A statistical look at G14

[Bongle]

Sorry about the repeated G14 threads, but trying to amass actual data about it is interesting to me.

So I was thinking about G14 the other day, and I realized that using previous-years scoring data could probably shed some light on what we can expect. I chose 2008 as my comparison because it was the last recent game that didn't have any big multipliers and your team's score was linear based on what it did. This should give a distribution of scores and blowouts similar to what we'll see in Lunacy.

So here are my results. I ran this on 2008 GTR, which had 88 matches.
Number of matches that would have had missing cells: 66
In matches with missing cells, average number of missing cells: 3.7

See attached chart for a distribution of how many matches were missing how many cells. In it, you can see that more than 50% of matches were missing 3 or more cells between the two alliances, which is pretty crazy.

You could say that G14 is kind of like pollution: your small contribution to your alliance's cell-loss doesn't matter if everyone else is picking up G14s as well, but it adds up and can really hurt your alliance.

However, keep in mind this is a worst-case: teams in 2008 were not caring about whether they doubled or tripled their opponents score. In this year, performance should be better in terms of keeping their cells.

I also did a run on elimination results, but there wasn't anything particularly insightful. The only thing I learned was that if you triple your opponent's score in finals, you lose all of your cells next match, which I really should have thought of ahead of time.

[bayesianlogic]

So, how was this done, precisely? Can you share the calculation logic or even the code?

[Bongle]

Originally Posted by bayesianlogic

So, how was this done, precisely? Can you share the calculation logic or even the code?

Pretty easy:

Copy/paste 2008 results from GTR into excel
Write VBA Macro:
for each match
  for each team in the match
    if this team scored 2x or 3x in their previous match, remove 1 or 2 balls and print that on the spreadsheet.
do statistic-y stuff.

Attached is my excel worksheet (with macros included). It's fairly ugly because I didn't write it to be pretty.

[Jared Russell]

Any way you can calculate the estimated effect of not having these balls?

E.g. estimate how many points a ball scores in the course of a match, and subtract these from the penalized alliances' scores. Is there a big effect on the W/L column?

[Laaba 80]

Did you do this calculations based on scores before or after penaltys? If it is after, your calculation will be completely off.

Joey

[Bongle]

Originally Posted by Laaba 80

Did you do this calculations based on scores before or after penaltys? If it is after, your calculation will be completely off.

Joey

I can't get un-penalized scores, so it's after. However, its usefulness as an upper-bound is not hurt by that:
-The losing team probably would amass more penalties than the winning team, though the distribution of penalties would probably be close to 50/50 or 60/40 for the winners/losers.
-Therefore, this will say that more cells would be missing than would really be the case, and is therefore still useful as telling us what the worst-case number of cells missing would be.

There aren't many games like this one. 2006 would be ok, but since autonomous granted a HUGE bonus both points-wise and gameplay-wise, the winning team would probably be 2x or 3x the losing team much more often than will happen in lunacy.

[XaulZan11]

I'm not sure how relevent this data is. Predicting scores based on a completely different game is very difficult and risky. Maybe if someone found data that shows the scores from past games are close, then this information may have more relevence.

Its as if the NFL decided that home field advantage is unfair and decided to give 2 points to all road teams. Then, to determine how many games would be affected, you looked at scores from MLB or the NBA. I know this is an exaggeration, but I think it shows the problem with relying on data from a completely different game.

I would be interested to see data after a week of regionals of alliance's record when missing Super Cells, though.

[Laaba 80]

Originally Posted by XaulZan11

I'm not sure how relevent this data is. Predicting scores based on a completely different game is very difficult and risky. Maybe if someone found data that shows the scores from past games are close, then this information may have more relevence.

Its as if the NFL decided that home field advantage is unfair and decided to give 2 points to all road teams. Then, to determine how many games would be affected, you looked at scores from MLB or the NBA. I know this is an exaggeration, but I think it shows the problem with relying on data from a completely different game.

I would be interested to see data after a week of regionals of alliance's record when missing Super Cells, though.

Completely agree with you. People always try to compare different games to give their opinions in rules. Like you say, they are different games. I would say it is more like comparing football to rugby. They are similar, but have very different gameplay.

[Bongle]

Originally Posted by Laaba 80

Completely agree with you. People always try to compare different games to give their opinions in rules. Like you say, they are different games. I would say it is more like comparing football to rugby. They are similar, but have very different gameplay.

Exactly. I was trying to get a round-about idea of how many cells you can expect to lose.

I didn't just decide to use 2008 as my proxy on a whim, I picked 2008 because the scoring style was similar:
-There were no big one-time finish bonuses to give lopsided victories
-Your score was linear based on how many tasks you completed, whether it was hurdles or laps
-It was a fairly high-scoring game.

2003 had multiplicative scoring. 2004 had capping as a doubler as well as the 50pt chin-up bonus. 2005 had lots of bonuses for the geometric arrangement of pieces. 2006 had a huge autonomous bonus. 2007 had exponential scoring. 2008 was the best stand-in for the scoring style we'll see in lunacy, though its autonomous bonus was still big.

However, 2008's usefulness as an upper-bound of how many cells a team could expect to lose still stands. Almost all the variations between Overdrive and Lunacy tend to make it so that an Overdrive score will be MORE lopsided than a Lunacy score. Therefore, we can probably predict that super-cell losses will be somewhat less than this thread predicts. At the very least, an analysis like this is a little better than the back-and-forth with very little evidence that exists in other G14 threads.

Though you give me an idea to check to see how similar scoring distributions are between years...

[Lil' Lavery]

Originally Posted by XaulZan11

I'm not sure how relevent this data is. Predicting scores based on a completely different game is very difficult and risky. Maybe if someone found data that shows the scores from past games are close, then this information may have more relevence.

Its as if the NFL decided that home field advantage is unfair and decided to give 2 points to all road teams. Then, to determine how many games would be affected, you looked at scores from MLB or the NBA. I know this is an exaggeration, but I think it shows the problem with relying on data from a completely different game.

I would be interested to see data after a week of regionals of alliance's record when missing Super Cells, though.

The difference here is that the NFL would have previous years' games to look at, FIRST doesn't.

We can't afford to go with "let's look at Week 1" method, because doing so is essentially screwing over week 1. We need to fix the problem BEFORE the competitions begin, rather than tell all teams competing in week 1 that they are just the test subjects.
And the best and closest examples we have, while not exactly the same, are previous years' games.

[XaulZan11]

Originally Posted by Lil' Lavery

The difference here is that the NFL would have previous years' games to look at, FIRST doesn't.

We can't afford to go with "let's look at Week 1" method, because doing so is essentially screwing over week 1. We need to fix the problem BEFORE the competitions begin, rather than tell all teams competing in week 1 that they are just the test subjects.
And the best and closest examples we have, while not exactly the same, are previous years' games.

Maybe I don't understand the purpose of this data. What problem are we fixing? (I sure hope your not using this data as proof that G14 is a bad rule and shouldn't be used).

I do agree with you and Bongle that this is the best data we have, but it simply isn't good enough to make any conclusive statements.

[Laaba 80]

Originally Posted by Bongle

Exactly. I was trying to get a round-about idea of how many cells you can expect to lose.

I have no problem with you showing this because you realize they arent the same. I have a problem with people posting on rules and saying "It was allowed/not allowed last year. Just look at the differences the past 4 years
2006 - insane basketball
2007 - extreme tic tac toe
2008 - race
2009 - moon
They dont seem very similar to me.

[Lil' Lavery]

Originally Posted by XaulZan11

Maybe I don't understand the purpose of this data. What problem are we fixing? (I sure hope your not using this data as proof that G14 is a bad rule and shouldn't be used).

I do agree with you and Bongle that this is the best data we have, but it simply isn't good enough to make any conclusive statements.

This data isn't the proof, the proof is the fact that G14 penalizes teams who have done nothing wrong. Why should my team be penalized for the actions of my alliance partner the match before?


This shows the upper bound of the scale of this problem.

<G14> is, by far, the single worst rule written since I've been involved with FIRST. There is no justification for this abomination of a rule.

[thefro526]

The Data is interesting but Overdrive may not be the best comparison. The distribution of points is entirely different than in Lunacy. This year we have one primary scoring objective that has a uniform point value with the opportunity for 15 point in bonuses where Over Drive had 4 different objectives with 3 different point values. I don't know what game would be the best comparison because I'm not familiar with the scoring prior to 2006.

If the Statistics for Overdrive were true though and I'm Reading the Graph correctly then my understanding is that 75% of matches would have at least one ball missing?

Thats rather scary now that I think about it.

[Daniel_LaFleur]

Originally Posted by Lil' Lavery

The difference here is that the NFL would have previous years' games to look at, FIRST doesn't.

We can't afford to go with "let's look at Week 1" method, because doing so is essentially screwing over week 1. We need to fix the problem BEFORE the competitions begin, rather than tell all teams competing in week 1 that they are just the test subjects.
And the best and closest examples we have, while not exactly the same, are previous years' games.

You see this as a problem (your own words), I do not.

There are going to be very few 0 point alliances as you start directly in front of your opponents PS and it will take a few seconds to get away (if you can). The real time scoring system will give the alliances an idea of where they are (not an exact number, but a fairly good idea). So all you need is a little communication within the alliance and some decent strategy and <G14> should not be an issue.

I, for one, would rather spend my time working through and planning around <G14>, to do otherwise is a waste of time and energy.