Encourage CCL late season play by using a win multiplier

At the moment CCL suffers from these major issues:

  • Coaches that can play many games in the first few days have a great advantage of not having to play TV down.

  • Starting a cup run after the first 3 weeks is strongly discouraging as you will often be TV down and may not have enough time to play the 42 game cap.

  • Population drops off a lot as the season progresses, with lengthened spin times.

To help solve the issues above, and make the end of season stage actually exciting, how about introducing a daily cumulative multiplier e.g. 0.02 (just an example) which gets applied to the number of points you get for each individual win draw or loss?

So on day 1, you win the normal amount for a win loss or draw. Day 2 you get x1.02, Day 3 you get x1.04 and so on. On the last day of the season you would earn (for example) x1.82 the normal rate for game played on that day.

The "early spammers", will still play their games early (as not playing TV down is such a huge advantage). But they may even play more than they do now in the latter half of the season with new races or even a 2nd team of the same race, to cover other teams starting with the more lucrative rate.

Coaches whose teams collapse after week 2, would have much more incentive to try again rather than just waiting for next season...

This could make the final weeks a lot more exciting for those who play in CCL.

It also would not be an administrative burden, as hopefully the day number could be plugged into a formula, which I believe is external to the Cyanide code base?

One disadvantage would be that numbers could not be plugged into goblinspy calculator like they are now, as each win, draw and loss has to have the day number applied to it.

Such a change could lead to many more final day multi-twitch streams, since leads would be more interchangeable on the last few days of the season.

last edited by RunningDragon

It's an interesting idea. I'll look into it.

@runningdragon said in Encourage CCL late season play by using a win multiplier:

To help solve the issues above, and make the end of season stage actually exciting, how about introducing a daily cumulative multiplier e.g. 0.02 (just an example) which gets applied to the number of points you get for each individual win draw or loss?

Teams aren't being given "points" - their rank score is their win% scaled by the number of games they've played. What you're suggesting is either a complete rewriting of the ranking system to use a different overall score, or having the win% be a weighted average based on those daily multipliers, and in the latter case you'll see very, very little movement in the rankings since the weighting of each value will be relative to the weight of the other values, not relative to a baseline of day 1 values.

It's also important to note that while we may want to encourage late season play in addition to early season play we don't want to encourage it instead of early season play. Making later games worth more does the latter... and while I understand its because you think it will coincide with higher TV rating disadvantage for those games, it's far from guaranteed especially if people do not stick to the current trend of starting early.

So, what you're talking about is a more generalized/less accurate system than applying an elo-style system to the matches to compensate for the higher challenge faced in later season games. To date, nobody has found an elo-style system that provides as accurate a ranking as the ranking system that is in place.... and really, we'd want it to be MORE accurate since the operating theory here is that the rankings are NOT accurate relating to late-season joiners.

I ran some numbers using CCL season 3 data, and found an acceptable multiplier, and a method to use it without having to re-do the entire ranking system... though it still requires tracking an additional variable for each team.

The multiplier begins at 1 and increases by 0.005 each day. The new variable is the cumulative "win values" for the matches, which is just the usual 1/0.5/0 for win/draw/loss respectively, multiplied by the multiplier on the day the match is played. The adjusted win% is used in the same formula we already use for ranking value.

Adjusted Win% = (cumulative win values) / (total games played * (((final multiplier - 1)/2)+1))

So, assuming 42 days in a season, the final multiplier will be 1.205, making the value the total games played is multiplied by be 1.1025 ... obviously this is predicated on the idea that we'll know how many days there will be in a given season ahead of time.

Using that formula to calculate adjusted win%, and using the existing ranking formula, we end up with ranking values that remain strongly correlated with the number of games played by a team and their win%, but which fails to correlate with the first day the team plays in the season, at a statistically significant level (p > 0.05). The resulting ranking value is also almost as strong a post-hoc predictor as the existing rank value... being only 0.5% less effective (half a percent).

It should be noted, though, that the correlation between starting day and final ranking value, while statistically significant, was fairly weak even in the current system, having an r value less than 0.1... as such, I'm not sure there's much practical reason to change the ranking system. I'll likely check again while controlling for covariance with games played... and we may find that even under the current system starting day fails to achieve significance in that situation.

UPDATE: Checked the correlation between starting day and final ranks while controlling for games played (meaning, controlling for the obvious covariance between the two owing to the fact that teams that start earlier have an easier time playing more games) to test to see if the shift in TV levels affects the rankings... and the normal ranking system fails to correlate significantly with starting day with that covariance controlled for. We create a correlation with the multiplier system... which is undesirable.

So, I'd say that while such a system is reasonably doable, it is not a good plan... and the idea that people who start late are at a significant disadvantage is not supported by the data, assuming they can manage to play a similar number of games to the people who started earlier.

last edited by VoodooMike

@voodoomike said in Encourage CCL late season play by using a win multiplier:

I ran some numbers using CCL season 3 data, and found an acceptable multiplier, and a method to use it without having to re-do the entire ranking system

Thanks for taking the time to look into this.

@voodoomike said in Encourage CCL late season play by using a win multiplier:

So, I'd say that while such a system is reasonably doable, it is not a good plan... and the idea that people who start late are at a significant disadvantage is not supported by the data, assuming they can manage to play a similar number of games to the people who started earlier.

I agree that a multiplier at the end of the season would be a bit wacky, but was just a fun idea, but isn't that the issue though: the average coach cannot manage to play a similar number of games to those who start early and are pros, playing up to 5 times a day.? For example, a Chaos Dwarf coach wanting to start in week 3 (who can only play once a day) doesn't have any chance to qualify compared to a coach who can play 5 games a day.

Is lowering the 42 soft limit a viable alternative? I know some folk would not like it to be as low as, for example 15, as bad coaches are more likely to just get lucky and sit on the lead. Also the coaches who can play 5 times a day could just get more attempts to go 15-0-0. Also less team development, which would be less interesting...

Could having a lower soft limit (e.g. 30) encourage later season play?
What was the logic behind choosing 42? Is one game a day a realistic expectation of most coaches? Don't some people have other commitments at least some nights of the week?! 🙂

I've used 30 (and 18 as the 95% point iirc, as opposed to the current 42 and 28) in an OCC competition before. It worked out ok. It's one possibility to reduce the grind, certainly.

@runningdragon said in Encourage CCL late season play by using a win multiplier:

I agree that a multiplier at the end of the season would be a bit wacky, but was just a fun idea, but isn't that the issue though: the average coach cannot manage to play a similar number of games to those who start early and are pros, playing up to 5 times a day.? For example, a Chaos Dwarf coach wanting to start in week 3 (who can only play once a day) doesn't have any chance to qualify compared to a coach who can play 5 games a day.

One of the overt design decisions behind CCL was to encourage people to play as many games as possible... what you're talking about is wanting to make games played less relevant or outright irrelevant, which is the opposite of that. Unless they want to change their stated intentions regarding games played, I don't think we want to go that route.

Instead what I was looking at was the assertion that people who start later are at a disadvantage because their opponents are mostly well-developed teams, which in turn makes the newer team's matches more difficult, which in turn makes achieving a decent ranking difficult or impossible. That's not about games played, that's about the environment penalizing people for starting late even if they do manage to play a lot of games.

The data I was using says "that's not happening", in essence. When we control for number of games played, the day of the season a team starts does NOT correlate with their ranking to any significant degree. I even looked at it by excluding teams that played less than 10 matchs, since there was a chance that all the early, 1-2 game throw-away teams might be clouding that correlation... but no, still nothing.

So yes, starting early gives people an advantage... but that advantage appears to be that it gives them more time to play more games during the season... and does NOT appear to be that they're having an easier time winning matches. This may go a long way in explaining why the elo-style systems haven't shown any improvement over the existing pro-rated win% ranking system.

@runningdragon said in Encourage CCL late season play by using a win multiplier:

What was the logic behind choosing 42? Is one game a day a realistic expectation of most coaches? Don't some people have other commitments at least some nights of the week?! 🙂

Technically the realistic "soft limit" is at 28, where your received 95.6% of your win rate as your ranking metric. The next 14 games only increase that by 5%, and after that the increase is trivial. So, that's 2 games every 3 days, overall, if you're looking to hit that general soft-cap. You're not seeing all the qualifying teams having 42+ games under their belt.

I think if it's decided that there IS an objective problem in CCL relating to play statistics, we need to decide precisely what those problems are, and then start asking people why they contribute to that problem... then we'll have a clear idea on how to solve them rather than throwing noodles randomly at a wall to see what sticks.

Looks like your connection to Focus Home Interactive - Official Forums was lost, please wait while we try to reconnect.