Data Mining

General Question Time:

How do you guys gather data during alpha/beta for analysis/balancing?
Do you notify people you are doing that?
What kind of data do you gather?

Thanks.

Comments

  • Hold on till wednesday ;)
  • dislekcia said:
    Hold on till wednesday ;)
    You releasing a new alpha to us and it has info mining? :D
  • A detailed approach depends very heavily on your game. I like to track a lot of detailed things about the design of the game.

    Valuable Analytic Data from my experience:
    Basics are:
    DAU Daily Active Users
    Retention Funnel - % of players that are still in the game at each point of interaction during the first session
    for example, before loading screen, after loading screen, after first tutorial instructions, after execution of tutorial instructions, etc. The more small steps you track the more you can identify problems and fix them.
    First Session retention - as mentioned above
    1 Day retention, how many players return after the first day
    3 Day retention - if 1 and 3 day retention is bad, it means that your core mechanics are the problem and you should scrap and restart.
    7 Day retention
    14 Day retention
    21 day retention
    30 day retention
    60 day retention
    90 day retention and so on

    You can also track
    Session length,
    Sessions per day.
    Average time of day per session - This you have to be careful as you have to work in averages, so split the day up into morning, afternoon, evening and very late at night. With this data you can better understand the player and what experience they are looking for.

    Then from here it depends on your game. In combat games I liked to track:
    Fail rate - how many retries it took to beat a check point
    death rate vs npc
    Loot drops if any
    time per battle
    shots to kill per npc
    shots to kill player per npc

    I think you get the picture. Anything that has a read out you can measure, you should group either by level or by npc, depending on what you have and what makes sense.

    I track these numbers to match with my balancing to see if my Master Document's formulas are correct in their predicted read outs.
    If you were creating your Master Document with graphs you can match the analytic data with the graphs and instantly see the result.
    Thanked by 2edg3 Tuism
  • I have similar questions and thoughts, and want to know mostly if you are already doing this how are you implementing it? I tried google analytics myself but found connectivity to it can be sketchy on certain platforms (my Windows Phone and brother's android tablet don't seem to do the analytics at all).

    Also, don't roll your own.

    Looking forward to Wednesday then @dislekcia!

    @iPixelPierre: I believe most platforms require you to have a privacy policy if you are sending any kind of data? Im not sure though. It is possibly best to have a privacy policy if you track any data at all anyway (to be legally covered).
  • We have in house tools at wooga for analytics, which makes it much easier, I know that supercell uses these guys http://www.appannie.com/app-store-analytics/#mobile and my friend, yuji head of appannie japan has done a few pretty good drunken pitches for me late at night while we were out partying. haha. That said I have never used it for in app tracking, but for tracking the market it is brilliant, and the intelligence function lets you see everyone elses earnings and downloads through the app stores.

    another well know tool is
    http://www.flurry.com/

    I hope one of these are what you are looking for. :P
  • Well, I did some prototyping yesterday and found a solution for me for now.
    Just sending data to a php/json page and using that to submit into mysql.
    Works fine for what I want to do at this early stage in development.
  • Would *love* to know how this stuff is implemented! looking forward to @dislekcia's VIDEO :)
  • We use GameAnalytics, so far its a pretty awesome and free package for tracking any kind of information possible :) its for unity. Heads up if your building for windows mobile or windows 8 store it causes errors XD
  • dislekcia said:
    Hold on till wednesday ;)
    Do tell :)
    Thanked by 1edg3
  • Here you go - DD gameplay stats as we record them.

    It's not a complicated system, we just ping information to a URL from the game and go from there. It's related to our crash information collection system that we used during the beta (and that's still running, no crash dumps since early Jan, hah). It does require people to be online though and sometimes doesn't complete because it fails silently and we don't bother caching stuff on the client-side. But it's not super essential to us figuring out the game at this stage, that's what the beta was for.
    Thanked by 2Tuism hermantulleken
  • edited
    @dislekcia I am curious, do you use A/B testing to check if new features and updates are hurting the game and if yes, which system do you use for assigning test groups numbered id with prime number group separation or a straight percentage split?
  • Nice man, looks sweet. Cards, must get them.
  • edited
    Bladesway said:
    @dislekcia I am curious, do you use A/B testing to check if new features and updates are hurting the game and if yes, which system do you use for assigning test groups numbered id with prime number group separation or a straight percentage split?
    No. We added features and made balance changes during the beta in response to particular needs and design issues that we saw. Sometimes we'd make a change, not like where that went or discover another knock-on balance effect and reverse the change or do something different after that, but the game in general isn't really something you could do A/B testing on in the first place: Cydstep costing 10 mana vs it costing 7 mana has several obvious balance implications.

    For balance we made the game obscenely difficult and ratcheted that down as we wanted players to explore more of the end-game once the runaway strategies had been discovered and nerfed.

    We didn't do stuff like move buttons around via A/B, when it was time to do UI stuff like that we just gave people both UI options as a setting and randomised what it started at. The plan was to use the most popular option at the end, but we kinda just left most of those settings in because people liked them both ways and we'd already done the setup anyway ;)

    I'm not super-convinced that A/B testing is too useful in actual game design. It's good for optimising towards specific measured targets, so I guess placing monetisation buttons and tutorial progression and stuff, but we just did the old time-separated testing and saw constant improvement.
  • @dislekcia nice one, start hard and then wind down sounds like a solid approach.

    On the A/B testing of core mechanics, I actually worked a crap load with this recently and I found that it is actually the most amazing thing to see. I think to some degree you could argue, as you well did, that intuitive design can already tell you that these things will be a success, but it is really great to see it in concrete comparable results so you can reason through them for a better understanding of the player's mindset. For me the main great thing was to see the effect you could have on players coming back, just simply by denying them one feature, or adjusting the difficulty. We did a test on our missions where we were able to identify specific missions that caused players to drop, so we changed the missions and improved our retention by a big amount (sorry can't say how much NDAs and all that). Overall for me it has been an awesome advantage.

  • Thanks for the article :) will be giving it a read this weekend :D
Sign In or Register to comment.