Saturday Feelings - Week 4
Oh man... this week I put a lot of hours into the model. If my boss at work is reading this, I promise I spent time on this only after working hours!
Many, many, many adjustments and additions made to the model including completely changing the output variable, adding new factors, and creating a probability measure. Read more about that here.
But now it’s a new week and I’m feeling confident in how the model will be predicting outcomes. Hopefully the probabilities along with the predictions will help garner a better understanding of how the model makes its guesses but we have to wait and see. Obviously, at the end of the day there is no way to predict with 100% accuracy what will happen in an NFL game but using these statistics and probabilities should give somewhat of an edge.
Week 3 in Review:
Not going to beat around the bush, I was very disappointed with the predictions for Week 3. I spent hours scraping data, training and testing a machine learning model, predicting outcomes and putting them on nice graphics for instagram all for a COIN TOSS to predict with the same accuracy. I was really frustrated. I really think the fact that I have hundreds of people reading my predictions keeps me honest and pushes me to create statistical models that actually have value.
There really was nothing worth celebrating for Week 3 if I’m honest. We had a clean sweep for the Steelers game, but that was not really much of a win considering how many incorrect predictions there were elsewhere.
One nice thing about the big L’s I took this week was the fact that I got some encouraging messages on instagram from supporters who reassured me that the model was fine and that no one really could have predicted the upsets for Week 3. It was nice to see that many people understand that predictions are just that: predictions. They use statistics and probability but at the end of the day, outcomes are very random and sometimes completely unpredictable.
I am cautiously optimistic that the model will perform better. I am also happy that I’ve included probabilities as this shows that these predictions have a chance of being completely incorrect. I hope that people can see this. I am not one of these shady sites that advertises you’ll “win big” and ask you to put a number of units on a bet and if you do that you’ll get rich. That doesn’t make sense. The sportsbook always wins. This is for the casual bettor who likes watching sports, wants to learn more about analytics, and enjoys placing bets here and there. Excited to see what Week 4 has!
Week 4:
The predictions this week all make sense to me! There is nothing too crazy and there are no predictions I can’t see happening.
The three predictions that have me wondering the most are the Buccaneers to only beat the Chargers by three, the Ravens to beat the Football Team by 11 and the Rams to beat the Giants by 10.
The Buccaneers prediction is interesting because every game this season the model has predicted the Buccaneers to do poorly (for the most part) and every game the model is off. I know that the model will start adjusting to this incoming data as time goes on, but I want the predictions to start lining up better!
The Ravens-Washington and the Rams-Giants games are interesting to me - both these favorites are favored to win by two touchdowns. The model will almost never predict a team to win by that much. And if it does, it’s almost certainly going to be true*. However, looking historically at similar matchups, there are instances in both directions where a team wins or loses by much more or much less. So the idea that a team winning by 14 or more is historically “rare” for these predictions is something to think about. In general, teams winning by 14 or more is sufficiently rare - blowouts don’t happen all that often in the NFL. But 14 sounds like so much until you watch the Ravens score three TDs out the gate and realize immediately that they might cover. We’ll see how these predictions play out and how, statistically, it is technically unlikely for teams to score like this.
The model is also predicting the Raider to upset the Bills. In terms of Raiders upset predictions, the model is 1-0! I am hoping to make that 2-0 by the end of the weekend. I really think it could happen too. Even though the Bills are 3-0, they won a close one to the Dolphins (31-28) and they narrowly beat the Rams last week after being up by 21-3 at half and by leading by as much as 25 points. They allowed 29 unanswered points and just barely came out on top. I think it will be a good game and I can definitely see the Raiders winning.
There’s not much else to do now but sit and wait for football! If you read all of this I thank you and encourage you to donate to our Patreon page linked here or our Cashapp account $bettorpicks. I do everything for free and any support is much appreciated.
As always, good luck bettors!