16 replies to this topic

[dubstep albatross]

How often are stomps actually close?

TL;DR Not often. It is nineteen times more likely that a stomp is actually a stomp and not a close match. In other words only about five percent of stomps are a close match. Scroll to the bottom for a colorful picture and a spreadsheet.

PREAMBLE

I was motivated by a post (https://mwomercs.com...ost__p__6450147) in the "Do Assaults Push First?" thread.

pbiggz, on 25 March 2022 - 10:01 AM, said:

the amount of "stomps" people experience, i suspect, is vastly lower than they think, because, even a 12/0 stomp might be 12 kills vs 0 kills, but the winning team has 15% total armor left. All things considered that's actually a pretty close match, both teams probably won an even number of trades, until the end, the end game screen just doesn't show it.

DEFINITIONS

• Match: A 12v12 game (no data from the recent 8v8 matches was used)
  
• Stomp: a match where there are no survivors on the defeated team and the victory team took three or less losses (12-3 or better)
  
• Damage done ratio: the ratio of total damage done by one team compared to the other; defeated team is the numerator, victory team is the denominator
  
• Mostly even: the defeated team had a damage done ratio of at least 0.85 (not more behind than 15%) and translates roughly to a 12-8 game
  
• Close match: one where the teams traded "mostly even"
  
  • Note that by these definitions, stomp and close match are not mutually exclusive
• Too many words: this post

TAKEAWAYS

• 26% (one in four) of 1,588 matches were considered a stomp
  
• Average damage done ratio (for the defeated team) is negatively correlated to the number of surviving victory team members
  
• Distribution of tonnage imbalance between stomps and non-stomps is very close, though for stomps the distribution may be slightly skewed in favor of heavier victory teams
  
• Very roughly, about half of non-stomp matches are "mostly even" (close) matches
  
• Victorious teams tend to have similar total damage done metrics irrespective of whether the game was a stomp and/or was close
  
  • This suggests that it is the defeated team that does or does not close the gap (but this could be because the winning team did not allow them to)
    
  • Damage done does have an upper limit (total armor and structure on team)
• It is unlikely anyone has gotten even this far in this post

CAVEATS

• It is next to impossible to reliably know how much armor (and structure) there is on any one team in any one match
  
  • Drop weights and team compositions can be wildly different
    
  • Something something mechlab
    
  • We can make assumptions such as average max armor/structure per category or we can even do some very detailed accounting of each variant and their max armors/structures, but there is a diminishing return here and the aggregates are in our favor for this analysis
• Total damage done by a team, on match-by-match basis, is not necessarily predictive of skill, success, or outcome
  
  • With the right shooter and the wrong shootee, damage done to destroy the mech can be shockingly low
    
  • On the aggregate there is likely just as much (if not more) sandpapering as there is surgical assassination
    
  • Matches do not always end in a total loss by one side (though a vast majority do)
• Damage done ratio is a proxy metric for how well the teams traded, with 1.0 being effectively even trading; it is not perfect and there are assumptions being made
  
• Data were collected from mostly Tier 3 games, with some Tier 4 and a very small number from Tier 5

CRAYON





SPREADSHEET




* several edits because whoops

Edited by dubstep albatross, 10 April 2022 - 09:28 PM.

[TheCaptainJZ]

Sounds plausible, though damage as a substitute for determining "close" is really more nuanced, but not something we can quantitatively describe so it's the best we can do (I wish we had paper dolls of damage for all mechs at the end).
MWO by it's nature often produces stomps. The main reasons? Numerous variables (mech variants, weapons, player skill, coordination, maps, etc., etc.) and no respawns.

[Nightbird]

Where did you collect the 1,588 matches of data? If you plan to reply from your own experience, you've wasted your time. What you experience differs based on your skill level and the data collected becomes entirely subjective.

Further reading if interested: https://mwomercs.com...thread-we-need/

[dubstep albatross]

Nightbird, on 04 April 2022 - 12:07 PM, said:

Where did you collect the 1,588 matches of data? If you plan to reply from your own experience, you've wasted your time. What you experience differs based on your skill level and the data collected becomes entirely subjective.

Further reading if interested: https://mwomercs.com...thread-we-need/

The data does come from my own games and I did call out what tiers the data came from. A claim could be made that my data are biased by my being a constant and thus not generalizable, but the data are not subjective. Some of the analysis is subjective, for sure. I did pick thresholds based on my personal experience in the game and clearly defined them.

As this is a post in a public forum, I am open to discussion. I have read that matchmaker post for 2019. How does it tie in to what I have posted?

This is a post about a game, to strangers on interwebz, of course it is a waste of time.

[DaZur]

Missing: Strength of force tipping point... In short, the number of combatant differential necessary to directly influence a "stomp".

Obviously there are outliers but in general In 12v12 the non-scientific number is 4 and in 8v8 it's 2.

The closer a team can maintain a +/- 3 or less (in 12v12) margin the closer the final match outcome ends up being.

[Nightbird]

dubstep albatross, on 04 April 2022 - 12:26 PM, said:

The data does come from my own games and I did call out what tiers the data came from. A claim could be made that my data are biased by my being a constant and thus not generalizable, but the data are not subjective. Some of the analysis is subjective, for sure. I did pick thresholds based on my personal experience in the game and clearly defined them.

As this is a post in a public forum, I am open to discussion. I have read that matchmaker post for 2019. How does it tie in to what I have posted?

This is a post about a game, to strangers on interwebz, of course it is a waste of time.

No, your data is subjective. Someone else with a different skill level will collect a different set of data such that if they applied the same analysis you did, they would come up with a different set of conclusions. The link I provided showed why.

Edited by Nightbird, 04 April 2022 - 01:05 PM.

[Bowelhacker]

Nothing triggers some people on here like a statistical analysis they didn't do themselves...

[pbiggz]

hello friends

[Nightbird]

Bowelhacker, on 04 April 2022 - 01:15 PM, said:

Nothing triggers some people on here like a statistical analysis they didn't do themselves...

Anyone who has been formally trained in statistics (or any other subject/profession) appreciates being told they've made a mistake. They might say Fuuuuuuuuuuu...... but understand it is part of the process.

Only shallow novices get truly upset.

[pbiggz]

dubstep albatross, on 04 April 2022 - 12:26 PM, said:

The data does come from my own games and I did call out what tiers the data came from. A claim could be made that my data are biased by my being a constant and thus not generalizable, but the data are not subjective. Some of the analysis is subjective, for sure. I did pick thresholds based on my personal experience in the game and clearly defined them.

As this is a post in a public forum, I am open to discussion. I have read that matchmaker post for 2019. How does it tie in to what I have posted?

This is a post about a game, to strangers on interwebz, of course it is a waste of time.

Perhaps having more people run the same test as you could overcome that issue.

See, im contributing instead of ******** on my fellow man. Take notes nightbird.

[Kanil]

We could also just consider how much the issue matters. This dude's forum post has some statistical flaws, oh noes.

For a while I tracked which maps and modes were voted over others, and I don't doubt that if someone else did the same, they'd come to a slightly "different set of conclusions", That said, I'm guessing my conclusion that skirmish usually beats assault in a head to head vote isn't at odds with most people's experiences.

[An6ryMan69]

Thanks for the analysis.

I suspect that the actual biggest factors in stomps are momentum and group drops. A battle can be sickeningly easily turned by four well organized pilots working in unison, regardless of anything else. Momentum can create a bit of an unstoppable wave that just eats up the opposition, even if it is frenzied and unorganized - think one team doing dealthball and already up three kills - nobody wants to be on the wrong side of that.

If those two factors combine - say an organized group wolfpack of fast moving mediums hits first and gobbles up a couple of mechs right away, its basically game over for the other team.

The other human factor to consider with all the talk about stomps is the tendency to be most emotionally affected by one sided ***-kickings. Many players can play several close matches but still walk away with that 0-12 stomp they suffered being what actually sticks in their brain.

NO stats were harmed in the making of this post, because stats are the Devil's work!

[Urbie Connoisseur]

Quote

Stomp: a match where there are no survivors on the defeated team and the victory team took three or less losses (12-3 or better)

26% (one in four) of 1,588 matches were considered a stomp

Data were collected from mostly Tier 3 games

Meanwhile, at least in my experience, in tier 5 nearly every single match is a stomp. I can probably count the number of fair games I've seen on my fingers out of my 254 matches so far.

[SFC174]

Just for reference I believe that PGI (aka Paul) defined stomps as 12-4 or worse when they reworked the matchmaker. I think the typical number PGI reported across all matches at that time was in the 30% range give or take. There are threads on it.

[dubstep albatross]

Marida Connoisseur, on 11 April 2022 - 12:37 PM, said:

Meanwhile, at least in my experience, in tier 5 nearly every single match is a stomp. I can probably count the number of fair games I've seen on my fingers out of my 254 matches so far.

It is interesting how different an experience people can have. I do not have much data from Tier 5 (I accelerate out of it quickly and do not underplay to artificially keep PSR low). What little data I do have (n = 131) suggests that I experience less stomps in Tier 5 than tiers 3 and 4. About 10% of the Tier 5 games I have data for are stomps, which is about half as often as tiers 3 and 4.

[dubstep albatross]

An6ryMan69, on 11 April 2022 - 09:39 AM, said:

I suspect that the actual biggest factors in stomps are momentum and group drops. A battle can be sickeningly easily turned by four well organized pilots working in unison, regardless of anything else.

I will say that I think groups can contribute to stomps in either direction. I have seen plenty of groups that put up the lowest scores on the team. I have put some thought into how to identify (and/or quantify) the effect of groups. Perhaps outsized contribution to team match score (or damage), and/or kills, could be used to identify stomps where group players likely contributed to the stomp more than non-group players. Perhaps I will look more closely into this.

My understanding is the matchmaker tries to put an even number of groups (perhaps players in groups?) on each side in the match. Chances are in any match there are grouped players on both sides of the fence and maybe even likely the same number of grouped players. Just identifying the proportion of winning teams with identifiable groups is not enough.

Some groups are obvious (shared tags), others are identifiable with a good degree of certainty (a lance of people in the same chassis and variant), and other groups are just not really identifiable unless you happen to drop with them repeatedly over a few games and recognize the clumping.

Quote

Momentum can create a bit of an unstoppable wave that just eats up the opposition, even if it is frenzied and unorganized - think one team doing dealthball and already up three kills - nobody wants to be on the wrong side of that.

The very nature of the game and its mechanics leads to a snowball effect. I think we all intuitively recognize this. If the opening moves of a match are highly asymmetric, it takes a disproportionate amount of effort for the team on the wrong side of that to recover.

I think some stomps are inevitable. Even if we had a (realistically) perfect matchmaker, there are just so many variables at play that the outcome is effectively stochastic. Based on the numbers in the analysis I shared in this thread, a stomp could be one in four matches (on average). I do not have data from other games to draw a comparison, so maybe this percentage is common or better than other games. Hard for me to say.

Quote

The other human factor to consider with all the talk about stomps is the tendency to be most emotionally affected by one sided ***-kickings. Many players can play several close matches but still walk away with that 0-12 stomp they suffered being what actually sticks in their brain.

In my opinion, this a non-trivial contributor to the topic. A genuinely-bad gaming experience such as a 12-0 stomp can cast a shadow on an otherwise decent gaming session. Thankfully (in my data at least), 12-0 stomps are very rare (1%) and 12-1 stomps are also rare (3%).

Edited by dubstep albatross, 14 April 2022 - 09:24 PM.

[Corbantu]

This may not be a big contribution to the discussion but a combination of human behaviors I've seen that contribute to a stomp is the combination of push and panic. When one side feels like it's time for the route they tend to rush fullstop. Because of this the defenders will panic and target the most threatening mech rushing towards them. But the thing is, the one rushing them most likely has the most remaining armor in the match, while their buddy who's pushing but more cautiously has no armor and a red core. I've been able to make a few kills at the end of a match while on the receiving end of a stomp by simply ignoring the most aggressive enemy.