Gonzalez Exchanges for Nationals 2017

Here we have a very long table for Day 1 of Nationals 2017. We will be playing 35 matches on Saturday. For many long reasons, the following table includes the pre-Nationals ratings for each exchange. The ratings are not updated round to round like they would be in reality.

For example, NSULA’s pre-rating is 40.640 and is for each of their matches. In reality their first match vs OSU will naturally raise or lower their rating, which in turn will affect the exchange against their second opponent, DePaul.

I guess to summarize, the purpose of this is to show you a close, potential weight each match has on each team’s rating. Exchanges are doubled for Nationals, so an even match will have an Exchange around 2. Most matches will fall within 0-2, while upset exchanges will be 2-4. 

Queue 1 High Seed Low Seed Rating Rating Predicted Exchange Exchange if Upset
J3 OSU NSULA 42.779 40.640 1.572 2.428
J4 PSU BGSU 43.156 42.329 1.835 2.165
S2 GVSU Akron 56.623 37.944 0.020 4.000
S3 MSU Kent 46.423 45.903 1.896 2.104
S4 Ohio DePaul 39.484 37.474 1.598 2.402
Queue 2
J3 UWP UNG 39.376 38.061 1.737 2.263
J4 CSU Miami 36.424 34.105 1.536 2.464
S2 UK UMD 48.235 36.149 0.020 4.000
S3 WKU UVA 42.223 37.425 1.040 2.960
S4 JMU SVSU 49.161 48.708 1.909 2.091
Queue 3
J3 PSU VCU 43.156 40.667 1.502 2.498
J4 NSULA DePaul 40.640 37.474 1.367 2.633
S2 GVSU BGSU 56.623 42.329 0.020 4.000
S3 Towson Ohio 48.819 39.484 0.133 3.867
S4 CMU OSU 49.717 42.779 0.612 3.388
Queue 4
J3 Akron UMD 37.944 36.149 1.641 2.359
J4 Kent WKU 45.903 42.223 1.264 2.736
S2 UK UWP 48.235 39.376 0.228 3.772
S3 SVSU Miami 48.708 34.105 0.020 4.000
S4 JMU CSU 49.161 36.424 0.020 4.000
Queue 5
J3 CMU UVA 49.717 37.425 0.020 4.000
J4 Towson DePaul 48.819 37.474 0.020 4.000
S2 PSU OSU 43.156 42.779 1.925 2.075
S4 MSU NSULA 46.423 40.640 0.843 3.157
Queue 6
J3 UNG UMD 38.061 36.149 1.618 2.382
J4 BGSU UWP 42.329 39.376 1.409 2.591
S2 JMU Kent 49.161 45.903 1.349 2.651
S3 SVSU Ohio 48.708 39.484 0.155 3.845
S4 WKU VCU 42.223 40.667 1.689 2.311
Queue 7
S2 CMU UK 49.717 48.235 1.704 2.296
S3 Akron UVA 37.944 37.425 1.896 2.104
S4 GVSU Towson 56.623 48.819 0.439 3.561
Queue 8
S2 MSU VCU 46.423 40.667 0.849 3.151
S3 NSULA CSU 40.640 36.424 1.157 2.843
S4 UNG Miami 38.061 34.105 1.209 2.791

The ease of the Gonzalez system makes it simple for us to post the rating charts almost instantaneously as they come in. Something that was impossible in the past 3-4 years. We have a lot to accomplish at Nationals, but the stated goal of this editor is to post the Gonzalez Rating Standings after the results of each queue. And maybe even the stakes of the next matches.

Look for a live feed on the website of these ratings, along with the results. Chaos permitting.

Observations

I love and hate working with a big schedule. Other than the complexity and working parts, there’s so many interesting observations you can make en-masse. Obviously, the outcomes of these 35 matches will be an excellent test of the system. People have said:

Arguably, this season has demonstrated the most parity the League has ever seen.

It’s not an argument. It’s clear, the records show it and this is statistically demonstrated; the League has not ever seen this parity from “top” to bottom. There are some solid sections of the rankings where teams are pretty close and with Nationals matches worth more than normal, expect a shakeup. 2-8 are close, 9-12 are close, and anyone 13 and below are also so close. That’s why you can’t use seeds to accurately compare teams. Use the ratings, you will be better off with their accuracy.

Alrighty, using the big table above and our historical records, we can test the upsets. I did a lot of this when vetting the schedule to make sure it was the fairest it could be under the circumstances how the schedule is created. Teams picking their preferred opponents are certainly a wrinkle, but a fun wrinkle. I still like fun.

How close are these games?

The Gonzalez System’s success rate is about 4 in 5. In straight terms, there could very well be around 6-7 matches that would be technical upsets (when the lower rated team defeats the higher rated team). But we can get even closer.

We calculated the Mean exchange of a Predicted Upset based on past history. We have 228 technical upsets of 1193 games to pull from. The Mean Predicted Exchange of a Technical Upset is 0.729 (1.458 in double Nationals exchanges). The standard deviation was 0.213 (0.426 when doubled). Simply, 68% of all upsets had a predicted exchange between 0.516 (1.032) and 0.942 (1.883). This table expands it out:

Std Deviation from Mean Percentile Equivalent # of Games % of Schedule
0.5σ 38.29% 21 60.00%
1.0σ 68.27% 2 5.71%
1.5σ 86.64% 1 2.86%
2.0σ 95.45% 11 31.43%

We can see that 21 of our 35 matches have an exchange that is super close to the Mean for a predicted upset. These matches have the greatest chance of becoming an upset, based on past results. 60% of the schedule includes these super close matches!

Meanwhile, there are still 11 matches that, if an upset where to occur, it would be statistically significant. A 95th percentile occurrence equates to the odds being 1 in 20. And this 1 in 20 chance of a significant upset is part of the larger 1 in 5 chance of any technical upset occurring.

What’s the chance of Overtime?

I’ve been trying to predict Overtime’s occurrence for years, but it is actually quite elusive. Mainly because we don’t have enough data to really develop an accurate opinion, the OT numbers are a bit too insignificant to provide an accurate prediction on where it’s going to happen.

The straight chance of OT happening is 97 / 1193 (8.13% or 2.8 matches)
Total OT at Nationals is 18 / 239 (7.53% or 2.6 matches) but parity is closer this year and this season has trended towards more OTs.

So Overtime will occur. We just can’t accurately predict where.

I did try. The Predicted Exchange of an OT was a mean of 0.630, one deviation would cover rating gaps up to 6.749 between Gonzalez ratings. That’s a significant portion of our schedule. 23 of 35 (65.71%) matches fall within one standard deviation (68.27%) of the mean predictive exchange for historical overtimes. These matches could be potential overtime situations, but they might not. There’s not enough data to predict where.

Author: Zigmister

DePaul Dodgeball #68 &
NCDA Director of Officiating

Leave a Reply

Your email address will not be published. Required fields are marked *