Let's talk esports

esports
ask

#1

Hi. I’m with DOJO Madness, managing an analysis platform for Counter-Strike. I come from a programming background, have been around esports in various capacities for over a decade now, and am now hoping to start connecting with the larger sports analytics community, now that I’ve kinda unwittingly crashed your party.

I know there’s a lot of FUD around esports, and there’s more misinformation floating around about this scene than you could shake a stick at. So as my first post here I wanted to open up the phone lines. Hit me with any esports questions - nothing is too small!


#2

Can you discuss an example of metrics Counter-Strike players are looking at to gain an edge? Is there an Actions Per Minute (APM) type of metric* for Counter-Strike, a simple thing that players are looking to get better at? How does your platform help train players?

*APM - The number of actions delivered to units in Starcraft and similar games, per minute. Beginners aren’t issuing, a lot of orders; experts have a high APM which increases the efficiency of their armies. The term is also used in fighting games.


#3

Can you discuss an example of metrics Counter-Strike players are looking at to gain an edge? Is there an Actions Per Minute (APM) type of metric* for Counter-Strike, a simple thing that players are looking to get better at?

Not currently. CS may be the odd-esport-out in this regard. But we may also just haven’t discovered such metrics yet, as my project is the first to really attempt to expand data collection beyond who won which round and who killed who.

But aside from honing individual skill in using the various weapons, CS is such a tactics/teamwork oriented game that individual performance indicators are often not correlated with match outcomes; or in other words, a player can have garbage stats in a match but still had a tremendous impact on the match. I could be wrong, but I wanna say that volleyball may be roughly analogous in this regard.

How does your platform help train players?

The platform as it exists today is best utilized in scouting an upcoming opponent’s play style on a particular map. In contrast to ‘MOBA’ games like Dota and League of Legends where teams will generally formulate a set of gameplans they employ and then execute on those plans regardless what the other side is doing, CS requires constant adjustment to the style and strengths of the opposing side to be successful. It’s one of the big reasons I love the game.

So the platform allows teams to efficiently review scores of rounds at a time, generate heatmaps, etc, so they can gain a clear sense of an opponent’s tendencies heading into a match, so they can employ tactics to most effectively counter an opponent’s preferred approaches.


#4

In what ways can players have a tremendous impact on a match but still have low ADR, HLTV Rating, etc?


#5

The two most popular metrics are Average Damage Per Round (ADR) and HLTV’s Rating metric


#6

Being an effective in-game-leader but not getting a lot of frags, using utility to maximum effect in setting up kills for teammates, getting dunked on most of the match but for some reason you win your team a few rounds in clutch situations - just a few. You can also have great ratings but had a low impact on a match, for instance getting most of your kills in situations where the probability of winning the round was low and you didn’t win the round.

ADR is pretty correlated with kill count, and speaks nothing to the quality of the kills; kills in an even or shorhanded situation (2v4, 3v3, etc) that lead to a round win can be said to have impacted the result of a round and thus the match more than a 5-kill spray down against an eco. A particular player might receive more pressure simply due to their opponents gameplan and so has more opportunities to boost ADR than a teammate defending the other side of the map.

HLTV ratings (both old and new) are highly correlated with kill-death ratio; and so while these metrics are interesting I don’t think they correlate as highly with actual impact to a match than is commonly thought. Old HLTV rating also arbitrarily put tremendous value on multi-kill rounds with no regard to context.


#7

i was reading about he hltv rating which includes things like opening kills and I’m wondering has there been any studies done showing the correlations between different metrics and winning or damage dealt or something of that nature


#8

Probably some neophyte level questions from me:

Would there be value in tracking and collecting accuracy data and shot placement data? Ex: a player has a higher accuracy rating and damage output with a specific SMG vs another rifle? From passive observations there’s an emphasis on roles, right?


#9

Not anything in terms of detailed papers, no. Best I can muster is a couple tweets trying to take stabs at what’s behind the HLTV 2.0 metric, since they published the exact methodology behind their initial rating, but have withheld details about their new metric. https://twitter.com/rxcs/status/895139882569351169 https://twitter.com/rxcs/status/895125093956730880


#10

No question is neophyte :slight_smile:

First, on roles: there are definite roles that teams must fill in order to be successful. Besides the ‘in game leader’/IGL (read: quarterback), you need an ‘AWPer’ who is primarily good with scoped single-shot rifles, an ‘entry fragger’ who is primarily good with the automatic rifles and is just a naturally good aimer, and you need a ‘support’ who is primarily good at setting up plays with well-thrown grenades. At the professional level, most all players are good in any role, but are just fractionally better at one aspect than 99.9% of the player base, and it makes all the difference.

Regarding shot placement and accuracy tracking, that’s a bit of an open question. We’re just beginning to work on how best to collect and use such data on our platform. There’s the potential for a lot of noise in such statistics as not all shots taken are necessarily at a known target; an attacker may ‘prefire’ around a corner having already lined themselves up towards a known common defensive position in an attempt to mitigate a stationary defender’s advantage without there being anybody there, or just pummel a permeable wall (commonly known as ‘spam spots’) playing the odds there might be someone on the other side that catches some damage, or just garden variety suppressing fire. All things that can pull a basic accuracy stat out of alignment. We might have some ways of filtering out these noisy cases, but just hypotheses right now.

Generally speaking, weapon choice in any given round is dictated most by how the game is balanced and the team’s economic situation in the round. The ‘holy triad’ of AWP-M4-AK has basically dominated CS since the beginning, and most weapons are underpowered to the point they barely if ever show up in competitive play. Typically if an odd weapon becomes in vogue for a time, it’s due to an inadvertent change on the developer’s part, usually some bug fix having an unintended side effect, so is hot for a couple months until it’s patched back over. If a team’s economy dictates that they need to empty their accounts and buy whatever they can get, because they need to win a round to stay comfortably in a match, then players will typically have a personal preference of what they buy in such situations, regardless of relative efficacy vs rifles.


#11

I think this would be a good place to start then with obviously a lot of trial and error to figure out which statistics actually are telling us something. I feel like there may be a lot of low hanging fruit to be gained. Do you also have the stats differentiated by game state such as even strength , 5v4, 3v4 and such. Also what about weighting for which side player is on? I would assume some maps are easier for CT than T and vice versa. Either way I’m really interested in this as a former cs player and would love to help out