Steam Survey for April 2026 shows Linux still trending well

Quoting: jjaksicSteam Survey is useless garbage. It's not humanly possible to do polling and statistics any worse than this. Please ignore it, it's just worthless random noise junk not worth anyone's 2 seconds of attention (except maybe to teach a toddler how not to do statistics).

You always this cheery or is it your birthday which caused you to just glow with joy and spread some love?

Last edited by TheSHEEEP on 3 May 2026 at 8:34 am UTC

Quoting: PlayingOnLinuxphoneSince people don't trust me, I also draw some lines:

External Media: You need to be logged in to view this.

Everyone can extrapolate the line themselves.

It's following a sigmoidal pattern not a heaviside linear ... Heaviside linear is means to approximate a sigmoidal

It can only ever follow a sigmoidal and the is determining the steady state value (it won't be 100%... ) and the peak acceleration/adoption point.

I pasted a few sigmoidal fits for the last few updates and I expected a slight correction this month and it doesn't do anything to the general fit. 2030 for 10% is still on track.

Quoting: Naib2030 for 10% is still on track.

Great. When do we reach 105% then?

Great, the exponential trend line follows the data a lot better. It's more representative than the linear one.

I don't get the fuss about trend lines. The problem I have with that is that slapping a trend line on the data is assuming that there is one function driving the shape of the actual curve that stays more or less the same throughout. While I am not a statistics person at all, to me this screams "oversimplification". If you assume instead that there are several factors at play on how the Linux share develops, factors that may interact with one another, that become relevant and possibly irrelevant again at arbitrary points of time, I don't see how a simple trend line is supposed to appropriately model all of that.

I mean, do you plot a person's height from age 0 to 13 and then predict how tall they are when they reach age 90? So what are those of you expecting from a trend line to do for you? Predict how unknown future events impact the development? Because in all likelihood, there will be such events. Of course you can say, "If the Linux share develops like it has, X will happen in Y years", but what purpose does that serve?

I don't mean to rage bait anyone or discredit anyone's opinion on that matter, I'm honestly curious what you expect of a trend line to do for you. All I can see it doing for me is separating noise from the signal, but that isn't as easy as finding a curve that fits nicely, either. At least I don't expect it to be.

I don’t get the fuss against it. People like different ways of looking at data 🤷‍♂️

The entire point is to cut through the sampling noise. Linear works to a point then the divergence becomes apparent.
Some form of exponent is then key to track it. At some point this will fail and a logistic is needed.

Quoting: Liam DaweI don’t get the fuss against it.

You can literally always find a polynomial that goes exactly through every point. You can literally always find a sine function that goes through every point.

Should you be plotting your points against some function your model needs explanatory power for why "percentage of Steam users running Linux" should be some factor of e raised to the power of "number of days since 2017," say, or "the tangent of days since 2017," or "the inverse of 1 plus e to the power of the negative of the number of days since 2017." Or whatever bullshit function looks a bit like it.

"Tee-hee I like de dot-to-dots" is meaningless gibberish.

That's why.

Okay, well if you don't like them, toggle them off when you visit the page as you can do that now 👍

Quoting: Liam DaweOkay, well if you don't like them, toggle them off when you visit the page as you can do that now 👍

Like I said, I understand why you put it there. No worries. And, yeah, being able to turn it off was a great move.

Quoting: NaibIt's following a sigmoidal pattern

As data shows, clearly not. I mean you can give me the formula for a smooth sigmoidal that fits into a 5.6 times increase from one month to another. That 5.6 times is steady for already 7 months. Your mistake is to not take the events into account and so you would over-simply what actually is happening.

Common growth is like accelerating a vehicle. It starts slow, will accelerate faster in the mid and slows down acceleration at the end again. But what happens with an event? A crash instantly slows down or increases the speed depending on the impact point and direction. And that was happening with Steam Deck and Win10 EOL as the data shows.

That being said, I also don't belief it will always stay a flat line. It can change at any point in the future. In my opinion it is just the best way to interpret (denoise) the current graph.

People coming to this site aren't data scientists looking to denoise a graph, they just want to know if Linux gaming is becoming more popular. You're overthinking it. Statistics are meant to represent information to a target audience in a way that's useful to them, not to be an arbiter of the "correct" way of measuring things.

Last edited by GoEsr on 3 May 2026 at 8:58 pm UTC

Not sure what you are trying to say. Denoising graphs are done to make it easier to read. GamingOnLinux does this, I chose another form, because I think it is necessary to read the actual growth of the half year and the prediction of the next months.
That I argued so much has nothing to do with science or overthinking, but with people that say I am wrong, so I had to deliver more arguments that I am not wrong (future prediction excluded - nobody can be sure about).