(Since this came up again this year, more than a dozen years after the option was introduced, I figure I owe you all at least the original design intent of the feature. I realize this does not make everyone happy but gamers can usually draw their own conclusions based on the reasoning.)
Team Specific Interceptions for SOM NFL (Notes for V10)
Sunday, November 1, 2009 5:32:36 PM
(Copyright The Strat-O-Matic Game Co, June 12th, 2010, all rights reserved)
OTHER SOM PRO FOOTBALL POSTS:
This what team interceptions look like, with the size of the bubble indicating the relative number of automatic interceptions a team should get. That SOM bubble is what the stock rule provides.
Okay, well, the other shoe fell. Now you see why my time was constrained. I had to research and quantify and try to distribute and modify int results for almost a thousand teams, and like every other thing I do, this was done entirely by hand. So it was like another season in terms of time, and yes, it stopped me from attempting something else this year, although there were other factors.
Well enough of that- how these models work of course is somewhat dependent on what you do as a coach. The model assumes you will not elect to return every int (because all teams still have that fumble on 11). It still depends on the flat pass return ints that the standard chart has. Because older seasons have more flat ints, and these become TDs more readily, there are subtle changes to the model as you go farther back, and ints become more frequent.
Smokey Joe hit the target right on, a TD is not worth as much as you might think in terms of yardage, that was what he contributed earlier that broke this wide open. A long return on a team's card, while visually appealing, is not as valuable as its nominal total. The reason why is that a longer return number has a better chance of turning into a TD from more places on the field, and of course, that pesky end zone gets in the way and shortens the return when this happens.
A reasonable but not perfect model is 10*log(yardage)^2.5 which looks like this:
Nominal Observed (what it’s actually worth in a replay)
5 4
10 10
15 15
20 19
25 23
30 27
40 32
50 38
60 42
75 48
100. 57
That's base ten log.
The marginal value of another yard on a return will decline as the return gets longer. This model is good to about 3% of observed. The idea that it is log limited makes sense to me, since the boundary condition of of where the int occurs can be modeled on field position after the int, and the impact on length is skewed, and not normally distributed. It is a geometric (hyperbolic) model with powers greater than one in the expression.
(Think prob*likely return length, the value is initially small, then becomes progressively greater farther down the field, then compresses again at the far end of the field.)
The goal of each team's distribution was to model as best I could:
1)The team's yardage per return;
2)The team's chance to "house" a return, based partly on the predicted or expected results from its yardages, and based on real life.
3)The actual yards found in a team's record, and the “chunkiness” or distribution of a type of returns, that is if roughly a quarter of returns were 28 yards
there is likely to be 7 or 8 chances for that length of return;
4) The CMs can and do kneel on returns; this will add arbitrary “0 yard” returns to the simulated effort;
5) Where I can, the teams longest return is on the chart. 99 yards is the max in this framework.
What this means is if a team had a high average and few TDs, you can expect that the TDs it would get would come from its yardages. But if a team had a higher TD percentage than its yards per return or distribution would normally account for, I might use the power of the automatic TD to get this team closer to its goal, and the league closer to its goal.
Based on the team's distribution of gains, if it looked like they returned nearly every return, they were going to get that type of chart, chunky returns get chunky charts.
These are the primary constraints. I could not balance all of them perfectly for all teams. But I did get them to work well, especially at the league level, where you can see the results develop in as few as four replays. Individual teams, of course, take longer to test.
There are a few assumptions and limitations-
1) I mentioned the fact not all returns are considered to actually be returned- there is some flex in the model for this, but in general I used the CM's logic. Remember- kneed returns usually count for a big zero for their teams, so if the charts look a tad high, they are, because they account for this.
Of course, now that coaches can see returning ints if you are in charge of '04 Baltimore is like having Rick Upchurch get TWO trys sequentially to roll a 1982 punt return on his card, this decision-making ruleset may not always be realistic. Teams may try more returns in human managed games.
2) The game is limited to 99 yards for the chart, so about a dozen or so teams with 100+ yard returns have 99s instead and this has been incorporated into their model. This is a PC constraint we have to live with. Cards - n -dice guys, though, could sub that yardage back for feel (Ed Reed will be happy.)
3) Obviously throw more or less flat and you can change your opponent's results. One thing this might do is make coaches more circumspect about using flat passes to kill the clock with a lead, which I like.
The goal of course was to see if we could get teams to exhibit that 600 yard return season if they did so, or be limited to their 100-200 yards if they did not. By and large this system is an improvement by quite a bit over the stock system, and since I argued for years that it would not be useful to even try this, I can only say one thing-
I was wrong.
Fred
No comments:
Post a Comment