So far, we have records from about 30 unique matches. Keep them coming!
Naturally, people want to know when they’ll see these scored and posted on the Balance of Power page. The primary holdup is that I’m going to be upgrading the Balance of Power skill adjust algorithm in a very significant way, but this will require some nuanced setup. I thought about just posting the un-adjusted results, but that would be misleading.
I’m hopeful that I’ll be able to get this posted up by the weekend, along with a few other critical arc-launching materials. Sorry for the delay, but on the other hand, this is going to be a dramatic improvement over the old algorithm.
- Original System: It vaguely occurred to me that there should be a skill adjust. I rated Bronze at 1 point, Silver at 2, and so on. I averaged these ratings for both teams and applied a point adjust based on the gap. This was better than nothing, but far from scientific. This was back before LeagueFactions.net or any of the fancy stuff. There was no staff. it was one step above “hey GD lets play noxus vs. demacia k?”
- Previous System: I asked a lot of people what point values seemed fair for different tiers of player (“If a Silver is worth 20 points, what about a Gold?”) and then hacked together an updated version of the above. This was better in that it made some attempt to value the tiers, rather than just arbitrarily lining them up on a 1 to 6 scale.
- Upcoming System: The first system truly grounded in empirical data. I’ve tabulated hundreds of matches according to the skill compositions on the winning and losing sides. In this way, I’ve come up with predicted winrates based on tier alone: “all else being equal, if one side is GGSSS and the other side is DGSBB, the GGSSS side wins about ___% of the time.” (Stats indicate the answer is about 44% of the time, for that matchup.) To use easier numbers, let’s say that, for a given matchup, the favored side is twice as likely to win (on rank alone: not counting faction strength, coordination, strategy, or morale) as the other side. If the favored side wins, the point value is cut in half; if the lower side wins, the point value is doubled. In this way, if the only difference between the teams were their rank compositions, no points would be won or lost overall: one team wins twice as much but gets half the value for it, the other team wins half as often but gets double the value. In other words, we control for rank and send the points toward the side that outperforms their raw ranked strength, e.g. by use of good faction strategy.
In other words, this is going to be a massive fairness improvement. Because of our somewhat limited dataset, though, I’m going to have to code it in a slightly complex way:
- First, we check to see if we have a meaningful amount of data (n of at least 10) on the specific matchup, e.g. GGSSS v. DGSBB. If so, use that.
- Else, convert ranks to the High-Gold-Low categorization. Diamonds and Plats become “H”, Golds stay “G”, and Silvers/Bronzes become “L”. For example, GGSSS becomes GGLLL, and DGSBB becomes HGLLL. This simplification dramatically improves the data-depth of our sample. If we have at least n = 10 for this, use that figure.
- Else, take a step back and cancel out mirrored team members, on the non-trivial but IMO acceptable assumption that teammates contribute additively to odds of success Thus, GGSSS vs. DGSBB becomes simply GSS v. DBB, because both sides have a Gold/Silver pair. Again, check to see if our database contains enough entries of this nature. If not, proceed to step four.
- Apply both steps 2 and 3 simultaneously: convert tiers to HGL, then cancel out mirrors. GGSSS becomes GGLLL, and DGSBB becomes HGLLL. Then, we cancel out the GGLLL vs. HGLLL mirrors, reducing it to simply G vs. H.
- If we still don’t have enough data for the matchup yet — and this would be rather surprising — we use an interpolation table. For example, let’s say that we actually just don’t have any HHGG vs. LLLL matches, but we do have a fair number of HHHG v. LLLL matches and a fair number of HGGG v. LLLL matches. As a rough estimate, we could use the average of these numbers to guess the “middle ground” value for HHGG vs. LLLL: surely, that’s not as bad for LLLL as HHHG would be, but it’s worse than HGGG.
Anyway, that’s what I’m working on now. I realize that Factions isn’t super-tryhard, and precise balancing is not the goal: indeed, the core concept, Lana, is that we have asymmetric matchups with different rosters. That said, I do care about fairness, and I’d rather a faction win out because it’s better organized, or its Champions are stronger, or whatever else, rather than simply because they have more Diamonds.
In closing, I do have a couple requests for match-runners:
- Make matches fair. Take the time to balance out matches as closely as you can. At the very least, make sure people are okay with the matchup before starting.
- Confirm who’s submitting the match. Before the match starts, make sure people know who’s sending in the results.
- Grab the LoLNexus.com screenshot. Yes, we can have the scorers look up each person individually, but it’s a lot more burdensome than just glancing at the LoLNexus lineup, especially when you’re talking about a large volume.
- Have fun! LoL is inherently sort of tryhard, but do your best to have fun and relax in Factions matches.
We’re rolling out a lot of new stuff this arc, and I think it’s going to make it a really good one. Please continue to bear with us for the next couple weeks as we get it all in place: we’d rather do it right the first time than implement something half-baked and then have to fix it.