Where players quit
Learning paths through this topic
Detect how long a traffic or supply jam lasted without the jam hiding itself
A jam is a drop below normal, but normal itself is measured from recent history, and during a long jam recent history is the jam. The detector then forgets there is a problem. This entry shows how to keep the reference fixed for the jam's whole length.
At which turn do players start taking forever, and can I trust the late turns?
In turn-based games the time per turn tends to grow as the board fills up, until it feels like homework. Finding the turn where it takes off is useful, but the late turns come only from the players who stayed. This entry shows how to avoid fooling yourself.
Tell apart tutorial steps where players get stuck and steps they just skip past
Two tutorial steps can lose the same share of people for opposite reasons: one confuses them, the other bores them. The fix differs, so the numbers have to say which it is. This entry shows what to compare and when there is too little data to say.
Spot sudden building losses that the player did not cause, not their own cleanup
A big drop in the number of buildings could be a fire, a flood or a player tearing down a block to rebuild. Only the first can make someone leave. This entry shows how to separate losses the game did to the player from changes the player chose.
Spot bursts of undo presses that show a player fighting the controls
One undo is a correction. Ten in a few seconds is a person who does not understand what just happened. This entry shows how to find such bursts in a click log without flagging a player who simply undoes things slowly over a long session.
Count how often and how long a city stayed unhappy without counting every wobble
A happiness number that bounces around a limit would count dozens of short unhappy spells when the player felt only one long one. Count it naively and your charts lie. This entry shows how to count real episodes, and what to do about one that is still going.
Which unlocked units do players never build, and is that a balance problem?
If players earn a unit and then never use it, either it is weak, too costly or nobody noticed it. To see this you need to count only those who had a fair chance to use it. This entry shows how to flag dead units without blaming short games.
Do not count players who leave right after winning as players who gave up
A strategy campaign has an end, and most people close the game soon after reaching it. If your drop-off chart counts that as leaving, the final level looks like the hardest place in the game. This entry shows how to split the endings so the chart means something.
Find story choices where one answer is never taken, without counting timeouts
If nobody ever picks one answer, the writing behind it is wasted, and it may be the reason a choice feels fake. This entry shows how to count picks fairly, including answers that were never chosen at all, and when there are too few players to say anything.
Did players who saw the early clue stay longer? A comparison that fools you
It is tempting to compare players who saw a helpful scene with those who did not, and to conclude the scene keeps people. But people who leave early never get the chance to see it, which quietly rigs the result. This entry shows who must be left out of the comparison.
Tell if players read a story scene, skimmed it, or just left the window open
Time spent in a scene looks like interest, but a long stay can mean the person walked away and a short one that they clicked through. This entry shows how to judge a stay against the amount of text, so your numbers say whether the story is being read.
Do long walls of dialogue make players quit? Ranking scene lengths with few visits
Group your scenes by length and one group with only a handful of visits will almost always top the list of worst offenders just by luck. This entry shows how to pull small groups toward the overall picture so the ranking points at real problems.