Eco Advantage and Round Outcomes

Counter Strike is a game dominated by material advantages, and one of the key ways that is expressed is in the balance of the economy. The insight.gg Performance Above Expectation and Win Percentage Added Per Round statistics both use research into exactly how the economy effects winning chances in the game, but they don’t expose some of the underlying research.

Here I’m presenting some of the data that insight.gg’s unique performance metrics are based on, and how they influence round outcomes.

Continue reading “Eco Advantage and Round Outcomes”

AWP Buys and Maps

The AWP is probably the most iconic weapon in CS. Players specialise in it, teams don’t really feel they’ve got a proper buy unless they’ve got one, and players will eschew basic protection just to get their hands on one.

In this article I’m looking at where and how AWP buys vary across sides, maps and teams, using data from the top level Lan tournaments in 2017. We’ll start with the very basics.

T vs CT buys

First let’s take a look at the T side in terms of AWP buys, this includes all types of rounds, eco, pistol, etc, so it’s not an expression of choice in all cases.

It’s important for the T side to have access to the AWP, but at the same time they will run a lot of rounds without it, and the double AWP is so rare it’s probably the result of scavenging AWPs left lying around by dead CTs as much as deliberate strategy. This makes obvious sense as mobility is a core part of the T side strategy, the ability to switch from one point of attack to another and the AWP restricts that.

The CT side has a clear difference. Not only do AWP buys make up over half of all CT buys, but the double AWP buy is also a significant strategy. Obviously for the team taking up the defensive positions mobility isn’t as important unless the map features make defensive re-positioning very hard, particularly if the map has good long sight lines in key areas like Train.

So far so obvious. What do we find from looking at things map by map? Starting with the T side

We have 3 classes of map here, the largest group with slightly above 40% AWP buy rates in Cache, Overpass, Train and Mirage. The 2 slightly below 40% AWP buy rates in Inferno and Cobble, and as always Nuke out on it’s own.

Nuke’s unique complex of tight corridors and indoor spaces don’t play to the strength of the sniper so AWP cat and mouse games are less common. Even so the effectively meaningless levels of double AWP buy makes the T side map stats fairly uninteresting.

Because there are 3 types there’s a bit more to discuss on the CT side of things, and we can see how the variations interact. Unsurprisingly Train, with 2 AWP friendly bomb sites, is the most common place for double AWP buys. This is also because CTs do well on Train so they more often have the money to make the buy.

Cobble, Cache and Mirage all have similar levels of double AWP usage. Inferno and Overpass are on the next step down, but whereas Inferno’s single AWP buy rates still lag behind non-AWP buy rates Overpass’s single AWP buy rates surpass the non-AWP buy. So while the single AWP on Overpass is top priority the double is much less popular.

I would lean my interpretation of that towards the AWP coverage of A being considered sufficient on Overpass. The diagram below show’s Fallen’s AWP kills on Overpass in 2017. The green spots are his position, the red his victim’s and the lines are the path of the shot. It’s clear that A constitutes the lions share of the kills and that B receives reduced coverage, although his tendency to attack A short rather than long might be considered a little unusual. Perhaps attackers against SK try to avoid him on long.

GuardiaN’s AWP kill map is even more pronounced. Although he obviously does get into B sometimes, there’s a very strong concentration of kills coming from A long from within the CT spawn on A, which is as you’d expect given the enormous sight line there.

Obviously the geometry of the map makes a big difference in this case. B is a much more compact point that doesn’t have such exploitable sight lines so teams aren’t defending it with the AWP in the same way they can defend A, and as these diagrams show you can defend B from directly underneath A although it’s a little difficult to get there.

But that’s not the end of the story for Overpass. It we look at double AWP CT setups and their win rates, there are some interesting results.

Obviously double AWP is a big investment and this represents the CTs having a lot of money so naturally win rates are high. The margins here are very fine, but there are some differences. Nuke has the highest win rate, but the Nuke data is noisier as it has a smaller sample size. However Overpass is played a lot and that has the 2nd highest win rate for a double AWP setup despite having a relatively low instance of buying on it.

Buying a double AWP setup is a big commitment so team don’t always do it unless they’re sure it’s worth it. As we’ve seen above success rate varies, so how often do teams turn their eco into double AWP buys when they have the chance?

Above in the success rate chart Inferno has the lowest double AWP round wins on the CT side, so it makes sense that teams would be tentative about using it. But Overpass has a very good double AWP success rate and there’s still only a 50% investment rate into double AWPs when CTs have the option.

On Cache, Cobble, Train and Mirage it seems almost everybody picks up the double AWP when the eco situation allows it.

Looking at which teams commit to the setup on Overpass when they have the resources to do so, we can see a big split in attitudes towards this approach. Of the big name teams, G2 are currently the highest rated that like the setup going for it every time it’s available.

If we look at a map of G2’s CT sided AWP kills on Overpass we can see that although we don’t have as much data for them as they have had a few long breaks last year, it’s much more evenly spread between the 2 bomb sites than the individual activity for Fallen or GuardiaN that we examined before.

There is plenty of activity in Jungle and around B in general instead of just moving to defend from under A.

Amongst the really big names though, Astralis are very reluctant to try this and SK err on the side of caution. Faze are 50/50, and with olofmeister as their secondary AWPer it seems there is an element of individual skill involved. Astralis are known as a strat and execution focused team, not short on skill but willing to use it within the framework of a set tactical approach. It appears double AWP is only a footnote in the Astralis playbook.

FaZe on the other hand have a lot of players that have high individual skill and they use it, so it makes sense that sometimes olof picks up the AWP as well to defend B while GuardiaN holds down the A site.

This also tells us however that the success rate of the double AWP on Overpass is not down to the better teams using it, it’s a mixed picture but the numbers imply there is no obvious reason to stay away from it.

One interpretation may be that in it’s current state it is an inherently random choice so hard to predict, and therefore hard to exploit. You would have to be a better CS strategist than I to make that determination, but maybe teams are just underestimating the value of the CT AWP in defending Overpass B.

Predicting the Eleague Major Redux, with Flash Gaming

With the exit of TyLoo from the Eleague Major and their replacement by Flash Gaming, as well as the tail end events of 2017 slightly altering a handful of team ratings, I decided to re-run the tournament simulations.

The results don’t differ much from the previous article so there’s no additional writeup, Flash Gaming are rated very closely to TyLoo in our rankings so there’s not much difference.

Space Soldiers have managed to gain a few extra rating points over a couple of weeks so they have slightly better chances than previously illustrated. Below are the revised charts.

The Map Meta in 2017 – part 4: Map Pool Twins

One of the interesting things I found mostly by accident when reviewing the map statistics from 2017 is that using similarity analysis it’s possible to identify closely related map pools between teams. In part 2 I looked at the concept of adjusting map win rates by confidence levels, and it was these adjusted rates that I compared to find map pool twins.

Once again the candidate teams are from our world top 10, and the matches vary in quality.

FaZe and SK

It stands to reason that extremely strong teams have reasonably similar map pool success just because they’re so hard to beat on everything and will generally face inferior competition, however there are a couple of obvious differences. SK plays Cobble more and has a great deal of success on it, while the reverse is true of Inferno where FaZe dominate. SK’s avoidance of Nuke is also a big difference, but in general both team have similar success on 3 of the big 4 staple maps.

Not identical but with a strong family resemblance. Stick them in the same coat and hat and you’d never know the difference

Astralis and NiP

One of our outstanding matches, only Overpass and Cache are obvious points of departure. Of course Astralis have to earn their win rates consistently facing the likes of FaZe and SK in late knockout rounds, but that doesn’t stop them being joined at the hip on their Mirage, Cobble and Train win rates.

Cloud9 and G2 Esports

Rather than being twins it’s more a case of little brother Cloud9 trying to do everything big brother G2 can do, only not quite as well. Cloud9 lags behind across almost the entire map pool but makes up for it by being a far more efficient team on Train.

Will Cloud9 finally eclipse G2 in Boston, or will G2 continue to make Cloud9 slap themselves in face with their own hand?

Fnatic and North

Admittedly the similarities are getting a little thinner on the ground here, these two are more like warring cousins than twins each laying claim to their own corner of the map pool and wrangling over a couple of close ones in between. North’s big advantage on Overpass gives them the moral victory here, but if there’s going to be a family tragedy they’ll strangle each other to death over Inferno or Cache.

Yes I’m going to keep stretching this metaphor to breaking point.

Gambit and mousesports

Although it was the last one classified the close matches on Train, Cobble, Cache and Mirage make a pretty good case for a separation at birth. Gambit’s superiority on blue chip maps like Inferno and Overpass indicate that they got the upper middle class suburban adopted parents while mousesports were sent to the trailer park to learn Nuke, a map that is surely favourite to be banned into oblivion in Boston and potentially removed from the map pool entirely shortly after.

So in conclusion not quite so many twins as close family relations but with more similarities than differences. A fun exercise, not really useful in itself but potentially a reference for your own chin stroking map pool prognostications before clashes between the featured teams.

Predicting the Eleague Major: Boston 2018, a 1 million iteration Monte Carlo simulation

Now TyLoo have dropped out of the Tournament the tournament probabilities have been recalculated here Not much has changed so the writeup below is still relevant. Consider exchanging any mentions of TyLoo with Flash Gaming as our algorithm has them very evenly matched

For those of you unfamiliar with the process for these predictions, I use insight.gg’s unique Glicko team rankings to seed a Monte Carlo simulation of the upcoming tournament, then run a million simulations of the tournament and collect the finishing places of each team and how often they finished in them. From this we can then build a picture of the most likely performances of teams in the tournament.

Continue reading “Predicting the Eleague Major: Boston 2018, a 1 million iteration Monte Carlo simulation”

Meta Changes in 2017 – Weapons

The Counter Strike meta is a constantly evolving landscape, sometimes gently prodded by Valve doing their best to prove the virtues of intelligent design, sometimes wiped out in a mass extinction event. 2017 has been no exception with sudden and gradual changes alike.

Starting this review I’ll look at one of the latter, the slow death of the M4A1-S rifle. Whether to pick the M4A4 or M4A1 used to be a favourite forum debate, but at the top level at least the debate is largely over.

Continue reading “Meta Changes in 2017 – Weapons”

The biggest plays from ECS Season 4 finals

Using insight.gg’s unique win percentage model we have identified the most impactful plays of the tournament in terms of swinging the round and game in one or other team’s favour. These aren’t about conventional crazy skill, they are about big moments that changed the direction of the game as a whole.

Continue reading “The biggest plays from ECS Season 4 finals”

8 things we learned from ECS Season 4 Finals

1. This was a very CT sided tournament
The ECS finals were an oddly CT sided affair. Only 2 teams broke 50% round win rates on their T side in Astralis and mousesports. Even FaZe Clan only managed 44.4% win rate on the T side on their way to taking the title. One of the reasons for this may have been the unexpected amount of Nuke in the maps played, which is generally a CT-sided map. Only Train and Mirage were played more often, and Cache was ignored completely.

Everything about the T side suffered, bomb plant rates were generally well below average except for FaZe, Astralis and mousesports but conversion rates from plant to round win were below global averages for every team in the tournament. On the flipside of that retake rates were great with 5 out of 8 teams posting above average numbers, with Fnatic nearly 15% over average.

Continue reading “8 things we learned from ECS Season 4 Finals”

Pistol rounds: The myth of the 3 free rounds

It’s often considered a truism in CS that winning a pistol round is like winning 3 free rounds as the winner gets to start building up their equipment while the loser has to eco until they can make their own buy. The problem is that we know instinctively that it’s not really true, we see top teams reverse the advantage of the pistol rounds on a regular basis, and not just by hitting a round of improbably headshots.

Before going into the method of how pistol rounds can be reversed, what are the actual stats?

Continue reading “Pistol rounds: The myth of the 3 free rounds”

Predicting the winner of ECS Season 4 Finals

Using the insight.gg CSGO world rankings system we ran a one million iteration Monte Carlo simulation of the Esports Championship Series 4 Finals. This is what we found out.

Firstly the groups look very one sided. In Group A there’s only one team rated over 2400 in mousesports. In group B 3 teams are over 2400, with FaZe over 2500 rating points. Liquid at 2305 is the weakest in group B, but they’re still rated above both Luminosity Gaming and OpTic in group A.

Continue reading “Predicting the winner of ECS Season 4 Finals”