Why we see patterns that are not there
People are good at spotting patterns, and online games give us plenty of raw material: win streaks, loss streaks, “good” servers, lucky times of day, a weapon that suddenly feels strong. Some of these are real. Many are what random data looks like when you stare at it closely.
The aim here is not to dismiss every pattern but to have a simple way of asking whether one is worth believing. The question is always the same: how often would chance alone produce what I am seeing?
Start with the size of the sample
Suppose you win 7 of your last 10 games and feel you have found a new level. If your true chance of winning each game were 50%, how surprising is 7 or more wins out of 10? Counting the ways: 120 sequences with 7 wins, 45 with 8, 10 with 9 and 1 with 10, out of 1,024 possible sequences. That is 176 out of 1,024, about 17%.
So a result at least that good would show up for a 50% player roughly one time in six. It is a hint, not evidence.
Now stretch it out. Winning 60 of 100 games is more convincing, since a 50% player would reach that only about 3% of the time or a bit less. The same 60% rate is much stronger evidence when it comes from more games. The core idea in sample statistics is that results from larger samples cluster more tightly around the true value, so a small sample can be far off without anything unusual going on.
Streaks are normal
Streaks feel meaningful, but random sequences are streakier than intuition suggests. A 50% player loses five games in a row with probability 1/32, about 3%, starting from any given game. Over a hundred games, a run of around six or seven identical results somewhere is typical. When people see a long losing run they often blame the game, their teammates or the servers, when the dull explanation is that long runs exist in random sequences.
This does not mean streaks never have causes. Tilt, fatigue and a change in your opponents can all produce real runs. The point is that a streak alone is weak evidence for any of them. If you suspect fatigue, check whether your results are worse late in the session across many sessions, not just in the one you remember.
The multiple comparisons trap
The more things you examine, the more likely one of them looks special by accident. Imagine you compare your results across 20 different maps, characters or times of day, and treat anything with a 5% chance of being a fluke as “significant”. With 20 independent checks, the chance that at least one looks significant by luck is 1 − 0.95²⁰, which is about 64%.
So finding that you “always win on one map” after scanning a long list of options is not much of a discovery. It is better to decide what you want to test before looking at the data, and then to see whether the pattern holds up on fresh games.
Questions that separate signal from noise
When you think you have found a trend, a few checks help:
- How many games is it based on? Dozens beat a handful, hundreds beat dozens.
- Does it hold up on new data? A real effect should keep showing up in games you did not use to find it.
- Is there a plausible reason? A map you struggle with because of its layout is more credible than a vague “lucky hour”.
- Did other things change? A patch, a new teammate or a different rank can all look like a trend in your own play.
- Could selection be fooling you? If you only remember or record the dramatic sessions, the record is biased.
Regression to the mean
Extreme results tend to be followed by more typical ones. If you have an unusually good week, the following week will usually look closer to your normal level, without anything having gone wrong. People often credit a change in setup or routine for an improvement that was partly luck, and then blame a “slump” on something else when results return to average. Recognizing this saves a lot of unnecessary tinkering. It also helps in judging game-wide data: a character that tops the win rates after a short window may simply have had a good run, and the figures often drift back.
What counts as a trend worth acting on
A trend deserves action when it is large enough to matter, supported by a decent amount of data and consistent with a reason you can explain. Even then, treat your conclusion as provisional. Track the numbers for a while longer.
Keeping a simple log helps. Record the date, mode, result and anything you changed. After a few hundred entries you can look at your win rates over blocks of games and see whether the line wanders around a level or moves steadily in one direction. Wandering is normal. A steady, sustained move is what suggests improvement.
For longer-term progress, thinking in terms of outcome distributions, the full spread of results you could expect, rather than single results, keeps expectations realistic. And since every online game also contains an element of random events, some variation in any record is guaranteed.
None of this makes the numbers a guarantee about future games. It simply makes it easier to avoid chasing patterns that were never there.