After releasing a video named Lack of Real Stats Experts, Duncan “Thorin” Shields has found himself once again in the all too familiar realm of controversy. Amidst a barrage of criticism levied against him by members of the CS:GO community and authorities in the scene such as lurppis and Petar Milovanovic, much has been made of Thorin’s alleged lack of understanding of statistics and his supposed denial of their effectiveness.

In this article, I want to clarify Thorin’s point in his video, and the misguided nature of some of the criticisms made against him as of late.

The Video

Thorin begins the video by rejecting the notion that statistics are the conclusive authority on performance. He goes on to argue that Counter-Strike statistics, in their present form, are inadequate insofar as they do not reveal many nuances in the game. As such, Thorin claims that those who make claims under the guise of statistics do so while also taking into account qualitative data, and his qualm is that they do not admit that they use the eye-test as well as the statistical data. It seems that much of the frustration coming out of both lurppis and Milovanovic stem from this claim.

He then proceeds to claim that many people who use statistics as a basis of judgement tend to omit a great deal of information. Thorin then states, rather boldly, that there are no CS analysts/pundits that have been or are experts in the area of statistical analysis, something that lurpiss later repudiated in defense of Milovanovic. He continues by stating that these people do not have the required education in statistics in order to adequately understand them.

From what I've seen, @Tgwri1s is the hardest working person within & watches THE most CS. I find it insulting to suggest he doesn't get it. — Tomi (@tomi) February 26, 2017

The Trial

Are HLTV stats great? Probably not. Are they good? Definitely.



Are they blatantly wrong? No way. — Tomi (@tomi) February 26, 2017

First, on the matter of lurppis’ tweet, Thorin never claimed that HLTV Ratings are blatantly wrong in general. Rather, through the example of Xyp9x, he argued that these statistical methods fail to reveal the specific nuances of a match, specifically impact frags. This is actually a very strong argument. Statistics in their current form, through no fault of Milovanovic or any other statistician, fail to capture some important details. For example, there are several areas in the game where statistics fail to provide enough information to make an informed analysis: what was the context of a kill? Was it an ideal position? Was the player in an advantageous position? Was he supported by his teammates through flash-bangs and smokes? What was the lasting effect of a kill?

These questions, as of now, are not answered by statistics, and yet they play a key role in judging the impact a player is having on a game. We can imagine a situation in which a player gets a kill in a 3v4 against a rotating CT at a terrible angle that opens up the map for his terrorist teammates to enter a site without fear of a flank. Effectively, that player won the round with their team, however; such details often get lost to a certain degree when analyzing match statistics. This is the foundation of Thorin’s argument, and a critical point to keep in mind. If we make our judgments based solely what statistics tell us, we omit many details of the game that might reveal critical information in judging a player’s effectiveness, his importance, and so forth.

My many-years-long investigation (listening to him analyze at events) has shown Thorin has below average understanding of stats in CS — Petar Milovanovic (@Tgwri1s) February 27, 2017

Again, when Milovanovic claims that Thorin possesses sub-par statistical knowledge, this seems to be an irrelevant straw-man, regardless of whether or not it may be true. It seems that this argument sprouted out of Thorin’s case about sample sizes and the claims that may be justifiably made out of them. As previously stated, it is not entirely unfair to criticize Thorin on this account, but it would be a shame to reject the notion of his entire argument on this point alone. In fact, Thorin’s primary argument on sample sizes, if one manages to overlook the rather pompous marks made towards statisticians, is again one of contextualization.

Thorin is bringing to light the fact that CS is a nuanced game. For example, I doubt that he's criticizing the use of an average of kills one gets on a map. He is not claiming that it is wrong to say that, hypothetically, seized averages 0.05 more kills on Train than Inferno. Rather, he argues that this means seized is better on train than inferno is a flawed statement. This is not to say that the statement is of no value, but rather that it may be omitting key factors that are also in play. For example, does this hypothetical Na’vi give seized better positions on Train than on Inferno? Does he need to give play more conservatively on Inferno in order to give Guardian more space to frag, since they play the same site? Is he relegated to support Guardian with flash-bangs instead of going in himself to find opening kills?

These are questions that, as of now, are relegated to qualitative judgments. As such, Thorin is not claiming that statistical claims hold no value in analysis. Instead, he is rejecting the notion that they hold absolute value, and makes the point clear that there remains a need for the eye-test in CS:GO. With all things considered, I think that lurppis and Thorin are misunderstanding each other.

Thorin never claimed that statistics are outright "wrong", but rather there are specific instances where they fail to reveal the full story. The example given is that Xyp9x may be bottom fragging and receive a significantly lower rating than his teammates, and yet have much greater impact on the game. Let us say that dupreeh has 21 kills in a game, and Xyp9x has merely 16, while both have 16 deaths. An HLTV rating would indicate that dupreeh would have performed better. Yet it is entirely possible that dupreeh got 6 of his kills in anti-eco rounds, simply due to his assigned position, while Xyp9x got 14 of his kills in gun rounds. It is clear that Xyp9x had more impact, despite what the statistics say. As such, Thorin is claiming that there is need for a healthy balance of both statistics and qualitative judgement.

Thorin is not, as some have tried to claim, making an either/or statement. To the credit of Milovanovic, HLTV has many statistics that help extensively in assessing performance beyond simply KDA. Thankfully, we now have statistics of clutch rounds, opening kills, ADR, among many others to offer more context as to what happened in the game and who had the most impact. I believe the purpose of Thorin’s video, at heart, was to demonstrate that the potential depth of statistical analysis is far from being reached. While this has been taken as an attack on statistics, partly because of Thorin’s somewhat belligerent exposition, it ought to be taken with excitement.

The very fact that statistics have a long way to go means that there is much room for innovation, and many more tools for like Milovanovic to use. As a fan, I am saddened to see such a fascinating discussion be brought down by all parties to such a low level of discourse. Nevertheless, I am excited to see a world in which statistics and qualitative analysis are seen as complementary, allowing the analyst to reach more profound judgments using both methods. Through this, there will be room for a deeper appreciation of players, strategies, tactics, and CS:GO itself.

Image credit: global-offensive