Free Allegiance Forums

I think what we're going to find out is that you cannot possibly measure the value of a player until you can capture in data form what it is that makes that pilot valuable.

The outcome of the game is an improper indicator of skill of the individual player. No matter how many samples you take, if the data isn't specifically tied to the pilot himself, then you're only measuring game statistics and applying it to players callsigns. The assumption is that over time, the bad data will somehow be shrunken into the overall picture once enough samples are taken. This is a great error. Bad data is bad data no matter how many samples you use. Over time, you'll just end up with statistics of how many winning games he was on. Not how good he was IN those games.

And there is no link between how good a player is and how often he wins games (assuming balanced games etc.)?

Raveen wrote:QUOTE (Raveen @ Nov 23 2006, 06:55 AM) And there is no link between how good a player is and how often he wins games (assuming balanced games etc.)?

The link is so small that it shouldn't even be considered. It assumes that the team won/loss because of the individual's actions. The outcome of the game is so greatly influenced by the team as a whole you cannot possibly evaluate the skill of the individual player by sampling how many winning games he was on. This assumption is so terribly flawed that even taking a million samples over a thousand years won't correct it. As I said before:

"rank is much more a measurement of picking the winning team than it is a measurement of skill"

Which is again, why I said -

When you can no longer simply pick the stack, the results will be valid.

See with enforced team balance it's not about picking the winning team... it's about joining the game.

Wasp wrote:QUOTE (Wasp @ Nov 23 2006, 02:20 PM) "rank is much more a measurement of picking the winning team than it is a measurement of skill"

You'll note I said balanced games, maybe I should have said autobalanced games with even commanders to be more clear. Do you still maintain that in this situation, a good player will still lose more games than they win? Because that opinion seems deeply counter-intuitive to me.

Raveen wrote:QUOTE (Raveen @ Nov 23 2006, 07:30 AM) Do you still maintain that in this situation, a good player will still lose more games than they win?

I'm saying that whether the team loses or wins doesn't indicate whether he was good or bad. One has so very little to do with the other. If this game was only a 2 player game, then this would work. Since it is so far beyond that, so much so that the individual's contribution is burried in the teams actions as a whole, the measurement is too greatly tainted by factors outside his real skill. You're trying to measure the sunshine by the amount of grass that grows while ignoring the rain and temperature. When the grass doesn't grow, you assume there's no sunshine even though you're standing in the desert under the blistering sun.

No offense, just an analogy

Pook wrote:QUOTE (Pook @ Nov 23 2006, 07:29 AM) Which is again, why I said -

When you can no longer simply pick the stack, the results will be valid.

See with enforced team balance it's not about picking the winning team... it's about joining the game.

Picking the stack would be easy. Ride the rank elevator down, jump on the team of choice without autobalance being able to force you elsewhere. You whore and win the game...ta da.....elevator going up you say?....jump in your hider nick and hop on the next one going down. Does rank now indicate my skill, or does it just indicate my elevator rides?

Wasp wrote:QUOTE (Wasp @ Nov 24 2006, 12:53 AM) I'm saying that whether the team loses or wins doesn't indicate whether he was good or bad. One has so very little to do with the other. If this game was only a 2 player game, then this would work. Since it is so far beyond that, so much so that the individual's contribution is burried in the teams actions as a whole, the measurement is too greatly tainted by factors outside his real skill. You're trying to measure the sunshine by the amount of grass that grows while ignoring the rain and temperature. When the grass doesn't grow, you assume there's no sunshine even though you're standing in the desert under the blistering sun.

No offense, just an analogy

http://en.wikipedia.org/wiki/Bayes'_theorem

Elo doesn't 'ignore' the rain and temperature. It takes all players on a team into account when calculating win percentage, as well as how long each player was on team. If, to use your analogy, you took a whole bunch of samples that include the amount the grass grows, the sunshine, the rain, the temperature, soil density, arability, pests, etc., you could in fact build up a statistical model for how the grass will grow under any given combination of contributory factors.

Wasp wrote:QUOTE (Wasp @ Nov 23 2006, 03:12 PM) Picking the stack would be easy. Ride the rank elevator down, jump on the team of choice without autobalance being able to force you elsewhere. You whore and win the game...ta da.....elevator going up you say?....jump in your hider nick and hop on the next one going down. Does rank now indicate my skill, or does it just indicate my elevator rides?

Given that you'd be forced to play even games, you'd be deliberately throwing a game in order to win another game later on. For one thing that's dumb. For another that's against the RoC (1 player destroying the gameplay experience of others is blatantly not allowed). For a third, if you're throwing a game the com will boot you unless you're really careful about it.

But yes, you can expend some significant time and effort into fooling ELO into thinking that you're less good than you are, which will all be overturned if/when you start playing properly again. How would any other system protect against this whilst allowing for an autobalance feature?

Raveen wrote:QUOTE (Raveen @ Nov 23 2006, 03:15 PM) How would any other system protect against this whilst allowing for an autobalance feature?

I'm glad you asked /smile.gif" style="vertical-align:middle" emoid=":)" border="0" alt="smile.gif" />


First off, in order for autobalance to work, the ranking system must be an accurate accounting of one's skill level compared to the levels of others (relative rank). This must be done by a panel of expert players who can truely evaluate the skill level of each and every individual.

If you try to mathmatically arrive at a skill level using statistics, you'll only get a rank that represents his statistics, not skill.