Division Strengths

[Rick TYler]

Did you calculate any measures of spread, like standard deviation? Is the median close to the mean?

Inquiring minds want to know.

[Ed Law]

Originally Posted by Rick TYler

Did you calculate any measures of spread, like standard deviation? Is the median close to the mean? Did you calculate the mode?

Inquiring minds want to know.

I did calculate it but I decided not to report it because looking at the graph tells you more than just another statistical number. If you are really curious enough, you can do that in Excel. Use =STDEV(B2:B88)

Ed

[Rick TYler]

Originally Posted by Ed Law

I did calculate it but I decided not to report it because looking at the graph tells you more than just another statistical number. If you are really curious enough, you can do that in Excel. Use =STDEV(B2:B88)

I did it for my own amusement:
Division: Archimedes Curie Galileo Newton
Standard deviation 953 973 894 929
Median team number 1,302 1,108 1,124 1,138
Mean team number 1,341 1,267 1,192 1,229

... and of course "mode" doesn't mean anything when no value repeats. Sheesh...

[smurfgirl]

Originally Posted by Rick TYler

I did it for my own amusement:
Division: Archimedes Curie Galileo Newton
Standard deviation 953 973 894 929
Median team number 1,302 1,108 1,124 1,138
Mean team number 1,341 1,267 1,192 1,229

... and of course "mode" doesn't mean anything when no value repeats. Sheesh...

What about doing it by years of involvement, rather than team number?

[Tom Saxton]

Originally Posted by smurfgirl

What about doing it by years of involvement, rather than team number?

I have a tab-delimited text file that lists all of the teams and their rookie years. If that's useful to anyone, DM me.

[Ed Law]

I want to emphasize that in my scouting database, I used average OPR and average CCWM of all the regionals and districts a team attended.

If you would like to use best OPR and CCWM, or most recent OPR and CCWM you can do that and the World Ranking will be different.

Next year I am going to change it to report a weighted average so that if a team attends 2 or more events, the earlier ones will weigh less than the later ones.

Ed

[Ed Law]

I did an interesting study. A lot of people from Michigan and outside of Michigan who watched the matches of the Michigan State Championship have made comments like
1) overall high performance and evenness of the teams in the matches
2) depth of the teams based on the fact that many good teams did not get into elimination round
3) very exciting to watch because there are very few lopsided matches
4) tougher than world championship and comparable to IRI

Team 2834 was at the State Championship and played in the elimination round so we have first hand experience.

I am not expressing my opionion whether the Michigan district and state championship model is good or not or whether other parts of the country or the world championship should adopt the model that robots need to qualify and teams who won rookie all-star award can have their robot compete. Some people may not like what I am showing below. I am just reporting on numbers so please don't shoot the messenger.

I overlayed the Michigan teams who made it to the State Championship with their OPR and CCWM before the State Championship. I "stretched" it horizontally to match the 87 teams in each division. You can see that in the attached file. This is my finding.
1) In general, Michigan teams are not stronger in OPR and CCWM than any of the divisions. The average OPR is 21.9 and CCWM is 6.4 and they are right in between the 4 championship divisions.
2) The upper half of the teams in the MI State Championship are actually lower than all the divisions meaning the good teams in the championship are better than the good teams in Michigan teams overall.
3) The lower half of the teams in the MI State Championship are much higher than all 4 championship divisions which is not surprising since robots have to qualify. This could explain why the MI State Championship seems to have higher performance because of the relative evenness of the teams. This could be an argument for the MI State Championship model if you want it to be more exciting.

What do you think?

Ed

[Rob]

Hi Ed,

I think this is very interesting, but I have a question. Please let me start by saying that I have a general idea how the math for this works, but I am not overly mathematically inclined so the details elude me.

My thought is that since the set of data that a teams performance is being calculated from comes from single events (each event having a set of teams), that the results of that data are best used to compare to other teams at the same event.

To properly compare one team from the midwest against a team from southern California, those teams would have had to compete at the same event.

Am I totally off base here of is there some validity to my assumption?

For example, if the MI state event had all great teams, could that not adversely impact the performance numbers of the top teams. Good teams score on you more (you get worse DPR) and are harder to score on (you get lower OPR).

I am not trying to invalidate your data, I think it is an amazing tool. I just have a hunch that it is most useful for comparing teams at the same events together. To compare teams globally they would have to be at the same events together (I think)

Now the top teams should still have good performance numbers no matter where they compete, but I think the numbers can only tell so much.

Sorry for the lengthy post, and thanks for making this data available, it does help quite a bit even if my understanding of it is limited!

[ScottOliveira]

Originally Posted by Rob

Hi Ed,

I think this is very interesting, but I have a question. Please let me start by saying that I have a general idea how the math for this works, but I am not overly mathematically inclined so the details elude me.

My thought is that since the set of data that a teams performance is being calculated from comes from single events (each event having a set of teams), that the results of that data are best used to compare to other teams at the same event.

To properly compare one team from the midwest against a team from southern California, those teams would have had to compete at the same event.

Am I totally off base here of is there some validity to my assumption?

For example, if the MI state event had all great teams, could that not adversely impact the performance numbers of the top teams. Good teams score on you more (you get worse DPR) and are harder to score on (you get lower OPR).

I am not trying to invalidate your data, I think it is an amazing tool. I just have a hunch that it is most useful for comparing teams at the same events together. To compare teams globally they would have to be at the same events together (I think)

Now the top teams should still have good performance numbers no matter where they compete, but I think the numbers can only tell so much.

Sorry for the lengthy post, and thanks for making this data available, it does help quite a bit even if my understanding of it is limited!

You definitely have a valid point, the data could be skewed by overall team quality at a given competition. However, it is generally safe to assume that the average team quality at all of the regional competitions is comparable. Consider that a lot of the good teams travel to multiple regionals, helping to balance things between.

Also consider that given the limited data available, this is roughly as accurate as you can get. There are too many variables for a perfect set of data. You could also consider that performances at regionals later in the season might be different, as those teams have had more time to code/practice, and are more likely to have already competed in a regional.

Yes, the data isn't perfect and there are many variables that could affect the rankings, but it's as good as is possible.

[Ed Law]

Hi Rob,

You are absolutely correct. However it is still good to compare teams who attend different regionals. I can not believe there is such a big difference in terms of competitiveness among most of the regionals. I am sure there may be a few that are particularly weak.

After we come back from Atlanta, may be during the summer months when I suffer from FRC withdrawal symptom, I will try to assemble all the teams and all the matches into one giant 1674 X 1674 matrix. Then all the interactions will be taken into account. There are quite a few Michigan teams that went to other regionals also. Since I don't invert the matrix to solve for OPR and CCWM, it should not even take that long to solve.

Ed

[Huskie65]

Originally Posted by Rob

Hi Ed,


For example, if the MI state event had all great teams, could that not adversely impact the performance numbers of the top teams. Good teams score on you more (you get worse DPR) and are harder to score on (you get lower OPR).

I am not trying to invalidate your data, I think it is an amazing tool. I just have a hunch that it is most useful for comparing teams at the same events together. To compare teams globally they would have to be at the same events together (I think)

I totally agree with this. With all good teams there are alot less blowouts. Harder to score. Your defensive and offensive ratings naturally go down by playing better teams.

[Ed Law]

Originally Posted by Huskie65

I totally agree with this. With all good teams there are alot less blowouts. Harder to score. Your defensive and offensive ratings naturally go down by playing better teams.

That is why I used the OPR and CCWM numbers before the MI State Championship for MI teams. After the Championship, about half of the teams' CCWM went down because even if all the teams are good, about half of them are going to get positive CCWM and half of them are going to get negative CCWM.

Ed

[IKE]

While the divisions overall may be reasonably even, if you look at the top 24 (those likely to play in Elims), there is a pretty considerable discrepancy.

Taking the top 2 and the 24th of each divsion, I can see a bit more of a discrepancy. Curie and Galileo look quite a bit stronger than Newton and Archimedes. There is also a big gap in abilities on the top two in Curie and Galileo which could be inferred as a better odds at one of those two making it to the top seed.

IKE

[Katie_UPS]

So while this thread is really cool, and the numbers are kinda helpful, and I can see the competitiveness of the devisions and all the wotnot, I just have to ask...

Where do you find the time?

From what you said, my basic idea is that you make a matrix for each team and then average it out. Thats.. what? 400 teams?