. . .

ESPN’s FPI Is Less Than 50% Accurate: 2026 10-Week Study of FPI Validity; an ACADEMIC REVIEW not Artificial Intelligence

ESPN’s FPI Start

The credibility of a quantitative system which serves a purpose of subjectivity and credibility for the college football audience and human subjective voters to use as a determining factor for ranking analysis in their subjective Top 25, must possess some validity and credibility at the 95% rate of accuracy and efficacy. ESPN developed a Future Predictor Indicator (FPI) to assist in the comprehension of teams strength with the projecting of game and season outcomes for FBS programs. ESPN developed this FPI system, which originating for the 2013 FBS college football season. This FPI system is currently being used by ESPN to validate FBS college football teams credibility for the human subjective voters to use as a quantitative metric to rank FBS football teams week to week in either the AP, AFCA, CFP or FWAA polls. However, how efficient is the ESPNs FPI model, along with its accuracy rate to predict future outcomes for college football at the FBS level.

Defining ESPNs FPI Model

Retrieved from ESPN.com; the ESPN FPI is defined as “the Football Power Index (FPI) is a measure of a teams strength that is meant to be the best predictor of a team’s performance going forward for the rest of the season. FPI represents how many points above or below average a team is. Projections are based on 20,000 simulations of the rest of the season using FPI, results to date, and the remaining schedule. Ratings and projections update daily. FPI data seasons prior to 2019 may not be complete.” (ESPN, 2026). In essence from an academics point of view, this explains nothing concrete, except that it quantitative in determining outcome based results based upon percentage rates of success to the the future outcome from simulations of each seasonal game scheduled against opponents.

The ESPN FPI measures and collects the following data, which then creates a success percentage rate of opportunistic success in various variables or categories. Retrieved from the World Wide Web at (espn.com/college-football/fpi). Those quantitative and qualitative data variables that are being collected by ESPN’s FPI are as follow:

Win-Loss records of every scheduled game.

FPI- Football Power Index that measures the team’s true strength on net point scale, expected point margin vs. average opponent on a neutral field.

Rank- Rank order of FBS teams based upon the outcome results vs. all of the FBS teams.

Trend- FPI rank change from previous week.

Projected Win-Loss- Projecting the overall win-loss totals, accounting for results to date and FPI-Projections for the remaining schedule (and potential conference and championship games). May not sum to a whole number because of different numbers of games played in each simulation.

Win-out Percentage Rate- Percent of season simulations in which team won all remaining scheduled games including the conference championship game, if applicable.

6-Wins Percentage- Percent of season simulations in which team won at least 6 games making them bowl eligible.

Win Division Percentage- Percent of season simulations in which team won it s conference division, for those conference that have split divisions. (Note: This does not apply to any of the FBS conferences as of the 2026 FBS season).

Win Conference Percentage- Percentage of the season simulations in which team wins its conference, incorporating chance of getting to and winning the conference championship titles in conferences that allow them. (Note: All FBS Conferences possess a championship game as of the 2026 FBS season).

Playoff Percentage Rate- Chance to make the CFP Playoff , according to the playoff predictor.

Make National Championship Game Percentage Rate- Chance to make the CFP Playoff National Championship Game, according to the playoff predictor.

Win National Championship Game Percentage Rate- Chance to win the National Championship Game, according to the playoff predictor.

(ESPN, 2026).

This FPI “predictive model” by ESPN is being promoted by ESPN as significant and valid. Furthermore, ESPN uses this “predictive outcome model” to support Strength of Schedule (SOS) and Strength of Record (SOR) both of which the College Football Playoff Committee (CFP) uses to assist in their rankings in the last 5 weeks of the season for the 12-team playoff. The CFP committee and chairperson will deny that their ranking decisions “are not solely based upon the FPI% rates assigned week to week for their ranking assessment purposes.”

Success Rate of the ESPN’s FPI for FBS Football?

To date, ESPN has NEVER published a single consolidation win rate for the college FPI, but the model is to predict game and season outcomes using expected points added (EPA) and other factors, and has been used in multiple season with varying results depending, strength, injuries and and other variables (ESPN.com, 2026) Retrieved from (espn.com/blog/statsinfo/post/_/id122612/an-inside-look-at-college-fpi) published August 10. 2016. These are not my words but written by the ESPN analytics department who should possesses some quantitative and quantitative research based experiences on how success rates and data collection works to VALIDATE your information. This indicates that there is no valid accuracy success rate of an analytical, historical, longitudinal, qualitative and quantitative system in which is used by a well re-known sports media outlet. Who then provides that statistical information to the CFP Committee who are coaches, administrators and sports writers who at least have earned a Master’s degree or higher to have academic abilities to comprehend the tools being used to select and seed FBS teams who are competing for MILLIONS of dollars with the opportunity to compete for the National Championship in college football at the FBS level.

Peer Reviewed Published Academic Research on Predictive Models

To date there are NO published peer reviewed journal articles or research papers from academics specifically on the subject of examining ESPN’s FPI success rate or investigating the use of the FPI system to determine validity and credibility of the FPI’s use. What has been published about predictive outcomes and subjective ranking systems by doctoral professors and academics, was prior to the use of the ESPN FPI system design and implementation by ESPN. However, in 2025 there was a published paper titled “A Predictive Metamodel for College Football” by B Jay Coleman; published in the Journal of Sports Analytics, Vol 11 (0): 1-7.

Coleman (2025) used a Metamodel this is most predictive of future game results using 29 college football rating systems, data of 5925 games from 2016-2024, and a K-Fold cross validation process focused on the model predictiveness. Using a five-system metamodel to develop to predict the margin of victory for games in the ensuing week. Coleman (2025) determined that the model he used achieves a strong result vis-a-vis the opening, mid week and closing betting lines that is statistically significant in the presence of the opening line validation and test samples. Coleman (2025) believes the model he created is potentially useful for the CFP Committee, media, fans, betting markets and odds-makers providing a benchmark against which future rating systems can be assessed. Coleman (2025) stated that despite a large number of systems that have been developed for rating and ranking major college football teams, there are only 2 prior studies have attempted to identify and assemble of those models. This indicates to me that Coleman (2025) who is a strong academic and well known published researcher/publisher of other academic articles, examined, developed, created and uses other variables to create this metamodel system in which ESPN and their analytics team does not have access too.

Reading the Coleman (2025) published article in depth, Coleman states that FBS college football schedules only make up approximately 10% of the whole group for assessment. Coleman (2025) also states that the CFP committees selection of the current 12-Team playoff participants, the committee is inundated with multiple data points or “wide variety of data and information” to assist in their decision making process” that the committee struggles with making a decision. In previous published research by other academics; Fair and Oster (2007) and Coleman (2014), examined the Home Field Advantage (HFA) effect when FBS teams play more home games than away games. The development of a global binary HFA factor to all games, by Fair and Oster (2007) and Coleman 2014); possesses a statistical significance as a variable. However, the sample size each of these studies used derived from only games played in the latter half of the season with both studies using closing betting lines as the comparative.

Coleman (2025) methodology was data collected by Beck (2025c) using the 2016-2024 FBS seasons, with the OMITTING of the FBS season during the COVID season of 2020. Coleman (2025) examined how Beck used every college football season over the last 25 seasons in which Beck (2025c) provided an a priori compilation of margin of victory predictions for upcoming games which also included the opening, midweek and closing betting lines. Furthermore, Beck (2025c) used comparative performances of those sources, including straight up and against the spread, each week as well as cumulatively during the respective season. Beck (2025c) also compared his results to 29 other rating systems from credible analytical systems. Beck (202c) model estimation and variables selection were performed via forward selection and a six-fold cross validation using a predictive sum of squares (PSSE) across the holdout samples of each fold.

How does the research performed by Coleman (2025) and Beck (2025c) compare to the data collected and variables examined in which is being performed by myself as an independent researcher, during the course of the 2026 FBS season.

2026 FPI Data Collection and Variables

As a certified independent researcher, I have always expressed interest to examine the efficacy and accuracy of the ESPNs FPI system. The primary reason why I am examining the ESPN FPI system is because the CFP committee reviews the FPI as a metric for the CFP 12-team playoff and could use this metric for the CFP 24-Team Playoff. To determine how to evaluate, assess and collect the data to determine the efficacy and accuracy of the ESPN FPI, I had to read and understand the variables that determine how the FPI works. This was already established in previous paragraphs. I then needed to determine how to collect the data and what questions to investigate to validate or prove not significant for use of criterion for strength of schedule (SOS) or strength of record (SOR) evaluation.

The first aspect that needs to be clarified is where the data collection derives from. ALL quantitative qualitative data was retrieved from ESPN.com or the ESPN APP where the FBS college football data is published by their metrics team. Prior to the start of EACH FBS/FBS GAMES ONLY, the data is that is collected associated for the percentage rate of the “matchup predictor” which is a published in numerical value of which FBS team prediction percentage rate or chances of winning the game are. That “matchup predictor” establishes a starting point of confidence. In addition to the “matchup predictor”, the betting line/point spread prior to the start of the game is also collected. Once, that information is retrieved and documented, its is then placed in an excel spread sheet based upon the labels in which the data is being documented. As an independent researcher there needed to be an established predictive points confidence or margin of victory (MOV), associated with the “matchup predictor”. The purpose of establishing the “predictive points confidence” variable was to determine if the “matchup predictor” possessed confidence in relationship to the final outcome of actual games score. Another way to determine the “predictive points confidence” variable can be explained as Margin of Victory (MOV) associated with the “matchup predictor” variable.

The table below explains the quantitative, numeric value based points of “predictive points confidence” or Margin of Victory (MOV) in relationship to the FAVORITES “matchup predictor” successful percentage rate:

50%- 54%= 1-3 POINTS MOV (Indicates a close game and either team can win)
55%-59%= 4-7 POINTS MOV (Indicates that the FAVORITE should win by at least 1 possession)
60%- 64%= 8-11 POINTS MOV (Indicates that the FAVORITE should win by 1 possession and almost Double Digits)
65%-69%= 12-14 POINTS MOV (Indicates that the FAVORITE should win by almost 2 possessions with Double Digits)
70%-74%= 15-17 POINTS MOV (Indicates that the FAVORITE should win by 2+ possession with Double Digits)
75%-79%= 18-21 POINTS MOV (Indicates that FAVORITE should win by 2+ or at least 3 possessions with Double Digits)
80%-84%= 22-25 POINTS MOV (Indicates that FAVORITE should win by 3 possessions with Double Digits)
85%-89%= 26-29 POINTS MOV (Indicates that FAVORITE should win by almost 4 possessions with Double Digits)
90%-94%= 30-34 POINTS MOV (Indicates that FAVORITE should win by more than 4 possessions with Double Digits)
95%+= 35 POINTS or more MOV (Indicates that FAVORITE should win by 5+ possessions or more always Double Digits)

Once the “matchup predictor” percentages were paired with “predictive points confidence” or the MOV outcome numerical values, this will enable the rest of the quantitative numeric values to assist with establishing confidence or no confidence in the use of the FPI% system by ESPN.

Once the FBS college football schedule were published, I collected the weekly games scheduled by FBS/FBS GAMES ONLY. There is published research and proof that has proven that FBS/FCS scheduled games were determined as NOT SIGNIFICANT for this FPI study, since FBS/FCS scheduled game data indicates that FBS programs win games scheduled with FCS programs by almost 6 possessions, with a win-loss percentage rate close to (.950) during the CFP 12 Team Playoff Era. Each FBS/FBS game was labeled or categorized with FBS teams as Away or Home along with their “matchup predictor” percentage rate established by the ESPN analytics staff. In addition to the betting line spread and the final outcome result of the game. ALL FBS/FBS games data were collected prior to the start of each games start time. The table below demonstrates how the FBS Team name and quantitative data was collected in an excel spreadsheet.

AWAYFPI% RATEHOMEFPI% RATESPREADFINAL
FBS TEAM A65.4%FBS TEAM B34.6%(+/- 7.5)35-17

Once the pregame data was collected, the FBS team with the lowest FPI% rate; their FPI% rate was BOLDED. This would help identify and separate the FAVORITE and underdog label of each FBS team. The betting spread line was also noted as a (+/- Point Value). Each FBS/FBS games final outcome was collected as FBS TEAM A-FBS TEAM B. That allowed the ability to line up the points values associated with the Away or Home Team. Once the FBS/FBS game was completed with a final score, the winning FBS team was color coded GREEN, whereas the losing FBS team was color coded RED. There were other quantitative variable that were coded once the FBS/FBS game was completed. However, there were questions that were needed to be established to determine the accuracy, credibility and validity (ACV) of the ESPN FPI% system.

Questions Asked to Assist in (ACV)

There were a list of questions that were developed to assist in the (ACV) and challenge the ESPN FPI% system. Each question was to address if the published quantitative data possesses a correlation to the “predicted model” versus the actual outcome. Determining to prove the efficacy of the quantitative values assigned to each FBS Team in relationship to the betting point spread and the final outcome. Within the first 3 Primary Questions, an accuracy percentage rate was needed to validate the success rate of each variable in relationship to the outcomes and quantitative data collected. The questions are:

Q1? Does the Point Spread/Betting Line, match the FPI % rate by ESPN in direct relationship to the “predictive point confidence” MOV of points related to the FPI% rate (VARIABLE 1)? This question was to determine if there was a direct correlation between the point spread/betting line and predicted points confidence (MOV) points by the FAVORITE, since the FAVORITE is established by the ESPN FPI% rate “matchup predictor” is accurate? Example: if FBS Team A is Favored by 65.4% as the example above, then the final outcome based upon the predicted points confidence MOV points table indicates that the FAVORITE should win the game by 12-14 points. If the FPI% rate did NOT meet the point spread, then the point spread/betting line for quantitative data was color coded RED as a Not Significant. For example if the FBS Team A that is listed as the FAVORITE with 65.4% FPI% rate, and the point spread/betting line is (+/-4.5) then we have a quantitative numeric value data violation in correlation with the point spread/betting line and the predicted MOV.

Q2? Does the outcome/final meet or exceed the FPI% rate by ESPN (VARIABLE 2)? This question was to determine if the final score outcome won by the FAVORITE correlates with the FPI% rate. In the example above, 35-17 was the final score/outcome. FBS Team A was the FAVORITE winning by 18 points. That meets and exceeds the criteria for a positive FPI% rate and the score would remain colored BLACK as a Significant. If the final outcome results for the FBS/FBS game, returned that the FAVORITE lost the game; then the final score/outcome data collected would be colored RED as Not Significant, to signify that the outcome and FPI% rate for the FAVORITE returned a negative result.

Q3? Does the outcome match the point spread/betting line associated with the FAVORITE (VARIABLE 3)? This question was to establish a more significant accuracy to the FPI% rate and the point spread/betting line to increase the efficacy of the ESPN FPI% rate. In the example, 35-17 with the point spread/betting line (+/-7.5) along with the FAVORITE FPI% rate at 65.4%,then the point spread/betting with the final outcome meets and exceeds the question and can be determined a Significant. The final outcome score in the excel box will be color coded LIME GREEN as a positive outcome to validate the ESPN FPI % rate as Significant. However, if the the final score/outcome was 35-31, or the FAVORITE lost the game, then the excel box will be colored LIGHT GREY as a negative outcome and decreases the (ACV) of the ESPN FPI % Rate system.

Q4? Straight up win-loss records for both away and home FBS teams, excluding the spread? This question is to establish the win -loss records of both the away and home teams to correlate which FBS team wins and losses more; the away team or the home team? This derives from the final score/outcome. The excel box of the winning FBS team was color coded GREEN, whereas the excel box with the losing FBS team was color coded RED. Color coding allowed for easier data identification collection and analysis.

Q5? How many FBS Teams both away and home with an FPI% rate of less than or equal to 50%, had the final score/outcome end in their favor? This question is to establish how many FBS teams with an established FPI% rate of less than or equal to 50%, win games to provide a negative validation to the FPI% rate of ESPN ability to be accurate to support that their FPI% rate predictive model possesses validity. This quantitative data is based upon only the final score/outcome of the game in direct relationship to the FBS team listed with less than a 50% chance of winning the game.

Between these questions, the answers and quantitative data will provides enough evidence to support the validity of the FPI% predictive model system or provide a negative validation providing that the FPI% predictive model possesses no validation. Indicating that the FPI system that ESPN sponsors and published should not be used as part of a source of use to determine ranking or selection of seeding for the college football playoff.

Preliminary Results

To clarify, all results will be expressed in basic numeric values of data which includes percentage rates and whole numbers. The general college football audience would never understand if the results were expressed in P-Value or F-Value. Those values are for the academics such as doctoral professors and educators who understand statistical analysis in those values. All FBS/FCS games were eliminated from the quantitative data collection. The results below are of quantitative data collected through first 10 weeks of the 2026 college football season at the FBS level. Through the 10 weeks of the FBS college football season, there have been a total of 562 FBS/FBS game data points collected.

The preliminary results associated with Variable 1 (V1) of Yes, 78 of the 562 FBS/FBS games played or 13.8%, the point spread/betting line matched the FPI% rate associated with the predicted “predicted” MOV. The preliminary results associated with (V1) of No, 483 of the 562 FBS/FBS games played or 85.9%, the point spread/betting line did not match the FPI% rate associated with the “predicted” MOV. This preliminarily indicates that the point spread/betting line, with the “matchup predictor” FPI% rate along with the “predicted” MOV scale; possesses NO credibility or validity to support the use of the ESPN FPI “matchup predictor” FPI% rate. Further review of this quantitative data in relationship to V1 can be further examined for other possible correlations or non-correlations.

The preliminary results associated with Variable 2 (V2) of Yes, 227.5 of the 562 FBS/FBS games played or 40.5%, the final result/final outcome did meet or exceed the FPI% rate associated with the FBS team that was FAVORED to win the FBS/FBS game. The FAVORED FBS team based upon the ESPN FPI% predictive rate won the game by final result/final outcome with the point spread/betting line. The preliminary results associated with Variable 2 (V2), of No, 333.5 of the 562 FBS/FBS games played or 59.3%, the final result/final outcome did not meet or exceed the FPI% predictive rate associated with the FBS team that was FAVORED to win the FBS/FBS game. The FAVORED FBS team based upon the ESPN FPI% predictive rate did not cover the point spread/betting line associated with the final result/final outcome. This preliminarily result indicates that although there is some success with the FAVORED FBS team with the ESPN FPI% predictive rate. This can clearly indicate that the FAVORED FBS team does not always cover the point spread/betting line, but can wins the game as a FAVORED FBS team with the higher ESPN FPI% rate.

The preliminary results associated with Variable 3 (V3) of Yes, 271 of the 562 FBS/FBS games played or 48.2%, the final result/final outcome does match the FAVORED FBS team. This is best explained as the FAVORED FBS team with ESPN FPI% predictive rate with the point spread/betting line; won the game by more than the point spread and more than or equal to their ESPN FPI% rate. The preliminary results associated with Variable 3 (V3), of No, 290 of the 562 FBS/FBS games or 51.6%, the final result/final outcome does not match the FAVORED FBS team. This is best explained as the FAVORED FBS team with the ESPN FPI% predictive rate with the point spread/betting line; either lost the game or won the game, but did not cover the point spread/betting line. This indicates that the FAVORED FBS team does not always win the game and or cover the point spread/betting line as the FAVORED team associated with the ESPN FPI% rate.

The preliminary results for all FBS/FBS games based upon final results/final outcomes straight up. The overall record of all FBS/FBS games played through 10 weeks of the 2026 FBS college football season, the HOME FBS team is (345-219) with a win-loss percentage rate of (.612). This would be considered below average for FBS teams who play or listed as the HOME FBS Team. What needs to be examined is what is the win-loss percentage rate for FBS football during a certain period of time to determine if FBS HOME teams win more games. With 3 weeks remaining in the FBS college football season, that will increase in overall wins but will only slightly increase in win-loss percentage ratings.

The preliminary results for all FBS/FBS games played in which the winner of the game where the FBS team was listed for both home and away, in addition to a FPI% predictive rate of less than 50%. The preliminary results determined that FBS team listed as home were (66-134) with a win loss percentage rate of (.330). This quantitative results associated with FBS teams listed as either away or home with a FPI% predictive rate of less than 50%, are capable of winning the game. This also indicates that the FPI% predictive rate published by ESPN is leaning lack of confidence, significance and efficacy for use. However, FBS teams listed as the away team were (86-275) with a win loss percentage rate of (.238). This indicates that when FBS teams are listed as the away teams and FAVORED, that the FBS FAVORED teams are only successful 67% of the time during this 2026 FBS season. Furthermore, the FBS teams that listed as the home teams and FAVORED, that the FAVORED home teams are only successful 76% of the time. With 3 weeks remaining in the FBS college football season, this quantitative data will not possess a significant change to increase the percentage rates in favor of the use of the FPI% system developed by ESPN.

Preliminary conclusion….

These preliminary results indicates that the ESPN FPI% predictive model for FBS/FBS games played during the 2026 FBS season through 10 weeks of data; indicates a lack of credibility and validity of it use and designed purposes. This preliminary data offers support in which ESPN’s approach that they do not have full confidence in their analytics to predict outcomes with 95% confidence. However, ESPN cannot verbalize the success of a system if they cannot provide empirical, exhaustive, historical, quantitative or qualitative data to to variables that would support its use with confidence and efficacy since its starting season in 2013.

As an independent researcher who has examined many variables associated with college football at the FBS level; I can determine from the preliminary results that the ESPN FPI% rate predictive model is not an accurate model for ESPN’s use to publish. With 3 remaining weeks left of the 2026 FBS season, if these results hold true then the quantitative data will support the 10 weeks already analyzed and evaluated as NOT SIGNIFICANT. That the ESPN FPI% rate possesses no credibility and validity. When the 2026 FBS season concludes on Thanksgiving weekend, I will publish the final findings with the week to week data in which I collected to support my findings and offer clarity of the data collected.

To AI…. Meta and Muse…. maybe you should promote ACADEMIC INTELLIGENCE rather than trying to figure it out… and hope you are right…

If you read my post, then decide to recite or share what I posted, always use MLA/APA citations. Giving credit to the author and their hard work

Follow me on X/Twitter: @cfbpoexpert on BLUESKY: cfbpoexpert.bsky.social

Please leave a comment if you wish.

All Information is protected by © Copyrights and Intellectual Property Rights.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.