32 replies to this topic

[Quicksilver Aberration]

KingCobra, on 11 December 2023 - 12:52 AM, said:

And it's not my armor amount I run full armor on 95% of my mechs it's my skill set I run 80% full defensive skills to even try to compete in battles. When i run a mech before i skill it up I'm usually killed in just a few minutes of game play or severely crippled as to only do a few hundred damage output.

MWO is now becoming like Mechwarrior4 at the end of its life cycle 95 percent of the mechs that were played were assaults no others but a few heavies ( 7 er nova) ETC could even survive as weapons were modified to be 6x overpowered compared to stock values, so Armor meant nothing in MechWarrior4 just like MWO is becoming now armor right now in MWO is totally useless and the only compensation to all the overpowered weapons in MWO are the skill set you can provide your mech for defensive ability's.

The volume of damage being thrown around is nothing new, this game has always favored heavier mechs. IE, this game has always been like MW4 because it is based on the same source material and didn't try to change enough. In fact I'd argue the only real thing it did that benefitted lights/mediums so hard is it undid MW4's undue fear of SL/ML boats (MLs/SLs did like half the damage they do currently).

Hitboxes etc is more subjective, ultimately you just want a signal that indicates a symptom, not the cause because that generally requires significant more investigation that likely goes beyond the metrics.

[Tarteso]

Der Geisterbaer, on 11 December 2023 - 12:10 AM, said:

Which numbers are you looking at exactly there to come to this conclusion?

• Lights have 309160 drops in total (lowest absolute number) across 22 Light chassis => Average drop number is ~14053 within just the Light class but drops to ~2811 across all 110 chassis.
  
• Mediums have 366429 drops in total (second lowest absolute number) across 29 chassis => Average drop number ~ 12214 within just Medium class but drop to ~3.331 across all 110 chassis.
  
• Heavies have 480076 drops in total (second highest absolute number) across 28 chassis => Average drop number is ~16554 within Heavy class but drop to ~4364 across all 110 chassis.
  
• Assaults have 565218 drops in total (highest absolute number) across 31 chassis => Average drop number is ~17663 within just Assault class but drop to ~5138 across all 110 chassis.

I certainly would not call that "similar" in terms of usage. In both ways of looking at those numbers there's a rather significant difference for both Lights and Mediums where Lights are the absolute losers, the relative losers across all chassis and Mediums being the flatout relative losers across weight class after the Light class that does better there but still falls farther behind Heavies and Assaults than it manages to outperform Mediums in this scenario.




[edit]Now obviously the data is not entirely complete (1720883 counted mech drops vs. 1720896 required for 71704 QP matches with 24 mechs each) but if we were to stipulate a desired equal distribution across 4 weight classes then the drops per class should work out to 430224 drops in each weight class which only Heavies hit within a reasonable margin of about +-10% and once again Lights are left in the dust along with Mediums[/edit]



I would say that your conclusion is a bit "off" there.

lol, I already knew you would come here

Absolute numbers and averages are catchy, but don't tell the whole story (and they can be misleading sometimes). This is my quick modest analysis. Feel free to perform your own (or anybody), maybe even more sofisticated, if you care.

And before somebody try to start a flame war about lights, please check my mech utilization stats.

Source data: link above, sheet 1; added new column for the class group: L(ight), M(edium), H(eavy), AS(sault)
Methods: Kruskall-Wallis & Dunn's test, alpha 0.05
Results (simplified):

No significant differences in: Total drops, W/L, Survival %, and Popularity (All); Note: Popularity (all) means % mech use.


Significant differences found for:

1. K/D: AS=H; H=M=L; AS significantly better than M, L
2. Kills per match: AS significantly better than H, M, L; H=M=L
3. AVG Score: AS significantly better; H=M; L significantly the worse
4. AVG Damage: AS significantly better; then H; M=L

Conclusion: lights were as used (between 2023-11-01 and 2023-12-05) as the other classes, and had a similar survival %.

Also, there were significant differences in K/D, Kills per match, AVG Score, and AVG Damage, which are metrics more related to firepower; so, it makes sense. However, lights did the job as good as mediums or even heavies. AVG Score could be improved by increasing the rewards for general actions like scouting, spotting or capture, just because lights are more capable to do so.

Edited by Tarteso, 11 December 2023 - 03:28 PM.

[Bud Crue]

Frost_Byte, on 11 December 2023 - 01:01 AM, said:

You can. I'm the owner of the spreadsheet he screenshotted, here's a link. Tabs 2 and 3 are what you want.
https://docs.google....dit?usp=sharing

We generate these for Cauldron on a quarterly basis, though this time someone asked I do a month capture to get a picture that has the loyalty mechs as well. As for balancing, this spreadsheet doesn't do as much as some would think. Cauldron balances based on power, not popularity. For example, the Executioner is one of the least used assaults yet it is ranked as one of the most powerful assaults. If you watched the MWO Championship Series this year, its usage rate was very, very high.

This is great!

Now as to balancing by power...I'm looking at you, Panther 10K and 10P.

[LordNothing]

Frost_Byte, on 11 December 2023 - 01:01 AM, said:

You can. I'm the owner of the spreadsheet he screenshotted, here's a link. Tabs 2 and 3 are what you want.
https://docs.google....dit?usp=sharing

We generate these for Cauldron on a quarterly basis, though this time someone asked I do a month capture to get a picture that has the loyalty mechs as well. As for balancing, this spreadsheet doesn't do as much as some would think. Cauldron balances based on power, not popularity. For example, the Executioner is one of the least used assaults yet it is ranked as one of the most powerful assaults. If you watched the MWO Championship Series this year, its usage rate was very, very high.

the exe really requires a special skill set to operate. im somewhat proficient with them, but they aren't my favorites. lots of masc and jumping to change directions rapidly and allow you to side poke with medium like speed. most players write it off as a knuckle dragger and pass on it, but thats a mistake.

[Curccu]

Tarteso, on 11 December 2023 - 03:24 PM, said:

Spoiler

lol, I already knew you would come here

Absolute numbers and averages are catchy, but don't tell the whole story (and they can be misleading sometimes). This is my quick modest analysis. Feel free to perform your own (or anybody), maybe even more sofisticated, if you care.

And before somebody try to start a flame war about lights, please check my mech utilization stats.

Source data: link above, sheet 1; added new column for the class group: L(ight), M(edium), H(eavy), AS(sault)
Methods: Kruskall-Wallis & Dunn's test, alpha 0.05
Results (simplified):

No significant differences in: Total drops, W/L, Survival %, and Popularity (All); Note: Popularity (all) means % mech use.


Significant differences found for:

1. K/D: AS=H; H=M=L; AS significantly better than M, L
2. Kills per match: AS significantly better than H, M, L; H=M=L
3. AVG Score: AS significantly better; H=M; L significantly the worse
4. AVG Damage: AS significantly better; then H; M=L

Conclusion: lights were as used (between 2023-11-01 and 2023-12-05) as the other classes, and had a similar survival %.

Also, there were significant differences in K/D, Kills per match, AVG Score, and AVG Damage, which are metrics more related to firepower; so, it makes sense. However, lights did the job as good as mediums or even heavies. AVG Score could be improved by increasing the rewards for general actions like scouting, spotting or capture, just because lights are more capable to do so.

Total drop %
Light 17,97
Medium 21,29
Heavy 27,90
Assault 32,84

How is 17,95% of total drops not significantly smaller than 32,84%?

And if we ignore significant part (and what is and isn't)
Lights have the worst... everything, not including survival %
that is the best, which makes sense as fastest and smallest mech can disengage and hide the easiest.

[Der Geisterbaer]

Tarteso, on 11 December 2023 - 03:24 PM, said:

lol, I already knew you would come here

While I in turn expected both such irrelevant "friendly jabs" as well as ...

Tarteso, on 11 December 2023 - 03:24 PM, said:

Absolute numbers and averages are catchy, but don't tell the whole story (and they can be misleading sometimes).

... mentioning trivialities that show that you didn't even bother to fully read what you quoted me on, because absolute numbers were only a fraction of what I presented there. But's what's even more important: If you really think that absolute values and 3 different types of averages (each for different views on the same data) are misleading here, then you'll have to properly explain why they don't work here.

Tarteso, on 11 December 2023 - 03:24 PM, said:

This is my quick modest analysis.

"Modest"? What has modesty to do with any of this?

Tarteso, on 11 December 2023 - 03:24 PM, said:

Feel free to perform your own (or anybody), maybe even more sofisticated, if you care.

I already gave you 3 different statistically correct forms of averaging on the key number concerning actual mech usage. That should be "sophisticated" enough until you manage to explain why those numbers I presented do not show that Light mechs and Medium mechs are the least played weight class and why you think that there is no correlation between those numbers and relative power.

Tarteso, on 11 December 2023 - 03:24 PM, said:

And before somebody try to start a flame war about lights, please check my mech utilization stats.

A pre-emptive red herring to counter a non-existent ad hominem. Relevance?

Tarteso, on 11 December 2023 - 03:24 PM, said:

Source data: link above, sheet 1; added new column for the class group: L(ight), M(edium), H(eavy), AS(sault)
Methods: Kruskall-Wallis & Dunn's test, alpha 0.05
Results (simplified):

No significant differences in: Total drops, W/L, Survival %, and Popularity (All); Note: Popularity (all) means % mech use.

May I ask what would constitute "a significant difference" in your mind?
Total drops: 17.9% Lights vs. 21.3% Mediums vs. 27.9% Heavies vs. 32.8% Assaults.

But hey, it's totally "modest" and particularly "honest" to chose a particular statistical method that doesn't actually deal with what you're allegedly trying to look for: Disproportionate usage numbers between the groups themselves (with an underlying correlation based an relative power).

Tarteso, on 11 December 2023 - 03:24 PM, said:

Conclusion: lights were as used (between 2023-11-01 and 2023-12-05) as the other classes, and had a similar survival %.

Conclusion: With an deliberately chosen method that isn't actually suited for the kind of analysis you'd be looking for you managed to conclude that a weight class that saw almost 50% lower usage than the most used class to be "as much used as the other classes". I guess congratulations are in order here.

[edit]Even more so as your conclusion ignores the results where even your own methodology found significant differences and put Lights at the low end.
[/edit]

Edited by Der Geisterbaer, 12 December 2023 - 04:19 AM.

[Aidan Crenshaw]

Never believe a statistic you didn't fake yourself

[Tarteso]

Curccu, on 12 December 2023 - 01:31 AM, said:

Total drop %
Light 17,97
Medium 21,29
Heavy 27,90
Assault 32,84

How is 17,95% of total drops not significantly smaller than 32,84%?

And if we ignore significant part (and what is and isn't)
Lights have the worst... everything, not including survival %
that is the best, which makes sense as fastest and smallest mech can disengage and hide the easiest.

In short, what the performed statistical analysis says is that any light mech was chosen as frequently as any other mech from a different class (talking about drops and the related popularity).
Problem with absolute numbers and averages taken alone is that you miss the data variability, which is precisely when statistics takes part to see if any given effect is (likely) real or casual.
But forget about stat methods: please, try representing data using a box (and whisker) plot instead averages plot. Quite a different picture, isn't it? Well, I hope so.

Edited by Tarteso, 12 December 2023 - 11:10 AM.

[Tarteso]

Der Geisterbaer, on 12 December 2023 - 04:00 AM, said:

While I in turn expected both such irrelevant "friendly jabs" as well as ...



... mentioning trivialities that show that you didn't even bother to fully read what you quoted me on, because absolute numbers were only a fraction of what I presented there. But's what's even more important: If you really think that absolute values and 3 different types of averages (each for different views on the same data) are misleading here, then you'll have to properly explain why they don't work here.



"Modest"? What has modesty to do with any of this?



I already gave you 3 different statistically correct forms of averaging on the key number concerning actual mech usage. That should be "sophisticated" enough until you manage to explain why those numbers I presented do not show that Light mechs and Medium mechs are the least played weight class and why you think that there is no correlation between those numbers and relative power.




A pre-emptive red herring to counter a non-existent ad hominem. Relevance?


May I ask what would constitute "a significant difference" in your mind?
Total drops: 17.9% Lights vs. 21.3% Mediums vs. 27.9% Heavies vs. 32.8% Assaults.

But hey, it's totally "modest" and particularly "honest" to chose a particular statistical method that doesn't actually deal with what you're allegedly trying to look for: Disproportionate usage numbers between the groups themselves (with an underlying correlation based an relative power).



Conclusion: With an deliberately chosen method that isn't actually suited for the kind of analysis you'd be looking for you managed to conclude that a weight class that saw almost 50% lower usage than the most used class to be "as much used as the other classes". I guess congratulations are in order here.

[edit]Even more so as your conclusion ignores the results where even your own methodology found significant differences and put Lights at the low end.
[/edit]

It is clear that you, "sir", have no idea what you're talking about. You "know" nothing about stats, given your "sophisticated" "analysis". Mine, even "modest" (does this mean the same for both us?) or even wrong (chose your own, but proper, method and probe it) was done considering "things" you have no idea about.
However, you have taken your time to write a dismissive wall of text, about a subject you have no idea, trying to disregard my "modest" analysis with your simple&clumsy, yet exquisitely elaborated, thoughts.
And, of course, as you have no idea about stats, you keep pushing your thoughts (in a rather rude tone) instead of doing some basic research. But you don´t know that things aren't so simple because you have no idea.

Your "passionate" defense of poor lights isn't a surprise, given your "impressive" 86% overall use (chaining several 100% use season after season). Your tone is, on the contrary, surprising to me, given that my ultimate purpose was just to be helpful by shedding some "light" about the shared data.

Good afternoon and good bye.

Edited by Tarteso, 12 December 2023 - 11:16 AM.

[feeWAIVER]

Tarteso, on 12 December 2023 - 11:09 AM, said:

Good afternoon and good bye.

Lmao Dodgeball Champ 2023.

[Curccu]

Tarteso, on 12 December 2023 - 09:57 AM, said:

In short, what the performed statistical analysis says is that any light mech was chosen as frequently as any other mech from a different class (talking about drops and the related popularity).
Problem with absolute numbers and averages taken alone is that you miss the data variability, which is precisely when statistics takes part to see if any given effect is (likely) real or casual.
But forget about stat methods: please, try representing data using a box (and whisker) plot instead averages plot. Quite a different picture, isn't it? Well, I hope so.

OK.. sorry it might be because I'm not native English speaker and/or because I have read D&D manuals past 3 hours...
How are lights chosen as frequent as other class of mechs if they have clearly least drops of all weight class?
Almost every third drop is in assault mech.

[ScrapIron Prime]

I see what he's trying to say, but he's got statistical analysis wrong.

LIGHTS are less used than any other class, but Locusts are used more often than Atlases. I believe he's equating different kinds of fruit there.

[Tarteso]

Curccu, on 12 December 2023 - 12:22 PM, said:

OK.. sorry it might be because I'm not native English speaker and/or because I have read D&D manuals past 3 hours...
How are lights chosen as frequent as other class of mechs if they have clearly least drops of all weight class?
Almost every third drop is in assault mech.

Nice to know you are a D&D fan. Oh, what a good times...

Ok, in short again: light class have less mechs than M, H, AS classes, so YES, the light class was picked less times overall in that time lapse. This agrees with the good correlation you can find (r= 0.79, p=0.2) between class % use and mech number per class. This is not perfect, however, because mechs are not chosen randomly: there are dedicated (or near to) players to 1, 2 or 3 classes, as well as players switching from one mech and class to other, depending on many factors. Also, a lot of new toys (assault mechs) have been added recently to the game so, who knows...

However, as Scraplrom Prime said, a given mech from a given class can be selected more, and less, than another given mech from another given class (the spreadsheet is full of examples). This makes that data distribution vary a lot within each class (you can get a picture by representing data or using a box plot as I suggested). So, speaking about drops, "no data point clouds are clearly higher" over the rest, to say it a bit too simple but clear. Some stat. methods analyze this dispersion along with the mean, ranks, etc. and let you to know if any group is "significantly" (lets say "clearly") different.

This is why a simple analysis can lead to confusion and you should analyze data carefully.

Hope it helps.

Edited by Tarteso, 12 December 2023 - 04:06 PM.