147 replies to this topic

[Ted Wayz]

"Vandalhooch brought up maths even though we all know math is hard."

On Loss: It is <101 = large drop, 101-250 = medium drop, 251-400 = even, 400< = small rise.
On Win: It is <101 = even, 101-250 = small rise, 251-400 = medium rise, 400< = large rise.

If my average match score is 251 what happens with a 1.0 W/L ratio? Bingo!

Also keep in mind 20 points for a win, 0 points for a loss so the table should actually read:

On Loss: It is <101 = large drop, 101-250 = medium drop, 251-400 = even, 400< = small rise.
On Win: It is <81 = even, 81-230 = small rise, 231-380 = medium rise, 380< = large rise.

So now it is even more biased to PSR rising.

So yeah, if you consistently do less than 200 damage you could be in trouble. But given random matchmaker is random you could also make it to tier 1.

[MrMadguy]

vandalhooch, on 11 January 2018 - 06:16 PM, said:

The people who claim that the system is just an XP bar are really just bad at math. They look at the list of all possible outcomes for a pilot's PSR at the end of match and see that there are more options for moving up than down and just assume that that's what will happen given time. None of them seem to comprehend that the possible outcomes are not all equally likely for any given pilot.

If you are a poor pilot and consistently do poorly in nearly every match, then even a W/L ratio of 1.0 will cause your PSR to go down! Not up.

PSR isn't rocket science. Treat results of matches as two related random factors: 1) Is it win or loss? 2) Match score. Then we have 4 match score intervals: 0-100, 100-250, 250-400 and >400. What values are assigned to this 4 intervals - depends on whether it's win or loss. We don't know exact numbers for rating weights, but let's assume, that Small = 1, Moderate = 2 and Large = 3, i.e. take any numbers, that follow Small < Moderate < Large rule. Therefore for loss it would be -3, -2, 0, 1 and for win 0, 1, 2, 3. Now we have bunch of matches. Result of each match R = one of this 8 numbers. All we need to know - whether ΣR is positive, negative or it's zero. That's it.

And there are 2 problems here:
1) No matter, what rating weights are - system is biased toward increasing of rating. Why? Just find summs of weights for every interval. We don't know exact numbers, but we know, that Small < Moderate < Large and therefore any numbers would show the same picture. For example -3 + 0 = -3, -2 + 1 = -1, 0 + 2 = 2 and 1 + 3 = 4. See? -3, -1, 2, 4. Wins give you more, than loses in any case. Your W/L should be < 1 in order to compensate this fact.
2) Simple fact, that you can't drop even with average stats below W/L = 1 and AvgMS = 200 suggests, that "Small rise" is actually way to big in comparison with "Moderate drop". I.e. it's not even -3, -1, 2, 4. It's something like -2.5, -0.5, 2, 3.5. I.e. drop for 100-250 on loss is so small, that it isn't even noticeable. That's why you have to have games below MS = 100 in order to drop.

If I'll have some free time, I'll make simulator, if you want.

[Bud Crue]

vandalhooch, on 11 January 2018 - 07:30 PM, said:

Because anecdotes are data now?

I don't know what data you have at your disposal but what I have is my stats showing how bad I am at this (I lose more than I win, I die more than I kill, etc. go look me up and see for yourself), combined with the little Tier 1 label to the left. Any other data is irrelevant to my anecdotal experience of being matched with and against people like you, despite clearly lacking your level of skill, because the math of the system says that is the way it should be.

If you have data that shows that my data is wrong, and that I am really Tier 3, and that this is all a big mistake, I would be happy to consider it; but the data I have at my disposal says I am not particularly good at this game, yet the Tier System has advanced me to the top of the heap because of its math. THAT is how the math has in fact worked, that is the objective reality here: the tier system says I am at the top despite all statistical evidence (data) to the contrary.

[Shifty McSwift]

What is data but an accumulation of anecdotal evidence?

[Burning2nd]

The data is a lie, you guys are over thinking this..

Its time played Ive got ten days of cockpit time in a pb,
days in ddc
days in a1(c)
days in a spider
days in a firebrand,

my bar is pegged out and has not moved since i got it there from teir 3 where i started.. which was moving up very quickly after they introduced it

[MrMadguy]

Here is my simulator:


Skill is assumed to have normal distribution. Avg and sigma should match your avg performance and give desired Avg MS at the end of simulation. PSR change is measured in same units, as Small, Moderate and Large values.

See? With this parameters PSR change should be -631. Series of simulations show, that this value is random value between -450 and -700.

Now, what we need to do - is to find such Small, Moderate and Large parameters, that would make PSR change positive.

Edited by MrMadguy, 12 January 2018 - 04:20 AM.

[arcana75]

MrMadguy, on 12 January 2018 - 03:49 AM, said:

Here is my simulator:


Skill is assumed to have normal distribution. Avg and sigma should match your avg performance and give desired Avg MS at the end of simulation. PSR change is measured in same units, as Small, Moderate and Large values.

See? With this parameters PSR change should be -631. Series of simulations show, that this value is random value between -450 and -700.

Now, what we need to do - is to find such Small, Moderate and Large parameters, that would make PSR change positive.

This is intriguing...

What are Small, Medium and Large for?

[MrMadguy]

arcana75, on 12 January 2018 - 04:47 AM, said:

This is intriguing...

What are Small, Medium and Large for?

PSR is actually rating, not just % of Tier. We see it as % of Tier, but it's some number - 1653 for example. And... Yeah, PGI have shown us a chart with some mysterious "small", "moderate" and "large" changes, but haven't given us any actual numbers. So, we can only guess, what is real relation between this numbers. May be it's 2x. Maybe it's 10x. Who knows.

Current suspicion - is that "Small" may be larger, than I assumed ("Moderate" is 1.3-1.5x, not 2x), that makes bias towards increasing even worse.

Edited by MrMadguy, 12 January 2018 - 05:29 AM.