Population & patches


In the weeks and months after Incarna was released, there was a great amount of debate about what effect the expansion had had or would have on the population and subscription numbers (the latter being a bit more difficult, but as the QENs and previous populations numbers had shown, online averages and subscriber numbers correlate quite nicely). The only useful source for this kind of information was (and still is) Chribba's population stats site EVE Offline, where he continuously collects data about the EVE servers through the EVE API. The problem is that, while very comprehensive, the data on EVE Offline is rather noisy largely because it is so comprehensive.

In an attempt to fix this, I started to collect the data Chribba published and re-render it in a more cleaned up format. At first, it was done in order to get a bit more readable data to back up (or refute) the many claims made at the time, but over time, I became interested in the more general patterns of what happens to the population around patches and across seasons.

Chribba's data goes back to early 2006 (Red Moon Rising) and is quite neatly contained in the XML used to drive the graphs on his site. To extract this data, I quite crudely copy-pasted that XML and washed it through an even cruder brute-force XSLT stylesheet to get a CSV that could be manipulated in any spreadsheet of the user's liking. For my part, I then chose to filter this data to only show Sundays, since they are consistently the days when most people play (as we will see later). This choice not made to inflate the numbers, but to have a consistent point of comparison and reduce the noise of the data that occurs when different days of the week consistently have drastically different averages.

Enough talk, graphs!

The first two graphs shows the daily population averages from RMR and onwards. The following larger expansions and patches have been marked in the graphs:

The first graph simply shows the daily average of all the Sundays from the beginning of the data collection up until the most recent data point. The colouration of the main line also shows the season for the data, for anyone wondering how summers affect the populations (answer: in general. they go up, contrary to conventional wisdom). The graph also contains the rolling monthly, quarterly and half-yearly average (the w-average heading in the legend refers to a rolling weekly average that is hidden here because of the noise it generates). The second graph shows year-to-year changes in the Sunday averages: the average of one date compared to the average on the same date the year before. This graph highlights the continuous growth EVE experienced up until mid-2010 or so, after which it started to sag and then become rather erratic (especially once Incarna hit).

Both graphs have a couple of highlights of note. The first is the break in the graph that occurs in late spring 2009, when the API was acting up and didn't provide any (or only very occasional) data, which generated day averages of zero or only a few thousand people. In the year-to-year comparison, this also creates a second gap in 2010, because there is no useful data available for comparison with the numbers collected for that year's spring months. The second is the sharp decline that happened just after the API was fixed, in the summer of 2009. This was when the famous Unholy Rage campaign ran, resulting in several thousand bots being removed from the game. Since these bots were online 23/7, they added a significant number to the daily averages, which was lost as a direct result of the campaign.

Both of these graphs show quite clearly the decline that happened after Incursion 1.1 had been released and the hype was dying down, and what happened after the famously botched Incarna release: the numbers started to go down to a level that was comparable to two years earlier. When Incarna hit, instead of generating at least a peak of initial interest, the decline only increased further — the peak in population came before the patch and was a combination of pre-patch hype and Alliance Tournament 9 drawing a crowd, as the AT:s usually do. This was the point where I started to collect daily data…

While the previous two graphs showed daily average, this last graph shows the daily Peak Concurrent User (PCU) count, starting about a month before the release of Incarna and continuing to the present day. Again, the main patches released during this time-frame are marked in the graph, and the data is separated into one graph for each day in the week. This is once again done to reduce the noise in the graph.

First of all, this demonstrates quite nicely the weekly flow of players: Sundays are consistently the most populated days, with a build up and and a dip on Saturdays and Mondays that equate to roughly the same amount of people on both days. While that number does not show up in this particular graph, it's also interesting to note that the Saturday and Monday numbers are usually a decent indication of what the weekly average will be (this is perhaps a bit more clearly seen in the raw data graph shown further down). From Mondays, the daily numbers usually sink lower and lower until the least populated day of the week: Fridays. It is interesting to note, though, that in the last couple of weeks (as of the time of writing), this trend has actually been broken — Fridays are no longer necessarily the least-populated days, and instead, it's Thursdays that come in last. Since this data only covers just over half a year, we can only speculate why this is. Maybe it's a winter thing: people aren't as keen on going out for ye olde Friday fun when it's freezing outside, so an evening of light EVE will have to do instead?

Another thing to note is how clearly some larger events in and outside of the EVE universe affects the population count. The sharp dip after Incarna aside, there are a number of points that show up very clearly in this graph:

  • The Fourth of July Monday that created an extra peak of U.S. players the week after Incarna hit.
  • The (U.S.) Labour Day Monday, which happened just after Incarna 1.1 was released, again creating an odd peak.
  • The start of the Goonswarm ice interdiction campaign which happened at the end of September, driving the Saturday and Sunday numbers down sharply until people(?) could adjust their play style and feel safe online again.
  • The release of the AAA titles slotted for the holiday-season in early November — people leaving to play MW3 and Skyrim for a few weeks until they're finished and/or the Crucible patch draw them back in.
  • …and finally, obviously, the Christmas and new-year's week-ends, when a lot of people obviously have better things to do than play spreadsheets online.

This graph also highlights the short-term effect of the Crucible patch: when the news of the patch started to come out in early October, coinciding with the Goonswarm ice campaign, there was an almost immediate response from the players. The numbers shot up instantly, and (slowly) started to reverse the downward trend that Incarna had caused (or at least failed to stop). Granted, this slow incline was abruptly halted by the AAA title releases, but the bump is there none the less. The effect of the expansion itself is beyond questioning, though: for the first time in five months, the PCUs started to approach and surpass 50,000 players with some consistency, and it looks like they're now climbing rather than going down, at least outside of Sundays.

As a comparison, here is the same data without any sorting by day of the week, and with a second set of graphs, along the right-hand axis, showing the difference between week-days compared to the week (or 2/3/4 weeks) before.

Again we can see how the weekly difference graphs are fairly well planted in the negative numbers after Incarna, and how they started to edge up towards positive (or at least near-zero) numbers once the news of Crucible, not to mention the patch itself, came out.

As a final note, while this post will be a referencing the situation as it is right now, I will continue to update the graphs with some (ir)regularity, so it might be worth coming back to see how things evolve from here (and maybe how the post suddenly seems to completely contradict what the graphs tell us…).