2021 Thanksgiving Football Odds and Predictions

This year I count myself thankful for many things, the Bears-Lions game on Thanksgiving day is simply not one of them. To prepare everyone for what to expect in this year’s thanksgiving games, we fired up our NFL model (using our Bayes Ensemble method) for the first time this year to understand the NFL landscape and give you some Thanksgiving football odds. Using this, We’ll make some predictions for the three thanksgiving day games and give our overall NFL rankings so you can find where your team slots into the overall NFL power rankings. At the end, I’ll include a short description about how our model computes rankings for those who are curious.

These predictions and odds are the result of our model which would be the fourth most accurate NFL model being tracked on ThePredictionTracker.com. As our model is more accurate than the opening, midweek, and final lines in Vegas, this model could give some insights not found in the general media consensus.

Thanksgiving Football Odds and Lines

We’re treated to a slate of three games on Thanksgiving Day 2021: Bears-Lions, Raiders-Cowboys, and Bills-Saints. Luckily, two of the three games are projected by our model to be pretty close. Our model functions by assigning a team a score (typically values living between +/- 10, average is 0) and the difference between two teams’ scores is a predictor for the margin of victory on a neutral field. Then, the neutral field line is adjusted 3 points in favor of the home team.

Let’s start with Bears-Lions. Our model rates the Bears as the 7th worst team in the league at -4.6 and the Lions as the second worst team in the league at -10.3. On a neutral field, this implies a Bear’s margin of victory of 5.7 points. But, the game being in Detroit shifts our predictions to Bears -2.7. Using standard point spread to winning probability calculators, this gives the Bears around a 56% chance to win. Money lines more favorable than Bears -130 or Lions +130 should be favorable. This matches Vegas’ lines pretty perfectly.

The second game of the day is Raiders-Cowboys. The Cowboys come in at a 4.8 while the Raiders are -2.7. Thus, the neutral site predicted spread is Cowboy by 7.5. Because this game is in Dallas, this updates our model to Cowboys -10.5. This is quite a bit more favorable for Dallas than the current Vegas line of Dallas -7.5. This means that some favorable bets should be available. Converting to winning probabilities gives the Cowboys an 86% chance of winning. For the money lines, you should be looking out for bets more favorable than Cowboys -625 or Raiders +500.

The final game – really the marquee game of the day – is Bills @ Saints. My model is very high on the saints this year, definitely higher than national consensus. They are certainly weird, though, with wins over Tampa, Green Bay, and New England but losses against the Falcons, Giants, and Panthers. On the other hand, the Bills might be just as strange. The Bills routinely flip back and forth between huge, convincing wins and absolutely questionable performance like the Jaguars debacle. So, who really knows what is going on with either of these teams.

My model has the Saints rated at 6.0 while the Bills are the second best in the league rated 8.1. This implies a neutral field margin of Bills -2.1. But, the game being in New Orleans leads me to predict New Orleans -0.9. This is a pretty significant deviation from Vegas’ Bills -6.5 number. This converts to New Orleans having about a 51 or 52% chance of winning. We’re looking for money lines then more favorable than New Orleans -110 and Buffalo -110.

2021 Week 12 NFL Power Rankings

Our model so far this season is really accurate. The model has correctly predicted head-to-head winners about 64.5% of the time so far this season. This puts it as the fourth most accurate model tracked on the internet – better than ESPN’s FPI, Jeff Sagarin, Pythagorean ranking systems, and the opening, midweek, and closing Vegas lines. This is pretty good for a model that doesn’t look at anything other than the final score. The table below contains the overall rankings and ratings for every NFL team so far this season.

	Team	Rtg
	Arizona Cardinals	8.2
	Buffalo Bills	8.1
	Tampa Bay Buccaneers	7.9
	New Orleans Saints	6.0
	Green Bay Packers	5.1
	Indianapolis Colts	5.1
	New England Patriots	5.0
	Dallas Cowboys	4.8
	Los Angeles Rams	4.7
	Kansas City Chiefs	3.9
	Seattle Seahawks	3.8
	Tennessee Titans	3.5
	San Francisco 49ers	3.3
	Baltimore Ravens	2.6
	Minnesota Vikings	1.1
	Pittsburgh Steelers	0.8
	Los Angeles Chargers	0.4
	Philadelphia Eagles	0.4
	Carolina Panthers	0.1
	Cincinnati Bengals	-1.2
	Denver Broncos	-2.5
	Cleveland Browns	-2.7
	Las Vegas Raiders	-3.7
	Washington Football Team	-3.8
	Miami Dolphins	-4.0
	Chicago Bears	-4.6
	New York Giants	-4.8
	Atlanta Falcons	-6.6
	Jacksonville Jaguars	-8.4
	Houston Texans	-8.5
	Detroit Lions	-10.3
	New York Jets	-13.6

Methodology

Our model is Bayesian meaning that it combines prior guesses about team quality with observed data to try to estimate how good a team is. Our prior guesses come from last year’s final rankings. The rationale is that a team’s quality at the end of the season last year is a pretty good starting guess for a team’s quality this year. The prior is weighted so that a team’s previous year quality is equal to five games of data.

Thus, if no games have been played so far, the prior does all the work; predicting winners in week 1 is done entirely by picking whoever our model thought was a better team last year. Predicting winners in week 5 is half who was better last year and half how good they’ve been so far this year. As the current season goes on, more and more weight is put onto the current season of data as memory of the past fades.

The rating for each team is a measure of quality. The difference between two teams’ ratings is a prediction for the margin of victory on a neutral field. The ratings are determined by looking at the past margins of victory and figuring out the ratings that best explain what we’ve seen before. Then, we mix in the proper amount of prior and we get our final guess for how good a team is. In this way, our model weights for both strength of schedule, quality of win (margin of victory), and prior expectations about team quality. It zips all three of these quantities up into a single number that is a pretty good guess for how good a team is.