Prediction Methodology

The complexity and scope of the analysis embedded in NFL Premium predictions go far beyond the free NFL predictions on TeamRankings.com.

The free NFL predictions on the site are based solely on the calculated power ratings of each team. We devised these power ratings to be "fair." They are an estimate of how good teams are, based on how they have performed so far this season. However, they only take into account past game scores, whether the games were home/away/neutral, and when the games were played. The free predictions are essentially the difference between the power ratings of the two teams playing. While they perform pretty well, they do not use a very large data set of historical information as an input.

NFL Premium predictions, on another hand, incorporate a wide array of historical information dating back for over 10 years.

NFL Premium predictions leverage unique predictive modeling techniques and software we've developed and refined over the past several years. They take into account many important factors and stats, several of which are either underutilized or flat-out ignored by many other handicappers out there (and, we're fairly certain, by Vegas too!). For example, how much travelling has a team done recently? How have they done in their last ten games? How does that compare to how they have done in their last two games? How have the two teams matching up fared against one another in the past? How have similar teams who matched up against each other in the past fared? These are all important questions in trying to forecast a game, and the key is combining them in the right manner. In addition, it takes a complex analytical system to comb through all of the available historical data and draw all the parallels and correlations -- a process that is simply impossible to do manually. If you've read Moneyball, you have a sense of what we're doing here. The numbers do the talking.

NFL Premium includes two separate predictions. The first predicition is the probability of each team covering the Vegas point spread.

This model was originally developed on data from the 1992-2001 NFL seasons, and it did very well the following three years. We continue to refine it on a nearly weekly basis, incorporating the latest NFL results as they happen.

The second prediction is the final game score.

This prediction computes the expected score, which means that it may not quite sync with the spread probability. Occasionally, there may be a case where the Bengals are favored by three points over the Steelers, for example, and we predict that the Bengals have a 55.6% chance of covering the spread. However, the expected outcome might be a Bengals win by only two points. Such a prediction implies that although we say that the odds are 55.6% that the Bengals win by more than three, the bulk of this 55.6% is going to be a win by just barely more than 3 points. On the other hand, the other 44.4% represents a relatively high likelihood that the underdog Steelers win in a blowout.

We realize all of this may be somewhat complicated and confusing for the average sports fan, but as mathematicians, we'd rather err on the side of being accurate than being simplistic or gimmicky. (Another reason why we can't stand terms like 'locks,' 'can't-miss picks,' 'sure winner,' etc.) Our hope is that this philosophy, which permeates the entire site, casts Team Rankings in a positive light with our users and customers. We are about the numbers -- sound sports analysis that can be directly validated by statistics and facts. Once you filter out the short-term effects that luck can have on making successful predictions, truly analytical modeling is the only way to evaluate and minimize the risk of being wrong.

Even with our NFL Premium preditions, we will certainly not be right 100%, 90%, 80%, or 70% of the time. We may not even be right 50% of the time for a given season. But back-testing our NFL Premium models has shown that over the long run, our 60%+ predictions are correct more than 60% of the time, and we have the data to prove it.

Sign up now
Copyright 2000-2008 Team Rankings, LLC. All rights reserved. All content on this site is produced for entertainment purposes only and is not intended to be used in direct or indirect violation of any local, state, federal, or international law(s). By using this site, you agree and consent to our Terms of Service and Privacy Policy.