I analysed 68 million lichess games to PROVE that the low time alarm makes people play worse

Forgive me for the clickbait title ;)

This is a project I cooked up yesterday and I think the results are pretty interesting. I pulled a month's worth of lichess games from database.lichess.org and tracket at what times blunders were most likely to occur.

Here is a sample graph:


And here is the full project with a writeup:
github.com/Antiochian/chess-blunders/blob/main/README.md

There are some interesting features of this graph to discuss here, but the main one I want to talk about is the big bump at the 20second mark, where players all of a sudden start blundering all over the place. Can it really be a coincidence that this moment of panic is the exact same time that the lichess low-time alarm beeps and the clock flashes red?

For 5 min game, the bump appears at 40 seconds, while for 10 minute games it appears at the 60 second mark - all, I believe, the exact times of the low-time alarm for the various time controls! Even 2000+ ELO players are not immune to this effect - if you click the above link, you can see graphs for all ELOs and all time controls. Hope you enjoy!

Do you have any other interesting insights/thoughts about my data you can share?
.
.
.
.
.
PS: if you are curious about the other features of the graph (eg: the dip at 90s which is caused by Beserk players) then a more thorough explanation of the various bumps and so on is included on the github page.

There's a feature to turn off the low beat sound alarm thing.

Also, I mean yes it's obvious people blunder more when the time is about to end...just like in anygame,

even boxing or any sports...when you are losing rounds and if it's towards to end of the game...last 20 seconds or so...

they tend to move faster and faster...sloppy

How does this prove that the low time alarm makes people play worse? I feel like it probably isn't the low time alarm itself, but actually the amount of time left: With less time, you have less time to think, therefore blunders increase. I somewhat doubt the "low time alarm panic" is the main factor here. However, an experiment could probably be devised to test this, and the results could definitely be interesting!

Also, when dealing with statistics, be careful of the word "prove". Correlation is not causation, and since your "study" is observation, not causation can be derived.

Anyways, cool project!

how to you analyze 68 million games when you only played 400?

I don't even notice the clock until the 20s alarm sounds. I also lean back when I think I'm winning, which decreases my level of play.

@TCF_Namelecc

The important note is the sudden "jump" at exactly the alarm time. There is a general smooth increase of blunders as time reduces (as you would expect), but the discontinuous "spike" at the 20s mark is very noticeable.

Further evidence that it is the alarm and not just the low time is that for 5min games it occurs at 40s, and for 10min games it occurs at 60 seconds. If all 3 time settings had the jump at 20s then I might think it was just a coincidence, but for all 3 to have their jumps at the exact (somewhat abritrary) alarm sound timing? It seems unlikely to just be general time pressure with this in mind.

Also yes, you're absolutely right, I haven't "proved" anything. Its just an interesting hypothesis that is somewhat supported by the data. Thats why I said "forgive me for the clickbait" :P

@DragonArcher
I'm a lot better at datascience than I am at chess, haha

Even so, this could just be that you notice that your time is now low and accordingly move quicker.

@lakeofcrackers

Good point, but I think there is something in the data which suggests otherwise.

If you zoom in real close on the data:


You can see that the number of blunders actually goes DOWN immediately after the alarm-time spike (before rising again once time gets really low)

Why would people play better at 15 seconds left than at 20 seconds left? It must be because the temporary shock of the alarm buzzer surprises people and makes them play much worse for one isingle move before recovering.

Or at least, thats my theory :)

If you really wanted to prove causation, you'd have to also establish a reasonable control.

still, pretty convincing :D

Fascinating. I don't know if this implies that the alarm makes you lose more often, I could e.g. see this alternative explanation:
Once the alarm goes people realize "Hold up, I need to move faster or I'll flag." Speeding up while playing worse obviously due to higher speed. Now those are worse moves however they mean you are less likely to flag because you sped up, I think overall the playing faster may be a good thing. (or at least what the players think is good)

An interesting experiment would be to have a high number of players, and have half of them turn off alarm and then a) look at their blunder rate but also b) look at their win rate. (or average rating gain or whatever) or maybe even c) look at their flag rate and see how much lower it is with alarm. (which itself might not say much but would be interesting to see too)

Though it's probably hard to find enough people to get a good sample size. But would be interesting.

EDIT: Concerning blunder rate going down a bit, may be due to people simplifying the position and so their job gets easier, or maybe starting to player slower again. (I probably have done the latter before, go super fast for a bit then slow down as my concentration drops a bit again :D )