Year-to-year ranking consistency

I’ve been wondering about how much (across the whole program) a team’s past performance correlates with their performance in any given year. I’d love it if one of the statisticians here would do some data scraping and run those numbers. It would be really interesting to see the results, and could start some good discussion.

There’s a thread here on CD that either covered or discussed this very subject in the thread, based on finishing places probably from TBA data

IIRC the top 3 are (in numerical order) 254, 1114 and 2056.

I know there was a thread (“FRC Dynasties” I think?) talking about the very top teams. I’m wondering about the rest of the pack.

1 Like

My Elo ratings for the start of a year are created by taking 70% of a team’s max rating from the year before, 30% from the year prior, and mean reverting this value 20%. This gives us a rough estimate that in a given year, each team can be roughly considered to be broken down into:
56% of the team they were the year prior
24% of the team they were 2 years ago
20% an “average” team

If you believe that Elo correlates well with performance, this gives you a ballpark for how performance carries over between years.

1 Like

Good resources + consistently returning students + long term mentors = continuity of knowledge + consistently of performance

I’d say it’s pretty independent of one years performance. Lots of teams have a few “good” years every now and then.

Assuming you’ve optimized well, all this means is that the previous year performance is a relatively good indicator. How good is it absolutely? To pick random numbers for an example, if a team ranks 10th at their one regional in year X, what’s their 50% confidence interval of ranking at their regional in year X+1? If a team’s end-of-year Elo rating was 1700, what’s the 75% confidence interval of their end-of-next-year Elo rating?

1 Like

stdev of end of season elo rating change between 2017 and 2018 was 86, so at roughly 95% confidence a team’s Elo stays within 172 of their last season’s Elo. Here’s some percentiles for some select teams:
A 20th percentile team in year X has 95% confidence of being between the 0th and 70th percentiles in year X+1
A 50th percentile team in year X has 95% confidence of being between the 10th and 90th percentiles in year X+1
A 80th percentile team in year X has 95% confidence of being between the 30th and 97th percentiles in year X+1
A 90th percentile team in year X has 95% confidence of being between the 55th and 98.5th percentiles in year X+1

Lots of assumptions baked in here, but that’s about where things stand.
image

1 Like

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.