Boards›Community Creations

How much impact do different ranks have in a mixed-rank game? Preliminary overview of findings.

CupcakeTrap·7/2/2014, 11:50:48 PM·3 votes·183 views

EDIT: I've since updated these stats, incorporating many more matches into the mix. I don't have time to update this post, which will be "expiring" soon anyway, but if you want to take a look, click here to see the LF.net post on the topic.

http://i.imgur.com/4bLUSWq.jpg

This is just a quick descriptive overview, carried out using data from about 200 Factions matches. Obviously, that is not the best data, and this is not a proper statistical analysis. I'm sharing these results in the hopes of encouraging participation in a just-launched project investigating rank impacts in mixed-rank games.

You can help by organizing or participating in some balance-testing matches.

*: Bronze includes unranked Summoners and Summoners below level 30. When viewed as separate categories, Unranked/Bronze are virtually indistinguishable, while sub-30 Summoners are only a few points behind.

Introduction: Some Interesting Questions

Let's say you arrange a match of 5 Silvers (SSSSS) versus 1 Gold and 4 Silvers (GSSSS). How fair is this match? If you held 10 such matches, how often would the GSSSS team beat the SSSSS team?

What about SSSSS vs. GSSSB? Which side has the advantage? Is the "upgrade" from Silver to Gold more or less significant in terms of game-winning impact than the upgrade from Bronze to Silver?

We can step back and ask an even more general question: how strong is the correlation between game-winning impact and ranked tiers in the first place? I conducted some surveys and received responses ranging from "rank is meaningless" to "Gold and below is all the same; Plats and Diamonds are godly" to "there's such a large jump between each of the ranks that adding a Gold to an all-silver match will totally unbalance it."

Surveys only measure perceptions. I've begun a project to put those perceptions to the test with some empirical analysis.

Without further ado: our initial results

http://i.imgur.com/4bLUSWq.jpg

The first image is a simple win/loss computation. Essentially, we treated each match as 5 "wins" and 5 "losses", one for each player involved. A victorious Silver player contributes one "win" to Silver's total, while a defeated Gold player adds one "loss" to Gold's total. Thus, if an all-Silver team beats a team of 3 Silvers, 1 Gold, and 1 Bronze, the results would be:

  • +2 wins for Silver
  • 1 loss for Gold
  • 1 loss for Bronze

The "blame" for the loss is evenly divided between Gold and Bronze.

I then divide the number of wins for each rank by the total number of wins and losses for that rank, generating a "winrate" for each level.

http://i.imgur.com/GcWuo7a.png

The second image is a more specific investigation into relationships between tiers. For example, the match above (SSSSS beats GSSSB) would be treated not just as "two wins for Silver, 1 loss for Gold, 1 loss for Bronze", but instead as:

  • increased support for the claim that "Silvers are better than Bronzes"
  • decreased support for the claim that "Golds are better than Silvers"

The "noisier" the match, i.e. the more tiers are involved, the less its total impact.

http://i.imgur.com/a9t7OfS.png

The third image shows the rank distributions in the dataset we're using. It's a Factions dataset, because that's the data we have right now. Obviously, we'd rather be using data that isn't complicated by (e.g.) whether Piltover is stronger or weaker than Demacia.

I must stress that these are only very rough, initial results, and further that they are primarily descriptive summaries rather than the results of a proper statistical analysis. (Model-wise, we're currently thinking of either running a WAR analysis or setting up a sort of Elo/Glicko rating system in which each rank is treated as a single "player".) We definitely need more data, and preferably more non-Factions data, before we can make any meaningful conclusions.

The first step to moving beyond this rough initial analysis is getting more (non-Factions) data. You can help by playing a quick custom match. Any help would be much appreciated.

Research Objectives

I think this is a pretty interesting topic in its own right. I know that I for one have agonized over rank quite a bit.

It's also of considerable importance for Factions, a game mode based around (e.g.) Demacia versus Noxus matches. We usually can't get perfectly "mirrored" teams together for these matches, and so we adjust the points that factions win and lose based on the evenness of the matchup. This requires us to come up with answers to the questions posed above.

"Your data sucks / Gosh CupcakeTrap, you sure could use some better data!"

Yes to both. Please help out!

The only reason I'm posting such preliminary results in the first place is to (hopefully) encourage a few of you, good Redditors, to contribute a little time to helping us gather more data.

What does that require of you? Basically, replace one of your normal matches with a balance-testing match.

Why?

  • Science! Contribute to greater understanding of LoL.
  • Fun! Ever wanted to smurf so you could stomp on lower-tier players, or to get boosted so you could have your shot at proving you belong in a higher division? Well, here's an ethical and "legal" way to do so.

6 Comments

GundayMonday7/3/2014, 4:56:09 PM2 votes

Cool idea. I think you need to take into account role as well. For example, I have ~3k games jungling, but ~8 games adc in ranked. I'm a much better jungler than adc. So even though I'm a platinum solo queue player, I'm probably ~bronze in skill level for my adc mechanics/knowledge/play. Do you account for stuff like that?

EDIT: Also, some roles are better when paired with more skilled players. Supports and tanks are a great example. As a support, I could set up kills, ward properly, and zone the enemy team, but if my lane partner isn't following up (or doesn't know how) then it's wasted effort. Same deal for a tank. I can engage and peel like a champ but if my adc is the first one to tower dive, my efforts still won't matter. Breaking it up by role would help you out a lot.

CupcakeTrap7/2/2014, 11:50:42 PM1 votes

Whoa, CB kind of exploded on the formatting. I think I've fixed it now.

Angry Monster7/4/2014, 12:31:09 PM1 votes

you really should try to get lyte to throw his 2 cents on this. He may have soem data he is willing to share.

Though i personally feel that strickly looking at pure divisions is a mistake. Lyte has said in the past the the difference between b5 and b-1,2 is like looking at the skill difference b 1,2 to S 5,4. Also this was stated for d5 to d1.

I have never been down that deep in the ranked. I would say that the low gold and high gold have a skill jump. For 2 seasons i hit the low gold barrier and finally raised my self beyond g5.

I honestly think that there is a massive skill gap between ranks (5,4) and (2,1). And that you may want to factor that in your research. Cause if I am right people lower or higher in a tier may change your findings.