After two seasons of winning only four games each, the 2025 New England Patriots have now exceeded that in seven attempts at 5-2. For Patriots fans, the team’s success stands as a huge contrast with the failures of coach Belichick in Chapel Hill, where his results have largely been disappointing. As most football viewers typically see it, New England’s current success is in large part despite a stretch of horrendous drafting at the hands of Belichick.
What isn’t mentioned is the role that recent some of Belichick’s picks have played in New England’s current success. Christian Gonzalez, the last first-round pick of Belichick’s tenure, has played like one of the best cornerbacks in the league. Christian Barmore, an interior defender that Belichick traded up to obtain, has been one of the team’s best defensive players this year. Defense is hardly the only side of the ball where a positive impact has been felt – two starting wide receivers in Kayshoun Boutte and DeMario Douglas have been reliable options for Maye this year. That’s not to ignore slot corner ace and punt returner star Marcus Jones either.
The recent success of Belichick draft picks had me question the common narrative of his perceived historic decline as a drafter. Was it possible that he merely had a cold stretch of picks? To answer this question, I built an expected value curve for each of the first 224 picks in the NFL draft based on 24 seasons of data.
Preparation
Due to an earlier project I pursued this year of creating my own draft value chart based on 21st century drafting data (2000-2020), I was able to use most of the same information I had already compiled from Pro Football Reference. The key difference this time though had to do with updating the data. Essentially, what I did was upload each NFL season’s worth of draft picks data for mostly the same period, making sure to add 2021, 2022, 2023 data. I then merged all of them together and kept the following features: year, round, pick, team, player, and DrAV. DrAV is a variant of Approximate Value – one of the earliest attempts to encapsulate player value in a single universal metric – but only for seasons where a player contributed on the team that drafted them.
With my previous research in mind, there were a few big problems with my old approach. First off, it’s not up-to-date on the more recent draft picks. More critically though, it’s extremely overfit by virtue of it excluding multiple outliers; the draft picks taken into account by our patterns are overwhelmingly in the IQR range (25th to 75th percentile), meaning that the large variance of drafting outcomes was not properly represented. This is an issue because it doesn’t capture the potential reward of gaining a star at any point in the draft 0 the most important factor that goes into a team’s drafting decisions. Lastly, the calculations for Expected DrAV are flat throughout our data set, which undervalues recent draft classes. This initially left me with a new task: make a new draft value chart with an updated methodology.
A New Expected Draft Value Chart
With DrAV as our dependent variable, as well as 24 seasons of draft data, I created a new column called Expected DrAV, which is calculated as a fitted value from a log-linear decay model. The reason I have chosen a linear log decay model is because a polynomial runs the risk of overfitting, and because outliers can distort our data into questionable conclusions. For example, Tom Brady being a success story in the 199th pick would inflate the value of such a pick versus the 198th pick. Therefore, I built this model assuming that an earlier draft pick is generally preferable to a later one.
Like last time I will also exclude outliers in which a drafted player with above 0 weighted AV did not play for the team that drafted them (such as Eli Manning or Philip Rivers), and then expand the data set to 2023. Although this excludes over 800 data points, I still had 5,256 instances to evaluate it. Finally, I decided not to exclude any outliers at all, since these are typically what teams are chasing more often than not.
Figure 1: Expected Draft AV by Pick Number
Our model explains roughly a quarter of variance in player outcomes (R-squared of 0.241). Though it’s not perfect, realistically speaking, a good model will only be moderately strong due to the incomprehensibly high number of factors that lead to career success for players on franchises that draft them. For now, it provides a strong foundation on which to evaluate Belichick as well as league-wide performance for drafting outcomes. 6,130 draft picks in total with roughly 871 outliers excluded among them.
Patriots vs. League Draft Value
With a new foundation for assessing team-wide draft success, I then decided to look at the New England Patriots’ during Belichick’s entire tenure against every other organization in the league. Using the existing data, I was able to examine the actual DrAV of the Patriots through their 24 drafts against their expected results versus the rest of the league. For this segment, I made sure to visualize the Z-Scores of each team’s AV Above Expected, since otherwise, recent years would disproportionately skew drAV values for the league much lower.
Figure 2: Patriots vs. League Draft Value Over Time (Normalized)
Here, there’s a fairly interesting story of someone who gradually went from greatest drafter in the league to consistently below average. After historic levels of early success – including drafting the greatest player of all-time with a sixth round pick, as well as landing first-round hit after first-round hit, Belichick regressed to the mean for the middle portion of his career before eventually hitting rock bottom near the end of his tenure.
It’s worth noting that the model only captures a little under a quarter of the variance (24%) in the original data set. However, what I think is inarguable is the fact that there was a decline from the beginning of Belichick’s tenure to the very end. In the next section though, I will visualize that dropoff and see if there is any statistical significance in Belichick’s drafting performance over time.
Measuring Statistical Significance Via Bins
For this section, I’ve split up my data into four bins: Phase 1, 2, 3, and 4. These correlate with six-season spans throughout 24 seasons. I chose these six-seasons as relatively illustrative sequential periods: the initial dynasty, the reload, the second dynasty, and then the end of the Brady/Belichick eras. For each period, I compared the Patriots’ DrAV Above Expected against every non-Patriots’ team’s DrAV Above Expected, making sure to normalize the scores for each period.
Lastly, I ran a series of ANOVA tests comparing each Phase to each other to assess the question of Belichick’s decline being statistically significant or not. Since this model had an F-Statistic of 8.116, I felt fairly certain about this model’s ability to detect statistical significance.
Figure 3: Net Value Difference by Era Phase
Belichick’s draft efficiency fell by roughly 113.0 AV from his early dynasty years to the post-Brady decline. From Phase 1 to each of Phase 3 (p=0.0245) and Phase 4 (0.0026), there is a statistically significant decline, while Phase 2 to Phase 4 (0.0430) and Phase 3 to Phase 4 (0.0206) each showcase statistically significant declines in drafting outcomes as well.
All in all, Belichick began his time with the Patriots drafting 1.31 standard deviations above every other organization. He eventually ended it drafting 0.57 standard deviations worse than them, reflecting a drop of 1.88 standard deviations across 24 years. On one hand, though weak, it’s not statistically significantly worse than league average; on the other though, this seems to indicate that Belichick’s early drafting success relative to league average was an outlier to the broader pattern of him performing mildly above league average to decently beneath it, though not significantly away from league norms.
Draft Efficiency Ratio
As a way of validating the expected value and Z-Score analyses, I developed a new metric called Draft Efficiency Ratio. This is a standardized measure of how efficiently each team converts draft capital into career value (DrAV in this methodology) on a season-to-season draft variation. The results will provide a more normalized and accurate assessment of Belichick’s drafting record.
Figure 4: Draft Efficiency Ratio Over Time
My findings mostly coincide with the previous results, but they also offer additional insight into the expected outcomes of drafting as a whole. Cumulatively, New England’s mean normalized DER above league average (+0.183) would suggest moderately above average drafting results across all 23 seasons, but the variation in drafts is fairly large, with 2005 (+1.23) as the best year for Patriots drafting on a per-pick basis and 2022 (-0.45) as the worst year. I think this passes the smell test; in 2005, New England drafted a Pro Bowler in the first round (Logan Mankins), three starters in subsequent rounds (Ellis Hobbs, Nick Kaczur, James Sanders) and even a future journeyman quarterback in Matt Cassel within the seventh round. Meanwhile, 2022 had the infamous Cole Strange overdraft, and the aforementioned Marcus Jones is the only remaining player on the 2025 Patriots.
Figure 5: Average Draft Efficiency by Era Phase
Breaking DER into four six-year phases reinforced the same arc observed in the ANOVA and Z-Score tests: a gradual, but statistically meaningful erosion of efficiency over time (though with a slight increase in performance from Phase 2 to Phase 3 rather than mild decline). In Phase 1 the Patriots held a massive advantage (+0.60) in being more efficient with draft picks than the standard team. Over time, with adjustments from opponents in Phase 2 (+0.084), the Patriots struggled to retain their edge (+0.084), though they eventually retained a slight lead in Phase 3 (+0.154). Phase 4, however, is where they took an even further leap into becoming a less efficient drafting team (-0.087).
On one hand, a decline in about 0.67 standard deviations may not be catastrophic. Though it’s notable, the data hardly proves that Belichick had totally lost his ability to assess talent. Moreover, in previous periods where the Patriots had seen their once-massive edge slipping away, they had adjusted to at least retain a slight advantage. However, to see where the edges faded over time, I looked at positions drafted in each of the aforementioned four phases. This shed light on both the timeless qualities of Belichick’s draft performances, as well as what aged less gracefully.
Where Did New England Lose Its Edge?
Figure 6: Positional Draft Efficiency Alignment
There’s no getting around the fact that Phase 1’s heavy success was in large part due to football’s equivalent of winning the lottery with a Tom Brady selection; conversely, the post-Brady era was a flop in large part because of the Mac Jones fiasco. But the key story here is less about what most people already know about Belichick’s consistent strengths (linebacker) and weaknesses (skill position players), and more about where his former strengths dwindled away.
If anything stands out, it’s Belichick’s decline in the trenches as a drafter for offensive linemen and defensive linemen. What ultimately doomed Belichick as a drafter was the fact that consistent picks like Damien Woody, Matt Light, Logan Mankins, and Sebastian Vollmer slowly turned into Antonio Garcia, Isaiah Wynn, and Cole Strange. In similar fashion, Belichick’s defensive line picks turned from Richard Seymour, Ty Warren, and Vince Willfork to far more modest, with Keion White, Christian Barmore, and Chase Winovich being a backup player, a decent starter, and a bust. Together with Belichick’s consistent struggles around the offensive skill positions in the modern era, it was clear that his positional alignment in drafting was away from the rest of the league, which contributed to his worse drafting.
Limitations & Considerations
Any quantitative assessment of drafting success must acknowledge the limitations inherent to simplified models built from public-facing data. Primarily speaking, the metric used throughout this study is a variant of Approximate Value (DrAV), which itself is focused around an individual player’s most recognizable contributions – not necessarily adjusted for impact on teammates or scheme fit. While the model does reflect outcome efficiency, its use of a relatively dated metric to evaluate player contributions does flatten the player-to-player differences.
A second limitation comes in the form of survivorship bias. Due to the nature of our data set, recent players are still undervalued in spite of Z-scoring by season. In similar fashion, aggregation of team-level data across phases still treats picks equally, which doesn’t account for draft-day or draft-pick trades, an additional variable to consider when it comes to assessing the Patriots’ drafting outcomes under Belichick.
Lastly, the model is simple by design, as it paints drafting success with a broad brush. It does not adjust for the quality of draft class in a given season or existing roster needs per team. Our calculations for phase-level ANOVA and Z-Score comparisons assume stable league baselines and a normal distribution of draft pick evaluation even though league-wide trends could shift drastically and follow a different form of distribution. In fact, this research doesn’t account for Belichick’s own role in shaping the successes or failures of his draft pick as a coach, as well as roster turnover in his managerial staff, coordinators, or other players.
Future research on this topic would require more features input into a draft picks table, as well as drafting teams’ prior records and performances in team-wide metrics. This would allow me to incorporate machine learning techniques (like Gradient Boosting or Random Forest feature importance) in order to capture nonlinear interactions between draft slot, position, long-term value, and existing team needs. For now though, this report should be read as an analytical approximation of the Patriots’ gradual decline in draft efficiency under Belichick, as well as a look at if, how, and why he lost his drafting touch. In my opinion, these limitations highlight that this captures relative efficiency fairly accurately.
Conclusion
After my wedding date, the happiest day of my life was February 1, 2015, when I went with my father to Arizona to watch the Patriots beat the Seattle Seahawks in the Super Bowl. It is no exaggeration to say that Belichick is directly responsible for some of the happiest moments I’ve ever experienced, as well as the careers of many of my favorite players. In fact, his approach to management and football brilliance has shaped my love for football analytics.
Who else would have turned Jamie Collins into one of the best linebackers in the league, or seen something special in Rob Gronkowski, Vince Wilfork, or Sebastian Vollmer? Who else in 2009 had the guts to go for it on fourth-and-two in overtime against Peyton Manning? Although Belichick himself is reportedly not big on analytics use in front offices, the irony is that modern-day football analytics partially descend from a system of thinking that Belichick popularized as the blueprint for success.
This project started off as an exercise in data analytics skills for an independent study. However, the more I worked on it, the more I realized it was really a story about legacy. The very analytical framework that I’ve used to measure his decline would not be possible without witnessing Belichick’s career-long search for his sport’s hidden edges. His decline is ironically the final proof of his greatness – through changing football itself, he ensured his destiny: a day where the game finally caught up to him.
Appendix