Forum adverts like this one are shown to any user who is not logged in. Join us by filling out a tiny 3 field form and you will get your own, free, dakka user account which gives a good range of benefits to you: No adverts like this in the forums anymore. Times and dates in your local timezone. Full tracking of what you have read so you can skip to your first unread post, easily see what has changed since you last logged in, and easily see what is new at a glance. Email notifications for threads you want to watch closely. Being a part of the oldest wargaming community on the net. If you are already a member then feel free to login now.

Do you think it would be possible to calculate ELO ratings for units in 40k? Drawing from tournament reults, could we calculate relative strength of units in the 40k game? Say attribute a win to a unit based on how much of the total points that unit occupies?

Yes it would be... possible.

Why don't you jump on Excel and get started!

Are you going to count just base unit (I.E. leman Russ) or variants (leman Russ battle tank) or specific loadouts (leman Russ battle tank with heavy bolter sponsons)?

Because this becomes more and more crazy the further you go.

Also, does a spammed unit in a single list count for more than a single unique unit? I.E. is a bloodangels smash captain going to be theoretically capable of being equal to his weight in guardsmen for example.

I feel like it would be really tough to quantify, especially for units with lots of options like an IG tank or infantry squad.

I mean anything is doable. it just going to take an ungodly amount of actual data input to get it right.

good luck mate.

If you can find an archive of army lists with win/loss records you might be able to do so, but getting a large enough sample of army lists and controlling scenario/matchup to come up with anything statistically significant is going to be an irritating problem.

You'd probably have to see if you can get the data from the best coast pairings app and go from there. Then lots of work. Definitely possible though.

It would also be cool if GW could track what sort of lists people were making when they come out with their list app.

What if we just took top 16 from major events and top table from minor events?

It would help too if GW put out an app that generated lists as well, then we could automate parsing the data.


Automatically Appended Next Post:

 MrMoustaffa wrote:
Are you going to count just base unit (I.E. leman Russ) or variants (leman Russ battle tank) or specific loadouts (leman Russ battle tank with heavy bolter sponsons)?

Because this becomes more and more crazy the further you go.

Also, does a spammed unit in a single list count for more than a single unique unit? I.E. is a bloodangels smash captain going to be theoretically capable of being equal to his weight in guardsmen for example.

I feel like it would be really tough to quantify, especially for units with lots of options like an IG tank or infantry squad.

Just a though experiment for now, but deffs a cool one. I wouldnt go too deep, just one for each dataslate.

Probably a waste of time honestly. Using ELO to rate individual parts is too complicated. There's too many variables. You could do it but the amount of work would probably be horrifically counter-productive. You'd never be able to keep up with the Meta as rules and game play changes. Plus with so many variables in play your data becomes less reliable because you have to make choices about how you're going to calculate different aspects of a unit.

A better way if you wanted to parse meta data for individual units is to collect information from tournament lists and parse out how commonly different units appear. As an example look at the way Smogon University classifies the Pokemon meta in tiers based on use (Uber, Overused, Rarely Used, Under Used, and Never Used). It won't tell you exactly how valuable a unit is on its own, but it will give you a gauge for where different units sit in relevance to how the game is being played. Arguably this is infinitely much more useful information than anything ELO will tell you and it's a hell of a lot easier and less time consuming to put together.

 davou wrote:
Do you think it would be possible to calculate ELO ratings for units in 40k? Drawing from tournament reults, could we calculate relative strength of units in the 40k game? Say attribute a win to a unit based on how much of the total points that unit occupies?

Would it be possible in theory? Sure, as long as you are clear about the game setup circumstances beforehand (e.g. for 2000pt games under ITC rules of 8th edition with X/Y/Z codex books and index lists using A/B/C missions).

The bigger issue is drawing enough data and analyzing it. An online game can draw from many thousands or millions of matches and monitor results on a constant basis. For 40k you'd have far fewer samples, and far more difficulty in recording and analyzing (and cleaning) that data.

In practice, I don't think it could be done feasibly.

 Vaktathi wrote:

 davou wrote:
Do you think it would be possible to calculate ELO ratings for units in 40k? Drawing from tournament reults, could we calculate relative strength of units in the 40k game? Say attribute a win to a unit based on how much of the total points that unit occupies?

Would it be possible in theory? Sure, as long as you are clear about the game setup circumstances beforehand (e.g. for 2000pt games under ITC rules of 8th edition with X/Y/Z codex books and index lists using A/B/C missions).

The bigger issue is drawing enough data and analyzing it. An online game can draw from many thousands or millions of matches and monitor results on a constant basis. For 40k you'd have far fewer samples, and far more difficulty in recording and analyzing (and cleaning) that data.

In practice, I don't think it could be done feasibly.

Even with that...stratagems could skew it quite heavily. You'd have to have rankings for Alaitoc SS, Ynarri SS, Saim Hann SS, etc.

 Daedalus81 wrote:

 Vaktathi wrote:

 davou wrote:
Do you think it would be possible to calculate ELO ratings for units in 40k? Drawing from tournament reults, could we calculate relative strength of units in the 40k game? Say attribute a win to a unit based on how much of the total points that unit occupies?

Would it be possible in theory? Sure, as long as you are clear about the game setup circumstances beforehand (e.g. for 2000pt games under ITC rules of 8th edition with X/Y/Z codex books and index lists using A/B/C missions).

The bigger issue is drawing enough data and analyzing it. An online game can draw from many thousands or millions of matches and monitor results on a constant basis. For 40k you'd have far fewer samples, and far more difficulty in recording and analyzing (and cleaning) that data.

In practice, I don't think it could be done feasibly.

Even with that...stratagems could skew it quite heavily. You'd have to have rankings for Alaitoc SS, Ynarri SS, Saim Hann SS, etc.

Not to mention the events themselves can skew results based on slight changes from FAQ's, house rules, local metas, or even that areas collection of terrain in both density and type, like if they own mostly ruins or purely LOS blocking like hills and rocks.

And then on top of that, all this data relies on what is put in by hand and must be combed through manually.

No, this isn't possible.

For an Elo rating to be meaningful, certain assumptions have to hold. These broadly do hold in chess. They don't hold in 40k between lists, much less between units.

You obviously aren't going to be able to rank lists such that the difference in their rankings tells you how likely each one is to win, right? There's not just one quantity of list-winningness that determines how well a list does in a game; some lists counter other lists which counter other lists which counter the first kind of lists. It's a rock paper scissors dynamic.

Most systems for ranking participants in a team effort (like the ones used for matchmaking in a game like League of Legends) depend on similar assumptions. They look at how much more likely a team is to win when a particular player is on them. This still depends on team-winningness being determined by a single quantity. But more than that, it depends on the players' contributions being independent. These systems don't notice that you do particularly well when on a team with a certain other kind of player, and terribly otherwise.

I would say that any effort to make this look rigorous is going to end up just being bs. You're much better off being clear from the start that this is a fuzzy sort of problem and then looking at a variety of metrics in hopes of catching outliers that are worth looking at more closely. So you could look at something like what you're interested in -- you could just go see to what extent the presence of or total points spent on a unit correlates with tournament placement -- but you should only use this to give you some idea of what units may be standouts when spammed and not as anything like a definitive ranking.

Possible? Yeah, sure. Probable? Not really.

There are already people out there that are really into mathhammer. They crunch the numbers of the statlines of every possible unit and compare it to their points values and put things together to determine what's the most efficient defense/offense per point or per model.

But there's a lot that that doesn't take into account. The size of the model and its base has an impact on how it performs in melee and in how readily it can find cover. The fact that the board, mission, and objectives change game to game also throws a monkey wrench into your plans. Then, you have to actually rate games to see what the actual statistics are. We know how well a plasma gun does against MEQ opponents, but how many kills are really attributed to plasma guns in the game? How many times are people overcharging and accidentally losing their own units instead of killing others? We know the math on IF you overcharge, but we don't know how often people ARE overcharging. Then there's stratagems, which vastly change how units can be used, but not all the time, and some units work much better with aura buffs versus no aura buffs, or with psychic power buffs or enemy debuffs.

If we take Overwatch for example, it's very easy to see the results of a lot of games and determine who the most popular heroes are depending on the map, team composition, and phase of the battle. But that is only available thanks to a high degree of automation and millions of matches being played over the course of a year, all the way from low tier to the highest grandmaster level.

For warhammer 40k, where one game can take up to 3 hours to complete, and may not even go all the way to turn 5 (let alone turn 7), these statistics become even more daunting.

And then, by the time you're ready to publish your findings, a dozen new codexes will have been released, along with FAQ's, erratas, chapter approved, supplements, campaigns, and other little tidbits that can throw everything out of whack.

When you look at it like that, ranking units and points and weapons seems incredibly daunting. Instead, why don't people just make fun lists with the units they like? Generally, by playing the game, learning your army, and feeling out different scenarios and deployments in your own experience, you'll discover which units, strategies, and other things work best for you.

 davou wrote:
What if we just took top 16 from major events and top table from minor events?...

You don't get data about a wide swath of the game that people for one reason or another don't consider worth taking in major events?

What's elo?

 koooaei wrote:
What's elo?

It was developed as a chess ranking system (it isn't an acronym, it's named after a guy called Arpad Elo, so technically "ELO" is wrong), but it gets used in a lot of other games as well these days. I don't know how applicable it is to describing specific units in a 40k game; it's supposed to be a dynamic system where your ranking changes based on the rankings of folks you play as well as whether you win/lose games.

Wikipedia's disambiguation page has also turned up a Samoan god of the underworld, a British rock group called "Electric Light Orchestra", and a breed of domestic dog, but since none of those are particularly applicable to 40k I'm going to assume the OP was referring to the rating system.

Best Coast Pairings is the place to start

I doubt it is feasible. Chess & Co. are obviously very static, so you can collect ELO ratings over time.

"40K ELO" (for units/lists, not players) would probably have to be calculated again from scratch after each new Codex or FAQ. Or inversely, a "40K ELO" equivalent to chess would probably still give you super-high ratings for Wraithknights, Grey Knights, Be'lakor, cheesve-Falcons, Rhino Rush Blood Angels, and more from 40K days of yore.

You would also need different ELO ratings for different 40K variants that are played. ITC is a very different game to Warhammer-World 40K or ETC or whatever, etc.., in ways that chess does not differ across regions and countries.

I don't think it works. Elo is for ranking players based on winning and losing vs stronger or weaker players. The issue with doing it for lists or especially units is untenable because the player matters to some extent.

You might be able to ELO units more effectively by just going to a larger tourney and removing 2% of the Match Point cost for each Vehicle/Unit/Character/choice that didn’t show up at all, and adding 1% to the cost for any of the same that show up more than 3 different lists gains 1% of match point cost.

At some point in the future the point cost should even out, and you will see a regular sway of a few things that sometimes raise a percent, and sometimes lose a percent.

Somebody get Nate Silver on the phone. We have a job for him.

Not enuff data, as previous posters stated. It should be theoretically possible, but there's effectively no way you'd get enough data to generate a model this complex before the next faq/codex/chapter approved/errata invalidates some, most or all of your dataset and you have to start over.

Also if you want to do Elo for units then you'd have to figure out how to handle buff/support units, in addition to the rock-paper-scissors relation between lists (so variable ratings per unit depending on the rest of the list and the opponent's list, the mission, the terrain, any house rules in effect, etc).

For example, a Tzeentch daemon prince is a lot better (more valuable) when its Strength is buffed to 8+, as this increases its killing power vs tanks and t4 units. However, this buff is worthless if the opponent's army consists of pure foot guardsmen. In fact, the daemon prince itself becomes less valuable in this scenario, as it is primarily geared to killing elite targets. How then could we meaningfully assign a single, general rating to a daemon prince*?

I agree with Dionysodorus though, there are definitely other related things you could analyze that are feasible and valuable.


* this is why skew lists work, and why even point costs remain an approximation.

 koooaei wrote:
What's elo?

A rock band from the 70s