Statistics for Melee

[JKJ]

I have been thinking for a long time now that gathering statistics for Melee would be of high value to the community, and would add another layer of depth to understanding players in all their strengths and weaknesses.
And, seeing as I actually have some free time for a little while now, would like to possibly undertake the gathering of these statistics.
What I'm wondering is what statistics should I gather, and how far back in time should I go? Also, what players should I gather statistics for?

I'm thinking right now that anyone who places top 8 at a regional or national consistently should have statistics drawn up for them, and I was planning to use any recorded video spanning from the first ROM onwards.

As for which statistics to gather, I'm thinking right now that I would like to gather stats for the following:

-Edgeguard Percentage
-Average Percent Reached (Survivability)
-Gimps Per Game (defined as a kill below 50%)
-First Hit Percentage
-Comeback Percentage (number of times coming back to win a match when down by at least two stocks)
-Average Match Time
-Percentage of Wins on Counterpicks (both the player's counterpick and the opponent's counterpick)
-Most Frequent Counterpicks vs Specific Characters
-Win/Loss Record in sets vs Top Players
-Average grabs per game
-SD's per game
-Successful Techchase Percentage
-Combo Percentage (as in how many combos are successful out of how many combos are started)
-Percentage of set wins after losing game 1
-Percentage of set wins after losing on player's own counterpick

I kinda also want to make a WTF Moment statistic, but don't know if anyone would take it seriously. As in WTF Moments per game, for example M2K's waveland nair vs Mango at Apex 2013, game one of Loser's Finals, or his double counter in the same set, last game. Or Armada's stitchface combos, any of them. Or Mango's shine nair on Hax at Apex 2013. They are easy to see, hard to quantify.

Please reply with which players you think I should gather statistics for, which statistics of these/statistics I didn't think of I should gather, and how to determine certain stats, like Combo Percentage or Techchase Percentage (What qualifies as a successful techchase?), and/or how far back the stats should span.

Thanks in advance! Looking forward to making this happen, hopefully.

 

[BTmoney]

Information is cool dude. I support this

 

[Pengie]

One of the most interesting statistics imo is hit percentage; i.e. percentage of moves that hit the opponent out of attacks done. This one is actually really cool because the game keeps track of hit percentage for you in the bonuses screen. Every now and then I like to check it to see how high it is and generally, I'd say that this is a good way to focus on fixing certain issues in your play because if you have a high hit percentage, it stands to reason that you aren't doing very many extraneous moves so if it's really low you can try to focus on streamlining your gameplay and getting rid of any unnecessary stuff.

 

[JKJ]

@Pengie
Hit percentage, while cool when seen from that angle, can also be deceiving. Many players do moves specifically to not hit, but to control space. Hungrybox is a prime example of why the hit percentage stat is not useful in determining how "streamlined" someone's play is in all cases. Using moves to deliberately control space is very common, and therefore undermines the usefulness of hit percentage.

 

[BTmoney]

One of the most interesting statistics imo is hit percentage; i.e. percentage of moves that hit the opponent out of attacks done. This one is actually really cool because the game keeps track of hit percentage for you in the bonuses screen. Every now and then I like to check it to see how high it is and generally, I'd say that this is a good way to focus on fixing certain issues in your play because if you have a high hit percentage, it stands to reason that you aren't doing very many extraneous moves so if it's really low you can try to focus on streamlining your gameplay and getting rid of any unnecessary stuff.

True!!!

Look at Mango vs Armada @ Evo he was especially accurate. That's so important with Fox

@Pengie
Hit percentage, while cool when seen from that angle, can also be deceiving. Many players do moves specifically to not hit, but to control space. Hungrybox is a prime example of why the hit percentage stat is not useful in determining how "streamlined" someone's play is in all cases. Using moves to deliberately control space is very common, and therefore undermines the usefulness of hit percentage.

Well obviously it depends on the character and to a certain degree the player. Breaking it down by character would be fine

 

[JKJ]

I think that a more useful statistic than hit percentage would be hit conformation percentage, the amount of opportunities converted off of hits

 

[tarheeljks]

i've flirted with this many times. just such a huge chore to do for a single match/set, much less to aggregate data with meaningful volume. i settled on matchups, stages, stock differential. haven't done it in a while though

 

[Pengie]

@Pengie
Hit percentage, while cool when seen from that angle, can also be deceiving. Many players do moves specifically to not hit, but to control space. Hungrybox is a prime example of why the hit percentage stat is not useful in determining how "streamlined" someone's play is in all cases. Using moves to deliberately control space is very common, and therefore undermines the usefulness of hit percentage.

Yeah, I should have probably mentioned that this is pretty character dependent. Fox and Marth I think benefit the most from keeping track of this because they're characters that want to be focusing on pressing buttons only when necessary because of how great their movement games are, and I think that one of the biggest hurdles that hold lots of Foxes and Marths back is that they do lots of extraneous stuff and press lots of buttons that don't accomplish anything. Keeping track of hit percentage gives the players a good way of seeing if what they're doing is necessary and can help promote more conscious and thoughtful decision making.

 

[-Ran]

I once broke down one hundred Brawl matches of top level players to determine how viable it was to come back in the game. If I recall, around eighty-percent of the time, the match was decided by the first kill if a quick revenge kill didn't occur. It'd be interesting to see what the static would be for Melee.

 

[JKJ]

Well since no one is suggesting new/different stats, I'm going to attempt to do this for Mango, since he just won EVO.
I'm going to use all recorded tournament sets I can find since ROM1. I'm going to include the WTF Moment Statistic as in WTF Moments per game, just because I think it's fun.

 

[tarheeljks]

Think you need some guidelines for the combo/tech chase stats for stuff like when a combo has been dropped vs when it was abandoned. This is subjective of course, but it's important to know the specifics of the data collection process.Maybe also an additional statistic incorporating damage to give some sense of the tradeoff.

Edit: I'd track grab % or success rate or w/e in addition to total grabs. Maybe also separate them into shield grabs vs other grabs. Also, how about conversion rates/damage stats on chaingrabs where applicable?

Countepart data would also be pretty useful for all these stats, by which I mean each stat applies twice. So if you combo me to 30% that goes into your "combo damage given" stat and into my " combo damage taken" stat. Of course, such a stat would need to be cross referenced with character matchups. That's generally true, as has been noted, but is esp true there.

 

[JKJ]

Think you need some guidelines for the combo/tech chase stats for stuff like when a combo has been dropped vs when it was abandoned. This is subjective of course, but it's important to know the specifics of the data collection process.Maybe also an additional statistic incorporating damage to give some sense of the tradeoff.

Edit: I'd track grab % or success rate or w/e in addition to total grabs. Maybe also separate them into shield grabs vs other grabs. Also, how about conversion rates/damage stats on chaingrabs where applicable?

Countepart data would also be pretty useful for all these stats, by which I mean each stat applies twice. So if you combo me to 30% that goes into your "combo damage given" stat and into my " combo damage taken" stat. Of course, such a stat would need to be cross referenced with character matchups. That's generally true, as has been noted, but is esp true there.

I'm definitely in need of guidelines on the combo/techchase dropped/successful statistic. I'm not sure what guidelines would work, though, and I'm looking for insight.
I wouldn't know where to begin with the stat you listed in the bottom paragraph, much less cross referencing it with character matchups. Im probably not gonna do that for the sheer amount of work it would take. The other stats take a lot of work as it is, but that stat would probably kill me.

 

[Landry]

The biggest problem I see with this is that some matches (early in bracket, pools, etc) will completely throw off your averages. I'd say you should probably only record statistics from winner's semis/loser's quarters on.

And as other people have mentioned, character is going to be very important. You'd have to do different stats for M2K's Marth and Sheik, Mango's Fox and Falco, etc.

It's a cool idea but it seems like a huge amount of work to me. If you're up to the task then more power to you.

 

[Problem2]

I once broke down one hundred Brawl matches of top level players to determine how viable it was to come back in the game. If I recall, around eighty-percent of the time, the match was decided by the first kill if a quick revenge kill didn't occur. It'd be interesting to see what the static would be for Melee.

oof!! That's a rough statistic. It's interesting because it sure feels like a mountain to climb after you lose your first stock, and then you put manage to put a number to that feeling and it looks surprisingly accurate.

 

[tarheeljks]

I'm definitely in need of guidelines on the combo/techchase dropped/successful statistic. I'm not sure what guidelines would work, though, and I'm looking for insight.
I wouldn't know where to begin with the stat you listed in the bottom paragraph, much less cross referencing it with character matchups. Im probably not gonna do that for the sheer amount of work it would take. The other stats take a lot of work as it is, but that stat would probably kill me.

Yeah, I completely understand. Just tossing out ideas, feasibility be damned lol.

@Landry, agree. I tracked all sets from winners quarters on, any sets with ranked players, and then some sets with players who were good, but not ranked for w/e reason

 

[kd-]

Kill move percentage (How many times was bair used to kill, upair, upsmash, etc.)
Average percent reached for each of Mango's stocks
Average damage done for each of Mango's stocks
Average kills for each of Mango's stocks
Average number of hits per kill, or his running total of hits per kills
Number of different recoveries per game (side b to ledge/stage/platform, fireanimal to ledge/stage, airdodge to ledge/stage)
Edit: Number of deaths from sides, bottom, and top
Average time spent on each of Mango's stocks

I wouldn't mind seeing all the statistics for just Mango's Evo top 8 alone, and maybe compared to the others in top 8 for throughout Evo

 

[JKJ]

Kill move percentage (How many times was bair used to kill, upair, upsmash, etc.)
Average percent reached for each of Mango's stocks
Average damage done for each of Mango's stocks
Average kills for each of Mango's stocks
Average number of hits per kill, or his running total of hits per kills
Number of different recoveries per game (side b to ledge/stage/platform, fireanimal to ledge/stage, airdodge to ledge/stage)
Edit: Number of deaths from sides, bottom, and top
Average time spent on each of Mango's stocks

I wouldn't mind seeing all the statistics for just Mango's Evo top 8 alone, and maybe compared to the others in top 8 for throughout Evo

I'm thinking of doing a preliminary bout of statistics going only on EVO, and hey, maybe top 8 is a good place to start. I'm gonna do this sh**.
I've decided on the following stats:

Average % reach for each stock
Average % dealt for each stock
Average kills for each stock
Number of different recovery mixups per game
Percentage of deaths from side blast zones, bottom blast zone, and star KOs
Edgeguard Percentage
Gimps Per Game (defined as a kill below 50%)
First Hit Percentage (for stocks and games, percentage of time the first hit goes to player)
Comeback Percentage (number of times coming back to win a match when down by at least two stocks)
Average Match Time
Percentage of Wins on Counterpicks (both the player's counterpick and the opponent's counterpick)
Win/Loss Record in sets vs Top Players
Grab success %
Average SD's per game
Average Strong Combos per game (defined as a string of hits dealing over 40% or resulting in a kill)
Percentage of set wins after losing game 1

I'm gonna start with Mango for EVO top 8 only to see how it comes out.

 

[Bones0]

I would advise heavily against any subjective statistics. They introduce bias no matter how hard you try. Even something like SDs can be controversial. If someone jumps off stage and gets edgehogged, is that an SD or is it a gimp? If the same situation happens and the opponent hits them once unnecessariily before edgehogging, does that change whether it was an SD or not? If you make judgement calls on all of these situations, you will inevitably skew the data. Your perception of whether something was a combo or just a separate followup can quite drastically change the number of combos you would count for any given match. I would cut down your list above to this:

Average peak % per stock
Average dealt % per stock
Average KOs per stock
Percentage of KOs and deaths for each of the 3 blast zones (side, top, bottom)
Gimps per stock (defined as a kill below 50%) [use per stock instead of per game since getting 4 gimps without losing 1 stock is obviously way different than getting 4 gimps spread across 4 stocks]
First KO percent [first hit doesn't really indicate all that much in the grand scheme of things, it gets messy with trades, and is skewed heavily by where the players spawn]
Average match length
Average stock length [simply divide the previous statistic by 4 if they lose; if they win, mark the time they lost their previous stock at and divide by however many stocks they lost]
Win percentage for game 1 and each player's counterpicks
Win/loss records vs. top players
Average grabs per stock
Average ledge grabs per stock [ignore ledge grabs done between an opponent's death and them leaving the respawn plat]
Rate of Avenger KOs (KOing the opponent after respawning without taking any damage)
Percentage of set comebacks (winning the set after losing game 1)
Favorite move (move used most often)
Favorite KO move (move that earned the most KOs)
Number of shield hits [multihit moves only count as 1 shield hit, ofc]

If you want to go really hard, you could tabulate how many times each move was used and the success rate of how often they hit body, shield, or whiffed.

 

[chenjesu]

wouldn't mind seeing a statistic for what actions people do out of shield, and how often they get hit out of them.

 

[JKJ]

So, before I read Bones' post, I already spent a few hours going through top 8 of EVO for Mango.
I omitted certain statistics because they felt sketchy; most wound up being the subjective stats to which Bones referred.

Here they are!

Mango, Top 8 EVO 2013

Average Percent Reached: 121.2105%
Average Percent Dealt: 139.8936%
Average Kills Per Stock: 1.0961
Edgeguard Percentage: .4516
Percentage of Wins on Mango's counterpick: 1.00
Percentage of Wins on opponent's counterpick: .875
Grab Success Rate: .84931
Average Time Per Game: 3m:50.067s
Average Time Per Stock: 57.516s

 

[Ussi]

How would Mango have a 100% win rate on his counterpick when he got 2-0'd by Wobbles in winners bracket?

 

[Ziodyne]

How would Mango have a 100% win rate on his counterpick when he got 2-0'd by Wobbles in winners bracket?

It was a 2-1

 

[Ussi]

Mango OP then, my bad

 

[tarheeljks]

Disagree very strongly about not attempting to track subjective stats. As long as you provide a clear description of your method so that anyone can replicate the process and compare results, or critique the method of course, it's fine. Any inference from said statistics is obv more tenuous, but that doesn't mean there is no value in their collection

 

[tarheeljks]

My bad, double post

 

[thespymachine]

Man, if Smash got only half as deep as baseball does with its stats, that would be crazy.

Something that came to mind:
Average Percentage at Death/KO - though this could show how efficient a player is at KOing their opponent and how good their DI, etc. is, this would mostly be used for MU statistics, I would think. My reasoning is that each character has very specific KO- and survive-ability in relation to the character they are playing against and the stage that they're on - so these stats would be much more useful in relating how players perform in certain MUs (or versus a certain player's character). Regardless, I still think it's good to keep track of.


I think that keeping track of what a player does/uses to KO or what the opponent did/used to kill the player is far more important (relatable to your "Edgeguard Percentage" and "Gimps per Game" stats).
So, the stats I'm interested in are: Location of Direct KO/Death (Top, Left/Right, or Bottom) and Type of KO/Death (Standard**, Gimp/Edgeguard, Edgehog, etc).

*Direct vs Indirect KO/Death - Direct would be when a KO/Death happens when the player has no chance of attempting a recover (due to hit stun, or whatever); indirect is when the player is sent off-stage and are able to attempt a recovery but don't make it back to the stage. (So, I'm interested in "Direct" only, as far as location goes, because all "Indirect" KOs/Deaths will be at the bottom - unless the player decides to use their recover attempt as way to kill themselves faster by going to one of the side killzones) Up for debate.

**Standard KO/Death - Tough to define, but it's basically any on-stage, non-combo KO/Death. Up for debate.

 

[Belgian Warrior]

I think the idea is very interesting but we have to be organized and a lot to collect all these stats. I guess the best is, at least at the beginning, to collect these news on EVO 2013 fr the TOP 8 only and to classify every idea in categories. For instance, basic to in-depth things, we can put the datas of JKJ in a basic category, those of thespymachine in the in-depth class, create others like miscellaneous (for instance percentage of succesful hits) and then analyse the sets or even the match-ups (here goes the last category but the most subjective). I'm pretty certain that if we are looking for this percentage in a match like Mew2King (Marth) vs Hungrybox (Puff), we'll see that none of the two players will exceed 20 % as much as possible.

@thespymachine. Does your avatar come from the movie Notting Hill ?

Sorry, it comes from 2001 : A Space Odyssey but it's shocking like a scene of Notting Hill has got a similar pic.

 

[thespymachine]

Actually, I'd like to see the statistics of Mango vs PPMD throughout this year (Zenith and EVO), if you wouldn't mind.

 

[JKJ]

Hmmm

I will see what I can do, @thespymachine

 

[MountainGoat]

Like this idea. Huge pain in the ass and time consuming though. Good luck.

Looking forward to comparing the top players in very specific areas. I agree PP vs Mango would be a good place to start since they both play spacies.

I think win percentages by matchup and by map would be really interesting.

 

[thespymachine]

This thread has been in the back of my mind since I first posted, and have never found/made the time to come back and check what was going on. Which is nothing, lol.

I'm going to watch The Big House 3 and do some stats on top 8 Melee. Hopefully by the end of the day I'll have the stats up.

 

[JKJ]

Dude you should do this
I just don't have time, I work and have school and tons of other ****, but it was really cool and rewarding to do.
Don't count on all top 8 results done by the end of the day tho.
this ****s hard dude

 

[thespymachine]

I'm doing very basic stuff, like: % at SL (Stock Loss), match time, stages, characters, first hit (projectile and physical), KO move, KO type (when the KO move came in: raw, end of combo, gimp/edgeguard, edgehog), location of SL (top, left, right, bottom). Then from those I'll do some average statistics and other things. Edit: Also, I'm keeping track of types of SDs as well (purposeful - the Mango, accidental - the Boback, circumstantial - being baited into airdodging only to realize that you are going to SD now [this is very boarderline, and I realize that, but I'm keeping it SD for now]).


I'm actually more focused on how to organize a stat-keeping page (ie, scorecard).
I used to do A LOT of stat-keeping in baseball, this is what their scorecard looks like:

I think by having a basic, organized (even standardized) scorecard it'll make way for great future stat keeping.

I'll post my results as soon as I can.
Watching now.

 

[JKJ]

awesome dude.
youd be like super champ status

 

[thespymachine]

Okay. Only did one match, since I spent most of my time trying to figure out the template of the scorecard.

["Prj FH" and "Phy FH" are supposed to be black, oops]

And then the only game I scored:

Explanations:
SL = Stock Loss (KO or SD)
Type = KO: Combo Ender (Combo), Edge-Guard/Gimp (EG), Edge-Hog (EH), and Raw (which would be any on stage, neutral game KO that isn't a combo, like a KO from a read or random attack)
= SD [none shown]: Purposeful (Pur), Accidental (Ac), and Circumstantial (Cir) [Like if you are baited into airdodging and fall to your death, or something]
% = Percentage SL occurs
Place = Placement of SL on deathzone: Left, Right, Top, Bottom
Move = Attack that caused SL (only applicable to KOs)

Brkt Spot = Part of Bracket and which Round the game occurs
Prj FH = Projectile First Hit
Phy FH = Physical First Hit


What is included in the scorecard and what means what can be debated, but I decided to make a first attempt at things (in a single game) that can be kept track, roughly, as the game was being played.

What do you think so far?

Edit: My first observation is that there could probably be some color or better separation in the scorecard. Looking at the baseball scorecard, which has a lot of things that you just check off (instead of write in), I could put at least four categories into a check-off format easily: SL, Type, Place, and CP/Strike.

 

[JKJ]

This is so godlike
thespymachine for best creator of stat sheets ever
whatd this take you like a minute YEEZUS dude you're awesome
run with this **** dude so much props

 

[thespymachine]

ver. 2
eta 10 min.

 

[JKJ]

My god
the dedication
I commend you

 

[nooga]

itshappening.gif

 

[thespymachine]

Took my initial observations and applied them.

Changes:
- Made some earlier write-in categories (SL, Type, Place, and CP/S) into check-off categories.
- Moved % down a spot to keep all the check-off categories together.
- Added a spot to put the "Best of ____" for the set.

Overall, I like the improvement, it looks more compact.


[Edit: fixed original image, it was missing %]


And here's a Best of 5 set version (if the players stayed the same characters throughout that is):

Okay, I'm done now. lol