SNAP College Football Predictive Model Deep Dive

Whether you primarily take in my content on here or across my social media accounts, you’ve probably heard of my SNAP CFB Predictive Model by now. But what exactly goes into making all the ratings, projections, etc? All that is explained (to the best of my ability, without giving the formula away) in this article.

Subscribe Below!

Subscribe

During the season, I’ll be posting my best bets every week (starting week five or six, once the model gets accurate enough) on all my socials, but if you want every single game projection with detailed matchup analytics, all you have to do is either go to my TikTok page and subscribe for just $3 a month, or subscribe at the bottom of this page for FREE! All subscribers on either platform will get full access to what I mentioned above, PLUS requests for hypothetical matchup analytics. For example – if you’re an Alabama fan and want to know what my model would think about a matchup with Clemson? Just go to the contact form saying so, and I will email you with everything you’d need to know about the matchup within two business days. Or if you’re a TikTok subscriber, just post your request on the subscriber message board. Anyways, enough ranting about subscribing, let’s get into how the SNAP Model works!

How It Started

Growing up, I always had a gift for numbers and was always a grade level or two ahead of my own in math. With my gift for numbers and love for sports came my passion for sports analytics. I’d always be comparing player stats and looking at percentages of winning games, winning conferences, etc. In the fall of 2021, my senior year of high school, I thought to myself: “Why not make my own analytics?” So that’s exactly what I did. I took the sport I was most passionate about, college football, and paired it with my gift with numbers and knowledge of google sheets, to create my first ever power ratings. Now I’m not going to give away any of my formulas here, but those power ratings were basically a very simplified, less accurate version of the SNAP Model we see today, that used only scores and opponent scores to calculate ratings and predict matchups.

I went on using that model through the ‘23 season, but in the spring of ‘24, I thought about improving the model. After some time to think about how I could make the numbers more accurate, I decided to dive into some research on the most influential, but predictable stats in college football. I looked at every game across the country from the last decade, tracked every stat and correlated them to wins and losses to find what I was looking for. What I found was that the four most influential and predictable stats in the sport are conversion percentage, completion percentage, yards-per-pass and yards-per-rush. With this knowledge, I still used the same formula as before to generate projected scores, but then looked at advantages and disadvantages teams may have in certain matchups to create my hypothetical “best bets” (I have never done any actual betting with this model, as I was under 21 until this past offseason and currently work for Iowa State Athletics). That was when I struck gold. I got over 57 percent of my “best bets” correct over the course of the season, the best I’ve done in the four years of its existence and better than most similar rating systems out there.

I thought that was going to be it. I thought this would just be a cool little private thing I did on my own, and maybe one day I could use this to bet games on the side of my actual job if that main job wasn’t working for a team. But then I got the idea to make it public. How could I make it public? Through a blog site and social media. And that, ladies and gentleman, is how Snap To Tipoff was born. I didn’t just want this brand to be about college football betting though, which is why you see my other content on here and my social media, but that’s not what this story is about.

I knew if I made my model public, I would have to try and improve it even more, and make it so that the four stats I mentioned earlier would be present in the projected scores. So from about April through June, I was trying to find every way possible to improve the accuracy of the ratings, along with finding the linear regression for of the four stats to assign them each point values. I put these changes, along with slight tweaks to the formulas, through countless field tests to gauge the accuracy on last year’s games.

After going through that, I then had to come up with a name for the model before I could make it public. I wanted to relate it in some way to my brand name, so I was throwing around names in my head for a few days before it came to me. The Statistical Numerical Analytical Predictive Model, or SNAP Model. Once all of that was figured out, I was finally able to make the model public on here and my socials.

How It Works

This model is the same type of thing you see from ESPN’s FPI, Bill Connelly’s SP+ and college basketball’s KenPom. It gives every team a  “neutral field predictability rating,” so for example if a team has a rating of 10, that team is about 10 points better than an average FBS team on an average day. We can then use this rating to project future results using a matchup calculation that predicts score, chance of winning and chance of covering the spread for each matchup.

This is purely a numbers game. I created the formulas myself, but I never manually change any ratings or game projections. As I’ve learned, the numbers usually know best, even if something looks strange. With that being said though, this system isn’t perfect, as it’s impossible to predict every game correctly. It’s an imperfect game played by imperfect players that changes from week to week. The goal is just to get more right than wrong throughout the season.

Like this content? Subscribe Below!

Disclaimer

If you are using my model to bet on games, do not bet money that you can’t afford to lose. As I mentioned above, it’s not perfect, and there will be games that it gets wrong. It’ll likely even have bad weeks here and there, that’s just part of this stuff. I’d also highly recommend not parlaying my picks. As I also mentioned above, the goal isn’t to get all of them right (although that would be great), the goal is to get more right than more wrong and end up in the positive by the end of the season. If you’re addicted to gambling and need help, please call 1-800-GAMBLER for 24-hour service.