted and interpolated from the known numbers.
• 29/june/2008 to 2/july/2008 (Unreal Tournament 2004): there was some kind of problem reading data for those days (it generated some super high numbers), so the game and mod data was deleted and interpolated from the known numbers.

Remember that this data belongs to GameSpy and they must be credited wherever needed. I merely store the statistics found on their page.

I started gathering the data from GameSpy on 14/12/2004 (for reference, that was shortly after CS:S release, if I remember correctly). The graphs on the site only show the last 365 days.

Like mentioned above, the data I’ve gathered is not 100% consistent, but it’s pretty close. I still stand by the data I gathered; while there’s no website with 100% accurate stats out there, it is my belief that GameSpy has been the most consistent one on average. There’s a lot criticism over it, as it should be (but also a lot of it unfounded or just plain false), but if one looks at the alternatives the critics point to, they leave a lot more to be desired and perform a lot worse on several other games, or just use a very different methodology; for example, it was common for people to point to Steam stats when stating that GameSpy numbers were wrong, but they failed to notice Steam stats also used local/singleplayer numbers on their counts, while GameSpy stats uses information from publicly known servers only. That is all to say, GameSpy stats was not perfect but it was the better of the bunch given my objectives.

It’s not my intention to continue a discussion that will never have an end, though. To be quite honest, this has been a great experiment for a few years. I certainly liked seeing how the audience for each game fluctuated over the years, and when new games were released. It’s too bad it has to close down; here’s to hope we’ll see something better or more accurate in the future.

Contrary to popular belief, this is not a sign of the apocalypse, that the website or its database has been fucked up beyond recognition, or (and this is my favorite) that this is some kind of conspiracy to make a game “popular”/”unpopular” artificially. Instead, this behavior is due to the incorrect reading of (more or less) one day’s worth of data. It’s not an ongoing error; it’s something that happened once and now is already back to normal.

The problem is, with the way I render the graphics - doing an evened-out 7-day average plot, to dilute the effect of weekends - having invalid data for a single date is enough to make the plot lines pretty weird for almost half a month. So that’s what you’re seeing right now; given time, things will look like they looked before, with game averages back to their original positions.

It’s important to notice, though, that this problem has happened before; sometimes GameSpy’s website is inaccessible, or my own website is offline, or the data is corrupted. This causes data to be invalid. When that happens, the solution is easy: invalidate the data read on those specific days, and then have the graphics be generated again. When generating the lines, invalid dates are actually ignored and the data they should have are interpolated from the two nearest valid dates. This is not accurate - of course, it’s just a mathematical guess - but it looks good enough.

For that, I have to go to the database and find the offending data and flag it down. This is not that complex - however, in this case, the problem is that I haven’t had the time to go inside the MySQL dungeons to find and fix the data as it’s some boring work that requires concentration and patience. Truth is, I’ve been really busy for the fast few months due to college and work (there’s a bit more information here) so I had to neglect a bunch of updates and maintenance to this website.

This doesn’t mean it’s going down the gutter. The data is still being read the way it should. That is the most important thing. I just have to go and fix the data later; this will “fix” the graphics, including for the past few days where it shows invalid lines.

Also, some of the graphics are also not being generated anymore; specially the graphics for mods. This is due to the way cronjobs are being ran now in my hosting provider (DreamHost). It’s a lot better and easier to maintain, but I still have to finish setting up the new cronjobs under the new system.

Really, this is just a quick, honest, unreviewed post to let people know what’s up: graphics are weird, but data is being read without a problem (save for the occasional issue that has to be fixed later, which is to be expected). I’ll fix stuff as soon as possible; they’ll be retroactive.

Update: at long last, I’ve fixed the erratic data which created those spikes on the graphics. It turns out it had some crazy numbers for Unreal Tournament and Unreal Tournament 2004 - in the millions, while both games ever go from more than a couple thousand players. Since two different games done by the same developer (and probably using nearly the same query protocol) were at fault, this was probably due to some problem in GameSpy’s server data reading.

In detail, the offending problems were as such: between june 28 and and july 4 of 2008, the mod “Tournament Deathmatch” of Unreal Tournament had those huge numbers registered. This contributed to the final (game) numbers being that high. At the same time, the mod “Frenzy” for Unreal Tournament 2004 - which never even shows on the stats! - had huge numbers between june 29 and june 30; the mod “xDeathMatch” had its numbers jinked on the same dates; and the mod “xBombingRun” had its numbers go nuts on july 1 and july 2.

I have deleted those entries and filled the gap with interpolated data from valid days, which seems to be pretty accurate since both games since to have a small but very stable following. Graphics will self-update later tonight.

Just to make it clear, the game is indeed doing very well online. If you check the ServerSpy or Game-monitor statistics, you’ll see it’s doing well enough to put it quite strongly at the spot of 3rd most played online PC FPS.

The reason it’s not showing on this website’s graphics is because GameSpy stats - which is the source I use for my data - isn’t covering it yet. They probably still have to code the server query logic into their system, and hopefully they will have it soon. So unfortunately the post-release data is lost, but it’s a sure bet the game is having a big impact.

And before someone points how I should switch to using some other data source instead, let me remind you that none of these FPS statistics out there is perfect. Each have its own little perks and little issues with specific games. It just so happens now that GameSpy had a major issue with CoD 4, but the sanest thing to do is to expect it to recover soon. Just switching the data source to another website would invalidate any possible comparison to the data already gathered, requiring a database reset, so it’s something I’d like to avoid.

Update: it’s working now, as CoD4 is being listed on GameSpy stats. Check the comments for more information.

The first thing noticeable, somewhat curiously, is that the game didn’t manage to attract any noticeable slice of the current Wolfenstein:Enemy Territory audience - and that’s the game that, gameplay and intellectual-property wise, could be considered its only prequel, despite a different setting and argument. Apparently, the hardcore following of a free game didn’t want to experiment a slightly different - and commercial - game.

Also noticeable is the apparently immediate surge on the game’s popularity from October to November. What happened there?

This is an topic that’s bound to cause some controversy, so before I get to that, here are some numbers: right now, over at GameSpy stats, ET:QW has 5167 online players, distributed over 891 servers. Over at Game-monitor, it shows 1059 players, distributed over 1157 servers. ServerSpy.net shows 2815 players, over 871 servers.

The surge in player numbers also coincide with the release of a patch that takes the game to version 1.2, so this is probably the indirect reason for the surge.

The direct reasons is either that the developers have fixed some problem with server querying that existed before, or that they’ve added the bot count to general player count. Unfortunately, the latter seems to be the case, specially because other statistics seem to properly filter out those players, even if the number of servers they cover is similar. This would be a huge no-no - it would inflate the number of players unrealistically.

Additional note: while there have been rumors other games - specially Battlefield 2 - have been also doing it, it’s important to remember Battlefield 2 didn’t even allow bots on public multiplayer games when it was released, and even when it started doing so, there was no change on the online numbers.

On the other hand, other statistic sources might be completely filtering out servers that simply allows for bot play, removing legitimate players from the total count. I wish I knew more about the issue. Either way, I think it’s fair to believe the real ET:QW numbers are much lower than what the graph shows, so keep this in mind.

Despite all that, how well did the game do online, compared to its direct competitors?

While it failed to attract Wolfenstein:Enemy Territory players, it apparently managed to grab a very small hold of Battlefield 2 and Battlefield 2142 players. Might might or might not be a good thing, given the fact that the typical Battlefield player is used to a gameplay that is not really similar to what ET:QW has to offer, despite some visual similarities (both features “vehicles” that travel over “large terrain”) and hyping a player to a game that offers a gameplay so different from what he expected is bound to cause online headaches.

No matter whether the current player count is inflated by bots or not, there’s also a visible downwards trend on the game’s popularity on the graph. It has also failed to take advantage of the Christmas effect - differently from what Battlefield 2142 did last year. The game will probably stabilize in the future as it grabs a niche of its own hardcore players, but right now, the future is not so bright for the game (again).

See more complete graphs on how well the game is doing online on the games page.

This move was needed because many full-fledged games based on the HL2 engine - games such as TF2, Counter-Strike:Source, Day of Defeat:Source, The Ship, Dark Messiah of Might & Magic, among others - are listed as HL2 mods. Technically, they might be mods after all, but since they’re sold as separate titles - with different SKUs and all - it makes sense to have them listed separately.

However, I went ahead and took the separation a bit further. Some original HL1 mods - Counter-Strike, Day of Defeat and Team Fortress Classic - are also separate now. This one is a bit controversial, as they are HL1 mods, and one can, technically, get them installed without the need to buy anything other than the original HL1 game. However, they’re also sold separately and generally considered to be very different games with separate communities and all. They also take a huge chunk of the online player base. So, as such, they are now treated as full games.

Due to this change, it’s now finally possible to see what kind of impact a few recent releases had on the online player base, specially comparing TF2 to the rest of the games. For example, it’s possible to see TF2 managed to get to the 4th place in online popularity quite fast, but now has dropped to 6th. I’ll have some more in-depth posts about this game and others in the future, as soon as December is over.

Finally, I will probably tweak these graphics - and what should and shouldn’t be considered a full game - a bit further in the future. The data remains the same - all mod-to-game “upgrading” is done when the graphics are rendered, not when the data is read - so I can always go back and undo anything.

The HL2 and HL1 mods are still listed on their original game pages, but that will probably change soon too.

As a follow-up, other interesting discussions about the statistics are taking place over at Rock, Paper, Shotgun and at ShackNews.

The big news is that this version of the survey also detects whether the user has some specific applications installed - on my case, it successfully detected FireFox, OpenOffice, and Zone Alarm. Data from those specific results are not listed on the survey results page yet, though. Personally, being both a FireFox and an OpenOffice zealot, I was pretty happy to see something like this popping up there.

Finally, Tom at The Steam Review has some additional discussion about it.

Update: awesomely-named RockPaperShotgun.com also has an analysis of some of the preliminary results of the new survey.

Lastly, just as a quick note: in the future I’ll arbitrarily separate some specific mod data from their specific game data, so games like TF2 and CS:S won’t show up as being mods of HL2 (they’re sold as separate SKUs after all). I’ll also have to remake the graphic generating algorithm to work around the server script execution time limit - the reason why the mods page isn’t getting updated. No data is lost, though, as gathering them is a separate process that’s still doing fine.