jimrtex
All American
Posts: 2,553
Joined: Aug 2021
Reputation: 263
I Root For: Houston, Tulsa, Colorado
Location:
|
RE: '22 Season End Simulations [not enough 1-loss teams]
(04-26-2022 02:23 PM)Crayton Wrote: (04-26-2022 02:05 PM)jimrtex Wrote: How do you do your sim?
If you had a set of teams and a schedule could you project the results?
What does 125b indicate?
I've got a database with the teams and schedules. There is also strength data pulled mostly from past years. I run a Monte Carlo simulation 10,000 times. I prefer Monte Carlo to a projection because results in one game often influence others (tie-breakers, correlation of strength, etc) Each "sim" gives each team a different strength value, loosely based around their expected strength. Each game of the sim is then assigned a point spread based on that strength and a random value along a normal distribution of that spread is assigned to the game. A crude algorithm then turns that point spread into a football-ish score.
The original purpose of the software was to rank teams. The results of each sim are then incorporated into the ranking function and every team is assigned a rank from 1-131. Division/Conference tie-breakers are incorporated and postseason games are also simmed. Technically the sim only completes the postseason games of: CCGs, semis, and National Championship; I usually hand-sim other games for the results I post here.
"125b" means of the 10,000 sims, this is #125. If I want to go back weeks from now and reference that sim, I know which one to pull up. "b" I used here because I have over-written the original set of 10,000 sims to incorporate the changes mentioned in my last post. "125" is merely the default number in the system; out of laziness I usually look at that one first.
I am learning SQL because I want the 10,000 years to be accessible online, so that users can filter the results to see "what if" results.
Is the "strength" comparable to a Sagarin rating, where the difference in ratings along with a home-field adjustment is the mean expected point spread?
For each run, you assign a different strength value to the team. What does "loosely" mean. I assume this is to give more variability to the results, and that your strength ratings are necessarily inaccurate (past results are indicative, not predictive), and perhaps to represent events that are difficult or impossible to model, such as a starting QB breaking his arm in first game or a coach having an Urban Meyer event. By varying the season strength you may increase the chance of a team running the table, that would not happen if you applied strength on a game by game basis.
What is the standard deviation for each game? I read that it is 14 points for an NFL game. If it were very large (say 30), it would be like they came out an held the coin toss. "Rams call head", (coin toss), "It is heads, the Rams win the coin toss and the game. Good luck next week gentlemen." If it were 3 points, it would be too predictable.
Do you adjust the strength based on the results of the game? If a game between East-West State and North-South University were a pick, it might still end up a 28-point rout. In your simulation you would roll the dice and they come up snake-eyes or box-cars. Improbable, but not impossible. Though there is no skill involved in rolling the dice, we can assume a 28-point win by E-W is in part due to E-W being the better team. An adjustment in strength would also increase the chance of an undefeated seasons.
|
|
05-15-2022 01:05 PM |
|