How did it start? Like most things, it started with a tongue-in-cheek remark that led to hours of going down various rabbit holes and eventually resulted in what you see today.
My husband and I are both involved in athletics in one way or another. We were talking about the season and how KHSAA creates RPI ratings. That system seems to make more sense for larger sports like football than it does for smaller sports like soccer. Knowing the tech and spreadsheet nerd in me, he mentioned that he thought I could probably come up with a better system.
Challenge accepted.
Starting with Spreadsheets
At first, I started experimenting with spreadsheets. I played around with different metrics, including winning percentage, strength of schedule, and margin of victory. The more I worked on it, the more I realized how many different variables could be included, especially depending on the sport.
On top of that, there used to be a website called Youth Soccer Rankings. It was a really cool site where you could essentially predict the outcome of any two youth teams based on a score they assigned. I actually reached out to them for their algorithm, but they politely declined to share it.
Even so, I still thought it would be a great idea to create some kind of metric where you could compare two teams, look at their rating scores, and try to predict who would win. So I kept experimenting in spreadsheets and eventually came up with a system I liked, although it has been adjusted a million times since then.
Using AI to Build the Website
At the same time, I was also starting to experiment with artificial intelligence. I had not used it much before, and I had never coded. AI seemed like it might be able to help, and it definitely did.
Long story short, I ended up with a website. While AI was very helpful in the beginning, it also seemed to make things worse every time I wanted even a minor tweak. Eventually, though, I was able to move the site from one hosting platform to the current one while keeping everything mostly the same.
The Goal of Zow Ratings
The goal now is to keep updating results and see how those changes affect rating scores throughout a season. While rankings are fun to look at, my main focus has really been finding a way to predict outcomes.
My confidence stats show that these systems can accurately predict results at around 75% to 90%, depending on the sport.
The Challenge of Predicting Margin of Victory
Predicting goal differential, however, has been much more challenging. That makes sense when you compare sports with very different scoring structures.
- Basketball and football can have very large margins of victory with no true ceiling.
- Volleyball has a much narrower scoring format, since scores are reported by sets and matches rather than a wide point margin.
- Soccer and lacrosse seem to fall somewhere in between, which may make them a better fit for this kind of modeling.
At least for now, those middle-ground sports seem to offer the best balance between preference and accuracy.
What Comes Next
I am sure there will be more changes to come. For now, though, I plan to keep things relatively stable and continue watching how the predictions perform.
I also tried adding NCAA men's and women's basketball, but that turned out to be incredibly difficult because of the sheer number of teams and conferences. If you're curious, you can still check out those results in the archived section. It is interesting to see how some teams stayed near the top of the rankings even when, looking at all the metrics, it did not necessarily seem like they should. At this point, I think I have even lost track of exactly how all the metrics interact to produce the final rating score.
Final Thoughts
That is a little bit about how Zow Ratings got started. Feel free to use the contact us option to let me know what you think.