2007 NFL Week 12 Predictions Summary

NFL Predictions listed on this page are based on the default settings of the Team Rankings GameZone NFL predictions tool, a 100% objective and data-driven system developed by Stanford engineers. Default settings loosely reflect our research into which statistics tend to be good predictors in most cases.

We encourage our users to manipulate the predictions tool for each matchup of interest, and to customize individual factor settings to test alternative strategies for predicting NFL outcomes. Although we will provide performance tracking information at the default settings, keep in mind the following guidelines:

• The system does not incorporate or attempt to model the potential effects of injuries, weather conditions, coaching strategies, specific player matchups, emotional factors such as team morale, or anything else not listed in the statistical factors box. Making any corresponding adjustments to predictions will require your own judgment.
• Every data-driven predictive system has a margin of error associated with it, and this system is no different. We believe that a 3% to 5% safety buffer around odds to win/cover is warranted, as is a 1.5 to 3 point buffer around expected margin of victory. For example, if odds to win for a given team are calculated at 53.5%, margin of error means that the actual win odds calculation may be as low as 48.5% or as high as 58.5%, based on the variance inherent in our data set and models.
• Every matchup is different, and tweaking the factor settings for a matchup based on the specific characteristics of the two teams in question may improve prediction accuracy. Simply adjust the importance sliders and time filters for any factor, and prediction information will update.
• Try to give the "Help" section a good read. It's lengthy, but it will enable you to understand how this product works and how to use it best. We're confident that if you read the help section, you'll be convinced of the analytical power and sound mathematical foundation of this tool.
• Please remember that luck and probability mean that anyone can win or lose big over one week, month, or even season. The key is to adopt a consistent strategy that you can measure, refine, and improve over the long term, and to filter emotion out of the process of making picks. This product can provide the basis to do so.
• Have fun, and please let us know your comments, questions, suggestions, and feedback!

Last Week In Review

At default settings, the prediction tool's analysis went 5-3 in 60%+ odds to cover games during Week 11, bringing total ATS performance in 60%+ odds games since its pre-Week 3 launch to 37-24 (60.7%). (See Season to Date NFL Prediction Results.) Note: We measure our ATS performance based on closing spreads, and spreads can and do change after we publish the weekly predictions summary commentary. To see closing spreads and results, you can click back to any Week 11 game using the drop down box in the top right of the screen. Week 11 60%+ odds to cover winners were Tampa Bay, Minneapolis, New England, Cleveland, and Philadelphia.

Default settings analysis slightly underperformed our benchmarks picking winners at the lowest 50%+ odds to win threshold, going 11-5. It was an odd week in that default analysis indicated four odds to win predictions that were very close to even (San Francisco-St. Louis, Arizona-Cincinnati, New Orleans-Houston, and New York Giants-Detroit), and three of these four games went the other way. Since launch, the tool is now 85-43 (66.4%) predicting straight up winners at the 50%+ odds to win threshold, still several percentage points better than our expert and consensus picks benchmarks.

We received several emails last week regarding the New England-Buffalo game. The common question: "How is it possible that a team's odds to cover can be higher than 50%, when the tool's calculations only favor that team by 12.1 points, and the Vegas point spread for that team is -16? It seems like there is an error in the calculations."

In reality, there is no error. Occasionally, and especially in the case of large spreads, this counterintuitive result can occur because we evaluate a team's expected win margin and cover odds as both dependent and independent events. Without getting too technical about it, think about it this way...

Imagine that in 100 similar historical games, the teams like the Patriots won 80 of them. The teams similar to the Pats won 60 games by 20 points each, won 20 games by 3 points each, and lost 20 games by 3 points each. Also, teams like New England faced a variety of different point spreads in those 100 games, and they managed to cover 70% of the time.

Using the simplest possible math and looking only at historical data, you would say that the Patriots have 80% odds to win the current game being analyzed (since teams like the Pats won 80 out of 100 similar games), and an expected win margin of 12 points (the weighted average of the win margins of the 100 similar historical games, assuming each game carries equal relevance/importance).

Now you find out that the spread for the Patriots in the current game is -16. So what are their odds to cover that spread? On the one hand, their expected win margin is only 12 points. However, in similar historical games, the Patriots won by 18 points 60% of the time, and a repeat of that performance would cover a -16 spread. In addition, you also know that the Pats generally covered the spread (which was not always -16) in similar games 70% of the time, another relevant data point that suggests greater than even odds for the Pats to cover, as long as most of those historical spreads were at least close to -16.

The math built into the tool is far more complicated, of course, but we hope this simplified example helps to answer the basic question. And as we all know, the Pats managed to blow away their expected win margin yet again. It's starting to make us wonder why Vegas still offers lines for their games...

In closing, one of the things we have been thinking about a lot lately is how to improve the selection and quality of statistical factors incorporated into the prediction tool. In this first version, we decided to keep it simple and, beyond the two power ratings included, use basic statistics that most people were familiar with. However, we know that there are more powerful stats out there that would enhance the analytical power of the GameZone predictions tool -- and its predictive power too. We've already received a number of informative emails on this subject so please let us know which factors you think would be good additions.

Shoot us an email with your thoughts and when you get a second, please check out our fantasy football player projections for Week 12, which should be posted by Tuesday evening.

NFL Week 12 Preview

One thing to watch out for this week is of course the truncated rest periods for teams playing Thanksgiving Day games, although with bye weeks now a thing of the past, we would be careful about making any assumptions about less rest affecting certain types of teams more than others. Right now, we don't have the appropriate data on rest days to make that sort of a judgment with a high degree of confidence.

Now that we are later into the season and power ratings are converging, we are seeing more and more games where the Vegas spread appears to be efficient, which is the way it should be. If your system is finding angles on 16 of 16 NFL games, then it's a safe bet that something is wrong with your system, since the Vegas lines are never inefficient that frequently.

An interesting study this week will be momentum. A season power ratings based system will always lag in "recognizing" teams who rapidly turn a bad season around and start consistently performing well, since their crappy early season performance can never be erased from the historical results on which power ratings are based. On the other hand, it is next to impossible to predict which underperforming teams truly have turned a corner while those teams are still in the early stages of such a turnaround. For example, most people jumped on the Saints bandwagon when they ripped off four straight wins. Then they lost to the Rams (ouch) and Texans. So much for a turnaround.

This week, we have some underperforming teams (hello Rams) with seemingly positive momentum, and some overperforming teams (hello Titans) with seemingly negative momentum. So do we assume that those trends will continue and make extra adjustments to predictions as a result, or do we believe that the last two games for these two teams were more likely just a blip on the radar screen?

A season power ratings based system effectively favors the latter approach, since in general, two wins and eight losses still makes a bad team. However, GameZone users wanting to make a more educated decision can modify the prediction tool's settings to decrease the importance of power ratings and increase the importance of other factors set to "Last 3 Games." See if the prediction calculations change significantly. If they do, you may have uncovered a situation where, historically, momentum seems to have mattered in the outcomes of similar games.

NFL Week 12 Predictions Summary

Below is our Team Rankings GameZone predictions summary for all NFL Week 12 games. Although we update this page periodically, predictions can change multiple times per day based on new information such as shifting NFL betting lines. Click on "View Predictions" to see the most recent lines and predictions and to customize factor settings.