- last post: 01.01.0001 12:00 AM PDT
Posted by: The Overseer 92
After briefly scanning through the thread, I never really found an explanation of Sigma. It might be worth noting for those that aren't as knowledgeable in statistics that in a Normal Distribution (http://research.microsoft.com/en-us/projects/trueskill/skill dia.jpg) Sigma stands for the Standard Deviation, which in basic terms - is the average distance of all the points to the average, the most accurate measurement of spread.
In the Trueskill system, the system's insecurity about your actual skill level is Sigma - and hence the lower your sigma, the more likely (according to Trueskill) your skill level = your Mu, which is the average point on the Normal Distribution. Statistically speaking, a nice way to consider your Sigma is the following:
"In a Normal Distribution, approximately 67% of the values lie withing two standard deviations of the mean"
This means that 67% of the area under the graph linked before, lies within the region between one sigma below and one sigma above the mean - eg, if your sigma=1 and your mu=10, then 67% of the data will lie within the skill levels of 9-11. In the context of Halo 3* this means that there is a 67% chance of your actual physical skill level being within one standard deviation of the skill level shown by the playlist. Hence, the higher your sigma, the more unsure about what the exact value of your skill level is - and the more willing the system is to change your rank.
*I am aware that there is a possibility that Halo 3 doesn't follow the Trueskill system 100%, and this statement may not be entire with it's accuracy, but generally speaking it's true ^^
True Skill's sigma calculations are not exactly what you would expect though. It is arbitrarily reduced at times even when you are winning, and there is a modifier added back after every game. It is extremely hard to increase your sigma, which is not what you would expect if you skill level actually changed. Basically, sigma in True Skill is much more of a plug than the real statistical definition of sigma/volatility.