More gambling yaaaay!
All 5 mil i had gone,nothing like last crates,i am getting unlucky lately, before i got 2-3 vehicles on acc and coupons.
1 Like
What vehicles did you get?
None this time, last crates i got MFG Tornado and some low tier smth
Ive spent 3.8 so far and have legitimately only gotten 15k silver lions, wagers, and backups.
Dang.
He didn’t link to yours at all, I doubt he has even seen it. As I said, if you confine this to result sets from single users of N>1000, like yours, you get about .004 as the average over the last ten SL chest events, with about .001 for coupon drops. There’s error bars on that.
Nothing wrong with aggregating smaller samples the way @war_Tinder has been doing this year, but the answer you get is always going to be near the upper probability bound, because the smaller the reward the less likely you’re going to take the time to fill out the communal spreadsheet. The lucky will always be just a little bit louder in aggregate. (Tinder actually concedes this point in his post today, when he says he doesn’t want any more single-box results, please which escaped this guy who linked to him I guess.)
I only said he was an innumerate because he said in this thread that because @war_Tinder 's last event communal result set had a vehicle drop chance of ~.0049 then obviously Gaijin must have set the one variable in their gambling code that decides this to exactly .005 (and .0005 for coupons) and no other value was possible. There is no way you could statistically support as precise a conclusion as that based on the probabilistics involved here.
Basically we’re agreeing: in this kind of game of small numbers, it’s put up 60mil+ SL or GTFO, tbh.
Fair, personally I’d probably have more fun with it. But to each their own I suppose.
This no single box drop thing is what I did with the Travel Suitcase. The issue was with the Toolbox in April. I already fixed this for the last set of drops so the Travel Suitcase drops don’t have this issue.
Also this time, I am tracking standard vehicles and coupons seperately so we can get chances for both.
Yes, I see that, but the effect on reporting persists, it just lessens. I do actually use your Travel Suitcase estimate which the guy was convinced I’d never heard of and didn’t want to read, had been referring to it all year, but as an upper limit on the probability, as I said. It’s a useful check on the larger aggregate results from CCs and whales alike.
As one example of the effect, let’s say you’re a content creator and you do a run of 100 for your audience, spent 6 mil. You get nothing. Are you always going to take great lengths to preserve that content, or maybe just meh and move on? Sometimes you will. But you’ll be slightly more likely to ditch it. But if you got a bunch of vehicles in that 200, it’s good content and you’ll be slightly more likely to keep it up, jackpot, fun times. Now do the same thing with 2000, that you spent 120 million SL on. Yep, you’re making a video of that stream and it’s going up on Youtube, no matter what you got, good or bad. And that’s what introduces the bias against smaller samples.
Same with the regular user. I do batches of 50 myself for these lootbox events so I don’t waste all the boosters, then another one when those boosters run out. I tracked those batches of 50 pretty closely on my own spreadsheet for a long while. But every now and again, I’d still have a little cash, or I got one box as part of an event reward, and I just burned that one extra box, and given the odds, probably got nothing. But I didn’t necessarily change my 50 to a 51 on the spreadsheet every time I did that, event was going and I had grinding to do. Nil results on lottery-type events like this are always always going to be underreported, so you need big samples so that the incentives are lined up to offset that effect.
Easy to prove me wrong if you want to try, of course. Just make your raw data available, or run regressions to exclude results below a certain sample size and graph what it does to your baseline P value. If I’m right as you increase your minimum permitted sample size N, your baseline P will tend to drop. As you said today, if N=1, you will get some crazy high P for the reasons above. N=100 will be more accurate. N=1000 even more accurate than that. :)
1 Like
Oh I fully understand. That’s why for the Travel Suitcase, I also added into that sheet videos of openings of boxes with giant amounts (e.g. openings of 500-1000 boxes) to make the results better. That’s why the Travel Suitcase sheet has over 4K boxes in it.
Ofc it’s still not perfect, but I am cognizant of that reporting bias issue and am trying to fix it as much as I can.
Once you need others rewards I can give you an indepth list of whatI got again like with the suitcases. Got pretty decent odds again. By the time crates end Ill have opened well over 3k of them
It doesn’t impact the results if you are smart enough to understand that if you don’t include the reward just because it’s small or whatever it will make the whole thing pointless.
For example i open boxes in batches of 10 and then write everything down without any exceptions, only then i go to open more.
What you describe only makes any difference if those sheets are made by people who have absolutely no idea what and why are they even doing. If you are deciding to create such list, by default you most likely understand what is it all about.
I agree with last statement though for the most part, i would just set the lowest bar around 20mln, by that point you are quite likely get 1 or 2 vehicles already. Unless they are the worst out of the worst you are good.
Yes, but when the sample is made by one person, in one “pull” of the odds basically, it’s going to be inherently more valuable than 10 smaller samples by 10 people aggregated together through a voluntary self-reporting mechanism after the fact. That’s the comparison we’re making and why I’m saying your ~.0045 is probably closer to the actual number here than Tinder’s ~.005.
(To be fair I don’t think it’s that big a difference, really, the real number’s in there somewhere, no one cares about the fourth decimal place, I certainly don’t: I was just reacting to someone up thread to saying it must be exactly .005 and anyone who thought it was something different must be dim.)
In survey science you do things to offset this like, for instance, getting your sample audience to agree to fill out the report beforehand and incentivizing to make sure they all do it even if it wasn’t fun or convenient or cheap, in the end. If you could do that here for a bunch of people the aggregate of their samples could be just as valuable as your single person big sample if the sample sizes were the same. Not really possible here, though.
Of course, the company could just tell us the real odds too, any time they felt like it. That would be nice :)
1 Like
3 Likes
It’s christmas and the reward probability is non-existant.
ontop of my 3.5million I spent another 180k for 3 crates and got just boosters and backups again. Still not a single vehicle or decal. I have 47ksl left. With that 180 being earned since the last time I posted about spending 3.5million.
bro the chance for a vehicle is about 0,33% (for me) Others say 0,5% but its always 1 vehicle per 300 crates for me
