Above are the graphs from the Boys MITCA race this past weekend. The line hits the y-intercept at 943, so the adjustment is 54 seconds faster than AF2018.
So for Brendan Favazza, the formula is:
(1560-(931.1+54))/3 = 191.6
He won the race, one loaded with all-state talent. THIS DOESN’T MAKE ANY SENSE. IT SHOULD BE HIGHER.
Here is the problem with the graphing method. It assumes that the field you’re comparing is equal.
Allendale Falcon is a pretty fair representation of an invitational. MITCA is not. Neither are States or Regionals. The faster runners are competing and many below our top sevens are not. The line for these races ends up trending towards too low of a number.
That’s where I use REFERENCE RUNNERS.
This late in the year, there are 35,000+ performances from either boys or girls, on average of about 6-7 rated races per runner. There is a built in expectation of one’s rating, and that’s what I used to determine the adjustment for MITCA.
Name
MITCA Time
Unadjusted Rating
Name
2019 Rating
Difference
Aidan
Hollandsworth
19:35.2
128.3
Aidan Hollandsworth
162.9
-34.6
Akshay
Reddy
16:44.3
185.2
Akshay Reddy
194.4
-9.1
Alan
Atkinsmith
17:37.0
167.7
Alan Atkinsmith
169.4
-1.7
Alex
Lawhorne
20:18.1
114.0
Alex Lawhorne
135.5
-21.5
Alex
Mathews
19:14.9
135.0
Alex Mathews
151.2
-16.2
Andrew
Daniels
19:13.7
135.4
Andrew Daniels
140.6
-5.2
Andrew
Grier
19:43.5
125.5
Andrew Grier
117.8
7.7
Andrew
Lane
15:56.1
201.3
Andrew Lane
204.5
-3.2
Andrew
Nolan
15:44.7
205.1
Andrew
Nolan
207.5
-2.4
Andrew
Sesti
16:20.9
193.0
Andrew Sesti
194.9
-1.9
Armaan
Lehal
21:56.1
81.3
Armaan Lehal
98.8
-17.5
Austin
Aldrich
19:06.8
137.7
Austin Aldrich
159.3
-21.6
Austin
Smelker
17:51.8
162.7
Austin Smelker
173.2
-10.4
Bo
Anderson
18:48.8
143.7
Bo Anderson
151.9
-8.1
Braxton
Shutko
19:03.7
138.8
Braxton Shutko
147.5
-8.7
Brendan
Favazza
15:31.1
209.6
Brendan Favazza
216.7
-7.1
Brenden
Schroeder
17:58.3
160.6
Brenden Schroeder
176.6
-16.0
Brenden
Wallace
20:59.6
100.1
Brenden Wallace
118.0
-17.9
Bryce
Brown
17:07.6
177.5
Bryce
Brown
178.3
-0.8
Bryce
Mathews
18:53.9
142.0
Bryce Mathews
142.8
-0.7
Here, I am determining how much slower Shepherd ran compared to the average rating. I had 143 matches, showing Shepherd was 7.9 rating points slower than the average rating.
But as with the graphing method, I want to eliminate outliers. Taking the average of the data points between the two lines, Shepherd ran 7.3 points (or 22 seconds slower) than the average rating.
So again, for Favazza:
(1560-(931.1-22))/3 = 217.0
Personally, I prefer using the graphing method first, especially in larger fields. We’re blessed here to have large meets opening the season, such as your Ann Arbor Early Bird, Pete Moss, Corunna, Portage Early Bird, or Under the Lights. Those help set a standard for using reference runners to begin the season. For the record, reference runners were the primary method used for Regionals and State.
Seth is a great follow for NBA stats in non-numbers way.
Came across this tweet the other day and it hit me pretty hard. Although I’ve posted a few times my methodology, it sometimes feels the word isn’t getting across. Which is 100% my fault, not that I’m seeking sympathy, but just giving myself an honest appraisal. I can explain it to myself in my head and in person to others, but usually in a convoluted ramble.
I’ll go into why times DO NOT matter but they really do kind of matter in a bit, but first a reminder of how ratings are created:
Boys Allendale Falcon 2018 X axis = number of runners Y axis = time (in seconds)
On my summer post, I mentioned how I was looking for a baseline. It wasn’t a Eureka! moment, but Allendale Falcon 2018 seemed to fit the bill. In comparison between that race and MIS 2018, the difference between the two ratings was very minimal (on both boys & girls), stdev of that difference also low.
AF 2018 was the baseline for races this year. It may be the baseline for races next year. Other reasons I like it are that it’s in the middle of the season, reasonably healthy competition, and has a large sample of runners.
That large sample helps smooth out the graph. The graph is an S-curve, which you’ll get in almost all races. Extend that out to almost 500 runners, it gives you some pretty identifiable patterns. I drew a line through the mid-section of runners. Why the mid-section? These are your “average” runners, and with a greater number/less variability between “average” runners, one can begin to make an assumption between the difference between two races.
That line hits the Y-intercept at 997 seconds. Invitationals of a decent size (>150) and of comparable quality can then rated against this.
Here’s the graph from Holly this year:
Holly’s line hits the Y-intercept at 1031 seconds. The difference between the two is 34 seconds, and that is the course adjustment FOR THAT DAY. There are some drawbacks to merging races across a day, but I’m strapped for time as is, so that’s what we’re going with for now.
So, Carter Solomon ran 15:26.3 at Holly. 15:26.3 = 926.3 The formula for ratings calculation is: (1560 minus (time minus adjustment))/3 (1560-(926.3-34))/3 = 222.6
You may notice his rating this year was 221.6. AF2018 was listed in 2018 as 3 seconds faster than the baseline that year. I decided to try to keep the numbers consistent through the years, so the formula for Solomon’s race was actually (1560-(926.3-31))/3 = 221.6
TIMES DO NOT MATTER
Let’s say someone missed a turn at Holly and led everyone on a path to make the course 300m long. Every single runner in the boys races added exactly a minute to their time. Carter Solomon runs 16:26.3 in this example. Here’s the graph:
The line hits the Y-intercept at 1091 seconds. 1091-997 = 94. With the 3 second adjustment, then the difference is 91 seconds.
16:26.3 = 986.3 (1560-(986.3-91))/3 = 221.6
You can repeat this exercise for adding 10 minutes, 3 hours, subtracting 5 minutes.
A runner’s individual time, independent of everything else DOES NOT MATTER.
But you said time does matter. Yes, it does. The time it takes a runner to finish a course in and of itself doesn’t. The time it takes a runner to finish a course compared to other runners on that same course the same day does matter. For every runner in every race ever run, there is a time differential between that runner and every other runner in the race (this is the reason you can still make ratings if you start the clock when the first runner crosses the line – it would just be in a different format).
Comparing runners against one another
Athlete
Team
Holly
Difference (WS vs Others)
Peter Baracco
Farmington
987.8
-73.9
Adam Hill
Royal Oak
1006.6
-55.1
Connor Myatt
Royal Oak
1077.3
15.6
Jakar Dhillon
Bloomfield Hills
1057.2
-4.5
Weston Schroeder
Royal Oak
1061.7
0
Ian Dunn
Farmington
1065.1
3.4
Kemp Winkler
Royal Oak
1046.9
-14.8
Kevin Miller
Farmington
1077.9
16.2
Bryan Hanes
Royal Oak
1049.1
-12.6
Jack Rickle
Royal Oak
1081.9
20.2
The chart here shows 10 runners that ran at Holly. Third column is their time in seconds, fourth column is the difference between their time and Weston Schroeder’s time. Adam Hill ran 55.1 seconds faster than Schroeder here.
Now at the OAA White Jamboree the next week:
Athlete
Team
OAA White 2
Difference
Peter Baracco
Farmington
1001.5
-59.5
Adam Hill
Royal Oak
1009.1
-51.9
Connor Myatt
Royal Oak
1046.1
-14.9
Jakar Dhillon
Bloomfield Hills
1046.6
-14.4
Weston Schroeder
Royal Oak
1061.0
0.0
Ian Dunn
Farmington
1061.2
0.2
Kemp Winkler
Royal Oak
1068.3
7.3
Kevin Miller
Farmington
1069.2
8.2
Bryan Hanes
Royal Oak
1071.6
10.6
Jack Rickle
Royal Oak
1085.3
24.3
Same format as before, now Adam Hill ran 51.9 seconds faster than Schroeder. Now Adam’s rating would still being higher, having run the faster time. But relative to one another between races, Schroeder would cut the gap between the two.
This is a tiny example of using reference runners. That is a different, yet still viable method of rating. I’ll get into it next week, but I believe around 100 runners ran at Holly and OAA White 2. The average time difference between the two had OAA White 2 around 24 seconds faster than Holly.
SUMMARY (or just a section where I use another source that might explain it better)
I’m stealing this from the Godfather, TullyRunners. Meylan’s explanations are much longer but cover all the bases:
