Conversation

Notices

1. Embed this notice
   Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:11:18 JST Asta [AMP]

   statistical model of steam game pricing that includes publisher history of sales, % of cuts, and the cut % percentage vs. publication date as a "publisher desperation" factor to determine the optimal time to buy, say, Dragon Age: Veilguard
   
   (also one that includes "did you ever fucking finish Inquisition you dweeb?" as a factor to instantly shortcircuit the model into a "do not buy ever" state).

   • Embed this notice
     Cassandra Granade 🏳️‍⚧️ (xgranade@wandering.shop)'s status on Friday, 13-Dec-2024 08:11:11 JST Cassandra Granade 🏳️‍⚧️

     in reply to

     • cthos 🐱

     @cthos @aud Seriously, there's a hell of a lot of steps in the causal chain between "see ad" and "spend money." Even if it seems direct, like a "spend money" button that you measure clickthroughs on, it's a hell of a leap to claim that the money was spent *because* of that ad, and not everything else that happened before that ad appeared.

   • Embed this notice
     cthos 🐱 (cthos@mastodon.cthos.dev)'s status on Friday, 13-Dec-2024 08:11:12 JST cthos 🐱

     in reply to

     @aud Oh even measuring the effectiveness of an ad buy is a lot of "vibes" anyhow so they might not even know.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:11:13 JST Asta [AMP]

     in reply to

     • cthos 🐱

     @cthos@mastodon.cthos.dev yeah, right? Like, god, if you're gonna burn up the earth to power thinking rocks specifically to build a massive global surveillance network, at least... well. I guess using it more effectively isn't good, so maybe just stop, go get a hobby.

   • Embed this notice
     cthos 🐱 (cthos@mastodon.cthos.dev)'s status on Friday, 13-Dec-2024 08:11:13 JST cthos 🐱

     in reply to

     @aud Yeah, it's a fun thing to think about because that's all directly related to how people spend money on ad buys and what the ad networks are set up to maximize their own profits.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:11:13 JST Asta [AMP]

     in reply to

     • cthos 🐱

     @cthos@mastodon.cthos.dev I guess that's the nice thing about them using (what I assume are) black box models for the targeted demographic. Like, it's certainly possible to infer more general categories such as "you bought a hot water heater, so you are building or renovating a house" and then suggest other things (these are all hypotheticals to me as a lifelong renter, for the record). But you know, extra computation is expensive, takes more work, etc etc etc, and does Lowe's even know or care they're getting stiffed on this deal? If it's marginally better than what they had before, then they're probably fine with it.

   • Embed this notice
     cthos 🐱 (cthos@mastodon.cthos.dev)'s status on Friday, 13-Dec-2024 08:11:14 JST cthos 🐱

     in reply to

     @aud Well yeah, that's the ad-network effect for hyper targeting which is a whooole other thing than video game statistics correlating to one another. Like, "what other kinds of games do people who buy my game also buy" (note, this is a thing that DTRPG tells me) can be useful to make a better game as you note.

     As a related thought point, that hypertargeting also kinda sucks and is why once you buy that bra you keep seeing ads for *more of the same bra* immediately afterward.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:11:15 JST Asta [AMP]

     in reply to

     • cthos 🐱

     @cthos@mastodon.cthos.dev to be fair, I suppose we're not usually rewarded with literal shinies for being tracked. Usually we're 'rewarded' with "more fucking advertising".
     
     But joking aside, I do wonder if that's also another reason (besides copy protection) that 'always-online' mechanisms exists even in single player games. If you were the sort, you could absolutely always the game in offline mode on Steam and never load it up while Steam is offline. Your save games would still back up, if you use the steam cloud, but the achievements would never trigger.
     
     (it's probably more of a happy coincidental side effect rather than a reason but)

   • Embed this notice
     cthos 🐱 (cthos@mastodon.cthos.dev)'s status on Friday, 13-Dec-2024 08:11:15 JST cthos 🐱

     in reply to

     @aud Part of the reason it's literal shinies is because putting phone-home telemetry in an offline game is generally a bad call (and IIRC wasn't allowed on most platforms besides mobile) and you do really want to be able to infer things about how people are playing the game, and the dopamine popups are handy for that.

     They're also fun to program, cause you can put in achievements that are just for funsies.

     Related inverse operation: "Gameifying" more traditional web experiences.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:11:15 JST Asta [AMP]

     in reply to

     • cthos 🐱

     @cthos@mastodon.cthos.dev Yeah. Thinking about this more, too, there's probably some element of what information achievements expose and who it is exposed to. I'm not mad about developers having some statistics to infer how players experienced the game as a general rule, but that comes from a context where we're discussing games as creative, interactive forms of expression. Like, yes, that is detailed information, but if it's exposed to the publisher/developer only and mainly consists of that kind of stuff, I'm fine with it.
     
     That doesn't mean that's universal, though; I'm sure social stigma and shame play into it. For instance, people who own an assload of hentai games may not feel as comfortable having statistics collected about that; it may feel (somewhat ironically, I suppose) invasive or be a sensitive, judgement filled topic for them.
     
     Even getting on to that, though, the kind of statistics collected online aren't like, that great, all things considered. For example, I don't really want a toothpaste manufacturer to know what underwear I buy (but inversely, I probably don't care too much if my underwear maker knows what toothpaste I use). I'm sure the degree to which the data is considered personal and abusable vs. how important it could be in someone else's hands is a factor. By collecting achievement statistics, a developer could potentially maybe improve a game in the future. We're talking very lengthy creative works, here, that, you know, we might not be willing or able to give verbal feedback to. And if someone knows I played Dragon Age, for instance, it's not like... a big deal? At this time? So the risk is rather low and the potential reward (better Dragon Age games!) is relatively high.
     
     On the other hand, my toothpaste vendor knowing what kind of underwear I buy... that's not really high quality information all on its own, I suspect (which is probably why they try and build profiles from as much data as possible, but I digress). It doesn't tell you that I liked it or that it's a good fit or if I had a specific purpose for it. it DOES feel highly violating that they might be parsing and using that information and it could certainly be used against me, as it gives insight as to what my body is or what I would like it to be, amongst other things. There's also not really much good to be wrung out of it, so to speak: if I buy the same bra brand, that's pretty much all they know. It's a common story that a lot of people buy the wrong bra size so you can't really say, "this is a good fit for her". You can try to infer about how long it lasts, or how my anatomy is shaped, but even then, they're not going to use that to build a better bra, are they? "Does your bra fit well? Did it last a long time?" are two quick questions with quick answers that summarize the experience pretty well. I don't need the toothpaste guy getting a profile with that info and the bra maker could ask me.
     
     So: scope, risk, reward, etc.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:11:16 JST Asta [AMP]

     in reply to

     I mean, I can't imagine this a is radical, unknown thing, right? Like this is probably openly acknowledged as the reason these things exist? It's a publishing data goldmine. "For whatever reason a lot of people did not make the step from achievement I to J so maybe people hated that part. Patch it?"

   • Embed this notice
     cthos 🐱 (cthos@mastodon.cthos.dev)'s status on Friday, 13-Dec-2024 08:11:16 JST cthos 🐱

     in reply to

     @aud I think this is pretty common knowledge yes.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:11:17 JST Asta [AMP]

     in reply to

     I don't know if this is an open secret or even documented thing in game publishing circles but I am quite sure the very common "you hit the start game button and watched to the end of the cutscene!" achievement and the "you did it! you beat the game at least once!" achievement (as well as achievements along story gated points) exist solely so the manufacturer can see:
     
     A. Who bought the game and actually made an attempt to start playing?
     B. Of those, at what story point did they fuck off and stop playing forever?
     C. What percentage of players who actually started... finished?
     D. Of those fuckers, who went on the magical extra secret game hidden cheevos?
     
     etc

   • Embed this notice
     Cassandra Granade 🏳️‍⚧️ (xgranade@wandering.shop)'s status on Friday, 13-Dec-2024 08:14:19 JST Cassandra Granade 🏳️‍⚧️

     in reply to

     • cthos 🐱

     @aud @cthos I would guess (and to be clear, this is *only* a guess) that there's enough of a causal influence that what they're measuring is a noisy, noisy proxy that has enough correlation that they can still meaningfully optimize it. Just, you know, as long as no one forgets that it is just a proxy.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:14:20 JST Asta [AMP]

     in reply to

     • Cassandra Granade 🏳️‍⚧️
     • cthos 🐱

     @xgranade@wandering.shop @cthos@mastodon.cthos.dev right? It's an entire industry that is basically just Lisa's rock that wards tigers away.
     
     easy money, I suppose.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:36:19 JST Asta [AMP]

     in reply to

     • Cassandra Granade 🏳️‍⚧️
     • cthos 🐱

     @cthos@mastodon.cthos.dev @xgranade@wandering.shop This is an issue that came up a lot during simulations: you know, you can calculate the confidence intervals of your distribution that you're sampling, but it's a measure about the statistical estimator you used on your distribution, not necessarily your sampling process.
     
     If you knew the ground truth answer, that could tell you whether your sampling is good. If your statistical estimate is far away from the ground truth, your measures of confidence can give you information about whether it's random noise, or whether you have a genuine problem with something but without that ground truth it's not sufficient to tell you whether your sampling process is good or accurate (like, it might just say, "your estimates are very consistent", or in this case, "our estimates of how many clickthroughs resulted in sales are very consistent" or inconsistent or whatever. And that number might seem high, but again, it's only a measure of the sampling process). And it's an industry that has a vested interest in not answering that question.
     
     Anyway... sorry, that was a bit of a ramble. I had to deal with this a lot in grad school (basically I would end up with answers with a super high degree of confidence that didn't agree with what I should have been getting and it sucked) since simulations are a huge sampling problem.
     
     also I know I am preaching to the choir here. Just sort of writing as I'm thinking about this since it's one I spent a lot of time on in the past... much to my chagrin.

   • Embed this notice
     Cassandra Granade 🏳️‍⚧️ (xgranade@wandering.shop)'s status on Friday, 13-Dec-2024 08:36:19 JST Cassandra Granade 🏳️‍⚧️

     in reply to

     • cthos 🐱

     @aud @cthos That's where the bootstrap method really helps... but I digress.

   • Embed this notice
     cthos 🐱 (cthos@mastodon.cthos.dev)'s status on Friday, 13-Dec-2024 08:36:20 JST cthos 🐱

     in reply to

     • Cassandra Granade 🏳️‍⚧️

     @xgranade @aud From what I've observed, it's just a lot of "we put the utm_campaign variable on the thing and it's getting a lot of clickthrus so it's gotta be working well".

     But for stuff like buying TV ad spots....... it's like zero data.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:36:20 JST Asta [AMP]

     in reply to

     • Cassandra Granade 🏳️‍⚧️
     • cthos 🐱

     @cthos@mastodon.cthos.dev @xgranade@wandering.shop ah, right, I can't believe I totally forgot about clickthrus. Well, it's not like cross site tracking isn't a thing, so that probably gives them ... I'm going to use the term "confidence", but I don't mean true confidence, as in, actual confidence that this is the truth. I wouldn't be surprised if they were abusing statistical terms for sales.
     
     As in, they probably have a statistical measure of confidence which itself may or may not be high (IE, this many clickthroughs correlated with this many sales, here's the 95% CI), but the degree to which that is actually correlated with reality.... mmmmm. How do I put it? If this was the case, I don't think there's a good reason to accept that quantification as being necessarily realistic or representative of whatever the "true" impact is. That is, I think it's an easy number to calculate and report and it looks good in sales pitches and slide decks from marketing. But it's still highly context sensitive: if I ran the only goddamn lemonade stand in the world and everyone had a disease that only lemonade could cure, I could spend an assload or no money on marketing and it wouldn't really make a difference, but if I did spend money they could certainly give me a number that made it look well spent.

   • Embed this notice
     Cassandra Granade 🏳️‍⚧️ (xgranade@wandering.shop)'s status on Friday, 13-Dec-2024 08:41:41 JST Cassandra Granade 🏳️‍⚧️

     in reply to

     • cthos 🐱

     @aud @cthos Eh. Confidence intervals are overrated anyway. (Says the person who printed up about a hundred t shirts saying "credibility is better than confidence" to troll a referee).

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:41:42 JST Asta [AMP]

     in reply to

     • Cassandra Granade 🏳️‍⚧️
     • cthos 🐱

     @xgranade@wandering.shop @cthos@mastodon.cthos.dev yep 😂 that's one method we used. a lot. for precisely that reason... aaaand that we really had no idea what our underlying distributions did actually look like. That can definitely help, but the term "confidence interval" is... mmm, now that I'm thinking back on it, I feel like it should get another name because it's too easy to make it sound like something it's not...

   • Embed this notice
     Cassandra Granade 🏳️‍⚧️ (xgranade@wandering.shop)'s status on Friday, 13-Dec-2024 08:48:55 JST Cassandra Granade 🏳️‍⚧️

     in reply to

     • cthos 🐱

     @cthos @aud It's always odd to see push polling but where the person being pushed to is some internal politics thing.

   • Embed this notice
     cthos 🐱 (cthos@mastodon.cthos.dev)'s status on Friday, 13-Dec-2024 08:48:56 JST cthos 🐱

     in reply to

     • Cassandra Granade 🏳️‍⚧️

     @aud @xgranade Relatedly: "We designed a survey that proves our point because we designed the questions to prove our point".

   • Embed this notice
     cthos 🐱 (cthos@mastodon.cthos.dev)'s status on Friday, 13-Dec-2024 08:48:56 JST cthos 🐱

     in reply to

     • Cassandra Granade 🏳️‍⚧️

     @aud @xgranade Yes, I'm still pissed at Proton.

   • Embed this notice
     Asta [AMP] (aud@fire.asta.lgbt)'s status on Friday, 13-Dec-2024 08:48:57 JST Asta [AMP]

     in reply to

     • Cassandra Granade 🏳️‍⚧️
     • cthos 🐱

     @cthos@mastodon.cthos.dev @xgranade@wandering.shop I look "forward" to the same abuse being used to justify further "AI" spend. "Look how many of your developers are using LLMs for coding! It's such and such percentage and if you're curious about the statistics look at our 'confidence intervals'" etc etc etc, completely ignoring "did they have a choice", "did they actually like it", blah blah blah blah. Like if your sampling was just "did they fire up an IDE that was connected to an account that has 'AI' integration" then that number might be very high and you might have a high degree of confidence that is the correct answer for your sample buuuuut your sampling sucks relative to the truth. But hey, does the C-suite really know enough statistics to sort that out?