Last month I had a post, A New Tool for Tracking EVE Online, showing how I used Google Trends to get an idea on how well EVE Online was doing based on search trends. Well, Google is at it again (NOTE: Figure of speech BTW, not meant to imply this is all that new. It’s been around since at least 2011.) They expanded their trend analysis project to include correlation analysis as well. You can find Google Correlate information here.
The easiest way to use Google Correlate is to put in a search term and let Google determine the correlation curve for you. I used the search term I used on by previous post – “EVE Online” – and got this result.
It is not completely unexpected that I’d get EVE Online related searches that correlated well with the “EVE Online” trend curve. The five best fitting correlations are specifically EVE related. There is one take away from it that I find interesting. They are mining and ISK related searches or related to ships in general. The first thing I wondered when I saw the results is, “Could this be related to RMT?”
That’s really cynical, I know. It could just be that people want to know how to make ISK in general. However, there are many ways to make ISK in EVE Online, but bots are only feasible for mining and missions. That would explain the correlations between mining, ships and ISK rather handily in my mind. And I can’t forget all that Russian activity I mentioned in the Google Trend post.
Anyway, back to Google Correlate. You can also provide your own correlation curves to try and match. Now, I’m no mathematician and I took Probability and Statistics a very, very long time ago, but I can still manage an Excel spreadsheet and that is all Google needs. So I decided to gin up a couple of slapdash curves and see what I got.
The first “curve” I based off the example Google gave. Their example looked for searches correlated to the Winter Solstice. They had a rather nifty equation for creating a COS waveform for weekly values. I decided I’d try the same thing but use EVE Online expansion dates rather than a solstice date. That was a little complicated. For one thing, Solstices happen on relatively easy to calculate dates: EVE Expansions… not so much. However, I decided I’d create a weekly COS normalization (I hope that’s the right term, damn that class was a long time ago) based on the days between expansions. Here is the equation and you can look at the raw Excel spreadsheet here. It’s nothing fancy, but the equation is there.
Value = 0.5 + 0.5 * COS(RADIANS(360 / (Next_Expansion_Date – Last_Expansion_Date) * (Week_Date-Last_Expansion_Date)))
Yeah, that’s a beast but it works. Here’s the curve it returned when I got everything done.
That looks really nice. Unfortunately when I copied the data into Google Correlate I got a big, fat nothing. There were no correlations found at all. So I changed tack. I decided to do a simple monthly data set where any month with an expansion got the value one and all the others got the value zero. However, to allow for any search lag, or whatever, if the expansion was during the last week of the month I also gave the following month a value of one. I thought that was reasonable. Here’s the curve I got for that method.
When I used that data in Google Correlate I got results; oh boy!
After looking at it for all of 2 seconds, I realized that they correlation was a false one. It didn’t tell me anything about EVE Online itself, it only confirms that quite a few expansions come just in time for Christmas. I had a really good chuckle over that one. However, it does give CCP something they could think about. Perhaps they could market EVE Online subscriptions as something to get dad for the holiday. I mean, for $15 a kid could get the old man something way better than a BBQ apron. They could get him a month of EVE Online, or enough ISK to buy the shiny new ship he wants! Let the world wide Christmas marketing of PLEX begin! Of course, CCP would have to figure out a gift PLEX program, but they’ve got lots of examples and a year to do it. Get cracking Yule Lads!
Anyway, I think I need to keep thinking about how to supply meaningful data to Google Correlate in order to get some better results. Until then there is always the method of entering a search string and letting Google provide the mathematical muscle. If I discover anything interesting I’ll be sure to tell you. If you give this a try yourself and discover something interesting, please share!