Linux remains above macOS on the Steam Survey for January 2024

Because the linear approximation systematically overestimates all of the oldest points and underestimates all of the youngest points (except a single one we would attribute to some measurement error, probably located in the far east). The line doesn't fit, and this not in a "jumpy" way (as to be to be expected), but in a systematical way. The curve to be approximated isn't linear, at least not over the whole time.

So we could either try two linear curves, or some curve with a slope (slightly) increasing over time.

It's not a fit to a curve. It's not supposed to be a fit to a curve, and it's not trying to be a fit to a curve. It's a trend line: a line which demonstrates the trend over time.

It fails to demonstrate the actual trend from 2018 to mid-2021 as well as in 2023.

Is there any reason not to look for a better representation than this?


Last edited by Eike on 2 February 2024 at 2:51 pm UTC

It fails to demonstrate the actual trend from 2018 to mid-2021 as well as in 2023.

The actual trend is that the number is going up over time. The trend line demonstrates that perfectly.

Is there any reason not to look for a better representation than this?

Arbitrary curve-fitting is not better in any way - it is much worse.


Last edited by CatKiller on 2 February 2024 at 3:04 pm UTC

Arbitrary curve-fitting is not better in any way - it is much worse.

Why do you think so?

I'm fine with not drawing two linear lines at hand-selected places, because it wouldn't be much better than having a single one, but there should be mathematical ways to find a much better fitting function.

A search dug up this, how about it?
https://tom-alexander.github.io/regression-js/

@Liam, do you have the numbers available computer-readable?


Last edited by Eike on 2 February 2024 at 3:18 pm UTC

I'm fine with the way it is. But it would be interesting to see a graph that sort of starts when the Steam Deck was released.

I'm fine with the way it is. But it would be interesting to see a graph that sort of starts when the Steam Deck was released.

You can already do that. Liam's graph lets you pick any start or end point from the dataset.

'm not sure if there are incentives to use the App Store over Steam. It seems like they get the same cut. I know macOS programs from the App Store tend to be gutted compared to programs distributed outside of it because of restrictions (DaVinci Resolve being one such example, Affinity Suite being another).

Well, the App Store is there on every Mac, you don't need to install it by yourself. Plus, users with iPhones will already know it's the place to get all software and Apple will actively steer people towards it. Also, IIRC, it would hide software that wouldn't run on your Mac (a simple check for OS version and 32/64 bits, nothing fancy like checking gfx card requirements).

There's a strong dislike for the built-in store on Windows AFAICT, this isn't the case on the Mac. People love it. And I'm sure devs who make software for the Mac do as well (except for the likes of Epic). After all, as we Linux users all know, hunting the web for programs and downloading installers is the silliest thing ever.

Plus, "tend to be gutted" is much too strong of a statement. Yes, software will need to be modified in some instances (like years ago LibreOffice not being able to be on there because of their Java dependecy, IDK if this has changed or not), but I'd say the vast majority is made with with Mac and App Store in mind from the start.

As for Steam, Valve doesn't need the Mac and Apple doesn't need Valve.

I can only speak for myself: I tried to use the App Store, but when I realized the functionality I needed to use in the software just didn't exist seemingly because of some weird App Store regulations, I gave up on it. I had a sample size of 2 programs, and they were both lacking something from memory. Of course, I didn't use it at all after that, so I can't say anything for other programs. I say this as someone who used an iPhone for 10 years and macOS for a little longer. It also doesn't have a great selection compared to say, homebrew. I think this is partly because open source programs aren't allowed on the App Store.[^1]

But in principle I agree with all your points. What better marketing strategy is there than to have your program displayed in the App Store, directly accessible compared to trying to find the developer's website? That's certainly reason enough to list your program in the App Store.

[^1]: I've tried to look into the situation. The developer for Raivo OTP claims the app can't be open source because Apple won't accept it. The FSF said 10 years ago that only the GPL/LGPL family of licenses weren't accepted. I don't really know what's going on here, but I'm assuming open source programs aren't allowed for some reason.

Apple allows programs published under(some) open source licenses in their appstore, but the rules of the appstore are written so incompatible with the open source principles that all versions of those programs that are actually on the app store aren't open source.
Apple wants all developers to actively forbid some forms of usage of their programs. This doesn't match the 0th software freedom or the open source principles.
The gpl is incompatible with it, because it forces all distributors of the code to allow their users to do whatever they want with the program.

As for why your choice of homebrew in the appstore is so bad. The reasons for that are more holistic. The smaller the developer the less resources they can put toward complying with store policies. Apple has the strictest store policies for general purpose computers(yes, your smart tv has stricter rules, but it's trying to be only a tv).
Same reason exists for Xbox(comparably strict rules, just not for general purpose), Playstation and to a lesser extend Nintendo Switch(yes, Nintendo is actually the least bad of the lot).
PC users just don't know how good they have it with their relatively open platform.
Microsoft has failed multiple times to cripple sideloading, thanks to interference of USA regulators Microsoft has to allow installing alternate OS'es and browsers, Intel had to allow AMD to compete, because IBM build that level of openess in it from the start.
Sure X86 is still proprietary, replacing your OS is hard, IntelME, etc., but compared to anything else on the market it's a breath of fresh air.
It's even possible to build a completely open source laptop.

Why do you think so?

Because it is.

As to why it is worse, fitting a curve is explicitly enforcing a particular model on the data. There is no model here; it is simply a record of sampled users over time. Curve-fitting would be wildly inappropriate.

No matter the numbers, I'm not going back to macOS, I'm not going to use Windows

The linear approximation to the curve really needs to be split into two parts! (somewhere in 2021?)

... or the software could come up with something non-linear?

Why?

Because the linear approximation systematically overestimates all of the oldest points and underestimates all of the youngest points (except a single one we would attribute to some measurement error, probably located in the far east). The line doesn't fit, and this not in a "jumpy" way (as to be to be expected), but in a systematical way. The curve to be approximated isn't linear, at least not over the whole time.

So we could either try two linear curves, or some curve with a slope (slightly) increasing over time.

Sadly, even the current trend-line is a Chart JS plugin, as it doesn't support trend lines at all.

Maybe I should look for one that does it better...

Or just split the DataSet in Pre/Post SteamDeck announcement :D

My take on some of the arguments here:

1. Chromebooks are pretty much an US thing. I know no one who owns a Chromebook here in Germany and when I looked for their prices, I know why. You get a pretty good Business Laptop for the price of a cheap tablet with a bad keyboard.

2. Not so sure about the Appstore. I can see that for single player titles, but for multiplayer? I really don't know. My guess is Apple Users tend to prefer console gaming.

Chromebooks are quite popular to hand out in schools here in Sweden.

Yes, they exist and are sold. I know that. Still, people living in certain regions might overestimate their impact. While around 30% of Laptops in the US are Chromebooks, it's 2% in Europe and close to zero in Asia. Current installations don't really matter either, since most of them are made to handle not much more then a browser. Yes, as Google opens up ChromeOS, beefier Chromebooks are sure to come, but chances they'll be popular here in Europe are much worse then in the US, where they are actually a trademark.


Last edited by const on 2 February 2024 at 4:42 pm UTC

Why do you think so?

Because it is.

Wow.

As to why it is worse, fitting a curve is explicitly enforcing a particular model on the data. There is no model here; it is simply a record of sampled users over time. Curve-fitting would be wildly inappropriate.

And you don't think developments in real life might be subject to mathematical description?

In my humble opinion, everybody with a bit of a mathematical heart should feel their toe nails roll up on the graphics.

If the linear line is only there to show that the numbers are rising, you could write it next to the curve as well instead: "These numbers are rising." Which would make very clear how much information this actually caries beyond what everybody can see in the actual numbers shown in the graph.

What the linear line is nebulizing is the actual change in trend, somewhere between the announcement and the release of the Steam Deck. Things were constant beforehand. Things went up since then. There's no rise since 2018, there's a rise since 2021.

Yes, they exist and are sold. I know that. Still, people living in certain regions might overestimate their impact.

I agree with that. While I did see some Chromebook adds lately here in Germany, after like a decade (I don't know) of only hearing about them from the US, I still don't think they play an important role here.


Last edited by Eike on 2 February 2024 at 5:03 pm UTC

In my humble opinion, everybody with a bit of a mathematical heart should feel their toe nails roll up on the graphics.

It's a shame that you don't understand the concept of a trend line, but I'm not your statistics teacher.

In my humble opinion, everybody with a bit of a mathematical heart should feel their toe nails roll up on the graphics.

It's a shame that you don't understand the concept of a trend line, but I'm not your statistics teacher.

I have to brag here: My statistics professor wanted to make me his tutor while I was still visiting his lectures.

But feel free to show me what supports you, e.g. from here:
https://en.wikipedia.org/wiki/Trend_line
I'm open to learning.


Last edited by Eike on 2 February 2024 at 5:24 pm UTC

I'm open to learning.

No, you aren't. You just don't like being told that your chuffing stupid idea is chuffing stupid.

You can fit those data points to an exponential curve. You can fit those points to a polynomial curve. You can fit those points to a sinusoidal curve. Absolutely none of those curves provide a model that has any predictive power to define the relationship between the date and what proportion of Steam users will be sampled using Linux, nor the proportion that would be sampled using Mac or Windows. It's just pointless fappery.

The trend is upwards, just like it's upwards for Windows and downwards for Mac. As shown by the trend line.

Were your statistical analysis skills as great as you claim, you would already know all of this; which would make suggesting such fappery even more chuffing stupid.

A completely different opinion here:

It might be that gaming on macOS becomes more of a thing, but rather than being through Steam, it's through the App Store, or the developer's website. If you've gone to the trouble of porting the game to Metal, you might as well use the App Store.

I'm not sure if there are incentives to use the App Store over Steam. It seems like they get the same cut. I know macOS programs from the App Store tend to be gutted compared to programs distributed outside of it because of restrictions (DaVinci Resolve being one such example, Affinity Suite being another).

Another thought: I would be completely okay with Linux hitting 15% and 14% of that being Steam Deck users. Unlike ChromeOS, Valve has made absolutely no effort to distinguish between the Steam Deck and Linux proper. They're all just one system. On the hardware survey, I would assume ChromeOS just shows up as Linux.

ChromeOS (now that Google has dropped Stadia) is making a push toward Proton. In November, Google partnered with ProtonDB: https://www.protondb.com/news/chromebook-reports

So...it's gonna be an interesting few years.

I have literally said that many times yet nobody listens here...
Mac gaming is moving to Mac App Store.

I'm open to learning.

No, you aren't. You just don't like being told that your chuffing stupid idea is chuffing stupid.

You can fit those data points to an exponential curve. You can fit those points to a polynomial curve. You can fit those points to a sinusoidal curve. Absolutely none of those curves provide a model that has any predictive power to define the relationship between the date and what proportion of Steam users will be sampled using Linux, nor the proportion that would be sampled using Mac or Windows. It's just pointless fappery.

I don't want to predict, nor did I say so. I want to describe what has happened.

The trend is upwards, just like it's upwards for Windows and downwards for Mac. As shown by the trend line.

The trend is upwards - but it wasn't for years. Half of the curve looks quite constant to me. If you're thinking that describing a constant function by an upward trend line is appropriate, I'm asking for sources for this. Derogatory terms don't convince me, by the way.

The point of this is that something has happened, somewhere in 2021, which had and still has quite some impact on Linux for gaming adoption, and is showing in the numbers since then.

The trend has changed. That's what the trend line is missing.

I don't want to predict, nor did I say so.

That's what fitting the data to a curve is.

I want to describe what has happened.

That's what the trend line does.

Which you already know because of your super statistical analysis skills, obviously.

Should you wish to examine the positive trend line for any range of the data, you already can. Slapping an arbitrary curve onto the data is pure fappery.

I don't want to predict, nor did I say so.

That's what fitting the data to a curve is.

I guess we found the misunderstanding. Of course, I wouldn't extrapolate this into the future. Nobody forces us to do that.

I want to describe what has happened.

That's what the trend line does.

As said and shown - not for like half of the curve.


The trend has changed, can we agree on that?

I guess we found the misunderstanding. Of course, I wouldn't extrapolate this into the future. Nobody forces us to do that.

That is literally the whole point of matching a data set to a curve.

As said and shown - not for like half of the curve.


The trend has changed, can we agree on that?

No, we cannot. The trend has always been positive. The gradient is steeper now than it was in the past, but that is not a change in trend.