Which Projections Are Most Accurate?

By Jesse Kartes | Last updated May, 2026

In this analysis, we dive into the performance of various fantasy football projection sources for seasonal fantasy football. Using Mean Absolute Error (MAE) to measure accuracy, R² to measure variance explained, and Mean Error (ME) to evaluate directional bias across twelve seasons of data, we explore which sources consistently provide the most accurate projections and where systematic biases exist. We break down the results by position, examine how accuracy and calibration have shifted in recent years, and offer final thoughts and recommendations for which sources fantasy managers should rely on moving forward.

Methods to Evaluate Projections

To assess the accuracy of fantasy football projections across various sources, we used three complementary metrics. Mean Absolute Error (MAE) serves as our primary accuracy metric, calculating the average absolute difference between projected and actual fantasy points over a full season. A lower MAE indicates more accurate projections. R² (R-squared) measures how much of the variance in actual player performance is explained by the projections, where higher values mean the projections better capture the spread of outcomes across players. Finally, Mean Error (ME) which measures directional bias. A positive ME indicates the source over projected (players underperformed their projections on average), while a negative ME means the source under projected (players outperformed their projections).

It’s important to note that these metrics capture different things. A low ME does not necessarily indicate accuracy. A source that over projected one player by 100 points and under projected another by 100 points would show an ME near zero despite wildly inaccurate individual projections. Similarly, R² can be misleadingly low when restricted to a narrow range of players (such as only the top projected players at a position), because there is less between-player variability for the model to “explain.” ME should always be interpreted alongside MAE and R² to get the full picture.

The analysis covers the 2014 to 2025 NFL seasons and focuses on the top 20 projected QBs and TEs, and the top 40 projected RBs and WRs for each source. We focused on this subset of players because they are the ones fantasy managers are actively choosing between during drafts, making them the most relevant group for evaluating projection accuracy. We evaluated two different sets of projections from Fantasy Football Analytics (FFA):

  • FFA Average: A simple average of projections across all available sources.
  • FFA Weighted Average: A weighted average based on the accuracy of each source in prior seasons. Sources with lower MAE in previous seasons were given greater weight, ensuring the most accurate projections had a stronger influence. Specifically, weights were calculated as the inverse of each source’s historical MAE (1/MAE), then normalized within each position to sum to one. Each season’s weights are calculated using all available prior data. This method prioritizes sources with a strong track record of accuracy, giving them greater influence on the weighted average. FFA Weighted data is available starting from 2015.

To ensure fair comparison, we focused our analysis on sources with sufficient historical data. For historical rankings, all eleven qualifying sources were included. For the “last three seasons” analysis (2023–2025), we only included sources that appeared in all three seasons. WalterFootball was excluded from the recent analysis due to missing 2023 data. Two sources were excluded entirely due to limited data availability: Yahoo (6 qualifying seasons) and FantasyData (2 seasons).

It’s worth noting that both FantasyPros and FFA aggregate projections from multiple individual sources. FantasyPros uses a consensus approach, while FFA offers both a simple average and a weighted average based on historical accuracy. We have included FantasyPros in our analysis for transparency and completeness, but readers should keep in mind that the strong performance of both FFA and FantasyPros reflects the power of aggregation rather than a single proprietary model.

The eleven sources included in our historical analysis were CBS, ESPN, FFA Average, FFA Weighted, FFToday, FantasyPros, FantasySharks, NFL, NumberFire, RTSports, and WalterFootball. For the last three seasons, ten sources qualified (all except WalterFootball). This methodology allowed us to evaluate the consistency and accuracy of each source while considering different strategies for combining projections, both simple and weighted.

Who Has the Best Projections?

Quarterback Projections

Quarterback has become the hardest position to project accurately, with error rates climbing sharply in recent seasons as the position has evolved. Despite this, aggregation-based approaches have proven most resilient. FantasyPros led QB projections over the full historical period with the best overall accuracy (61.0 MAE), followed closely by FFA Average (61.7 MAE). Both outperformed all individual sources, reinforcing the power of combining projections. FFA Weighted ranked third (63.0 MAE), with CBS (63.1 MAE) and FFToday (63.6 MAE) rounding out the top five. FFA Average delivered the top QB projections in 2020 and posted a top-four finish in eight of its twelve seasons. CBS had standout accuracy in 2014, 2015, and 2019 but saw its performance decline sharply in recent years, falling to eighth in the last three seasons (82.1 MAE). ESPN posted impressive early results, including the best QB projections in 2016 and 2017, but ranked last among all sources over the last three seasons (86.7 MAE).

In the last three seasons, NFL emerged as the top source for QB accuracy (71.2 MAE), followed by FantasySharks (72.5 MAE) and FFToday (74.0 MAE). FantasyPros (74.7 MAE), FFA Average (75.9 MAE), and FFA Weighted (76.1 MAE) finished fourth through sixth. The 2022–2025 period has been particularly challenging for QB projections across the board, with MAE values rising significantly for all sources. This likely reflects increased quarterback volatility driven by the continued evolution of the passing game and rushing QB production. NumberFire and RTSports consistently struggled at the position, ranking near the bottom in both the historical and recent analyses.

Running Back Projections

Running back remains one of the most volatile positions in fantasy football, but it is also the position where projections explain the most variance suggesting that while individual outcomes are noisy, the overall structure of RB production is more predictable than other positions. CBS led RB projections over the full historical period with the best accuracy (52.2 MAE across 11 seasons), followed by FantasyPros (52.4 MAE) and FFToday (52.8 MAE). FFA Average ranked fourth (53.4 MAE), tied with ESPN (53.4 MAE). FFA Weighted finished sixth overall (54.1 MAE). CBS earned the top spot in 2018 and 2025, while FFToday led in 2015, 2017, 2021, and 2022. FFA Average claimed the best projections in 2014, while FantasyPros led in 2019, 2020, and 2023. NFL (57.3 MAE), FantasySharks (59.1 MAE), and RTSports (62.8 MAE) rounded out the bottom of the rankings.

Over the last three seasons, CBS maintained its lead (54.2 MAE), with FantasyPros close behind (56.1 MAE). FFA Average and FFA Weighted finished in a virtual tie at 56.9 MAE each, followed by FFToday (57.6 MAE). The FFA projections showed remarkable consistency in the RB space, finishing in the top five for accuracy across recent seasons without the sharp year-to-year swings that other sources experience. RTSports continued to lag behind the field (64.3 MAE), finishing last among all qualifying sources.

Wide Receiver Projections

Wide receiver is the most tightly clustered position for projection accuracy, with the top seven sources separated by just 2.1 MAE points historically. This compressed field makes consistent top-half finishes especially meaningful. FantasyPros was the clear leader historically, posting the lowest MAE across its eleven qualifying seasons (40.2 MAE). FFToday ranked second (41.1 MAE), followed by ESPN (42.0 MAE), CBS (42.1 MAE), FFA Average (42.2 MAE), and FFA Weighted (42.2 MAE). The FFA projections again demonstrated their consistency at the position as the FFA Average finished in the top three for WR accuracy in four of twelve seasons and delivered the best projections in 2018 and 2023. NumberFire (42.3 MAE) tracked closely, while NFL (44.9 MAE), FantasySharks (46.2 MAE), RTSports (49.2 MAE), and WalterFootball (53.7 MAE) struggled throughout.

In recent seasons, FFToday continued its dominance at WR (40.9 MAE), followed by FantasyPros (41.9 MAE) and NFL (42.8 MAE). FFA Average (43.8 MAE) and FFA Weighted (43.9 MAE) finished fourth and fifth, maintaining their steady presence near the top. CBS dropped from fourth historically to ninth in the last three years (47.5 MAE), illustrating the kind of volatility that individual sources can experience. RTSports again finished last (58.0 MAE), more than 17 points worse than the leader.

Tight End Projections

Tight end has historically been one of the most difficult positions to project accurately due to heavy touchdown dependence, yet the recent trend is encouraging with most sources posting improved accuracy in the last three seasons compared to their historical averages. FantasyPros led TE projections historically with the best overall accuracy (31.4 MAE), followed closely by FFToday (31.5 MAE), NumberFire (31.9 MAE), ESPN (31.9 MAE), and FFA Weighted (31.9 MAE). FFA Average ranked sixth (32.3 MAE). The gap between the best and worst sources is substantial with more than seven points separating FantasyPros from WalterFootball (38.5 MAE). CBS (33.8 MAE) maintained a middle-tier ranking, while NFL (35.0 MAE), FantasySharks (36.2 MAE), RTSports (37.2 MAE), and WalterFootball brought up the rear.

Over the last three seasons, FFToday maintained a narrow lead (29.4 MAE), with FantasyPros (29.5 MAE), ESPN (29.6 MAE), and NFL (29.6 MAE) all within a single point. NumberFire (30.2 MAE), FFA Weighted (30.4 MAE), and FFA Average (30.5 MAE) finished close behind. The TE position saw overall improvement in accuracy across recent seasons, with most sources posting lower MAEs than their historical averages. CBS (34.3 MAE), FantasySharks (34.4 MAE), and RTSports (36.3 MAE) continued to struggle, finishing in the bottom three.

Variance Explained (R²)

While MAE measures the average size of projection errors, R² tells us how much of the variance in actual player performance is explained by the projections. Higher R² means the projections better capture the differences in scoring outcomes across players.

R² varies considerably by position. Historically, running backs show the highest R² (20–28%), likely because their production is more volume-dependent and workload is somewhat predictable from preseason expectations. Wide receivers fall in the middle (14–19%), while QBs (7–15%) and TEs (16–26%) round out the picture.

Historically, FantasyPros posted the highest R² at QB (15.1%), CBS led RB (27.8%), NumberFire topped WR (18.7%), and RTSports led TE (25.5%). FFA Average ranked second at QB (14.4%) and WR (18.1%), fourth at TE (19.7%), and fifth at RB (25.9%), demonstrating strong explanatory power across positions. FFA Weighted tracked close to FFA Average across all positions, with its strongest showing at RB (23.7%).

Over the last three seasons, NumberFire posted the highest R² at QB (12.0%) and WR (16.7%), CBS led RB (30.2%), and ESPN led TE (28.0%). FFA Average ranked second at QB (6.3%), fifth at WR (14.3%) and TE (19.9%), and sixth at RB (26.4%). FFA Weighted tracked close to FFA Average across all positions, with its strongest showing at RB (26.5%).

The relatively low within-position R² values across all sources highlight why minimizing error matters so much because even the best projections explain only a fraction of within-position variance, making aggregation strategies like FFA’s particularly valuable for reducing noise. For fantasy managers, this means that even using the best available projections, you should expect significant surprises. The value of projections lies not in perfectly predicting outcomes, but in tilting the odds in your favor across many decisions over the course of a draft and season.

Calibration: How Projections Exaggerate the Spread

Beyond overall accuracy and variance explained, it’s worth examining how projections miss. Calibration plots, which chart projected points against actual points, reveal systematic patterns in projection errors. In a perfectly calibrated system, all data points would fall along a 45-degree line where projected equals actual. Deviations from this line reveal where and how projections go wrong.

Using FFA Average projections across all twelve seasons, a clear pattern emerges at every position, projections exhibit an exaggerated spread. The highest-projected players tend to underperform their projections, while the lowest-projected players in our sample tend to meet or slightly exceed theirs. Statistically, this shows up as a regression line with a slope less than 1.0 and a positive intercept. At QB, the calibration slope is just 0.67, meaning for every additional projected point, actual scoring increases by only about two-thirds of a point. Wide receivers show the strongest calibration (slope 0.85), with running backs (slope 0.79) and tight ends (slope 0.72) falling in between. 

This exaggerated spread pattern is not unique to FFA and is present across all projection sources. The implication for fantasy managers is straightforward, the gap between a source’s top-projected player and their fifth-projected player is almost always smaller in reality than on paper. Elite projected players carry more downside risk than their projections suggest, while lower-ranked starters often offer more upside than expected. This doesn’t mean you should avoid drafting top players, but it does mean that the perceived drop-off from, say, RB3 to RB8 is likely overstated by raw projections.

Projection Bias by Position

Having established the structural pattern of exaggerated spread, let’s examine how directional bias varies across individual sources. We measured bias using Mean Error (ME), calculated as predicted minus observed. A positive ME indicates the source over projected (players underperformed their projections on average), while a negative ME means the source under projected (players outperformed their projections). In the seasonal context, the biases are consistently positive and substantial across all positions and sources, meaning projections tend to be too optimistic for the top projected players. This reflects the downside risk of projecting elite players, whose underperformance outweighs the modest overperformance we see from lower-ranked starters. For a comparison of how these seasonal biases differ from the smaller, more balanced biases observed in weekly DFS projections, see our companion article, Which DFS Projections Are Most Accurate?

Quarterback Bias

Quarterbacks show the most dramatic increase in over projection in recent years. Over 12 seasons, the average QB ME across all sources was +22.5 points, meaning the top 20 projected QBs consistently underperformed their preseason projections. In recent seasons (2023–2025), this bias has nearly doubled to +46.5 points. The calibration plot reveals that this bias is not uniform across the projection range. QB has the lowest calibration slope of any position (0.67), indicating the most exaggerated spread between projected and actual outcomes. The highest-projected QBs tend to fall short of their projections by the widest margin, while lower-projected QBs in the top 20 come closer to meeting expectations. This pattern, where the top of the range underperforms more than the bottom, is what produces the large positive ME at the position. RTSports posted the highest historical QB bias (+44.5 ME) and the most extreme recent bias (+64.6 ME), while NumberFire also showed heavy over projection (+58.8 ME recently). On the other end, FFToday was the most conservative QB source both historically (+9.6 ME) and in recent years (+32.8 ME), suggesting their methodology may better account for QB variance. FFA Average posted a moderate historical bias (+18.9 ME) and recent bias (+45.6 ME), falling in the middle of the pack.

Running Back Bias

Running back bias has remained relatively stable over time. The historical average ME across sources was +23.2, and recent years show a similar pattern (+21.9 ME). The calibration slope at RB (0.79) is the second closest to 1.0, but the highest-projected RBs still tend to fall short of their projections by a meaningful margin. As projected points decrease, RBs come closer to meeting or even exceeding expectations, which is consistent with the positive overall ME being driven primarily by over projection at the top of the range. RTSports consistently over projected RBs the most (+41.0 ME historically, +40.1 ME recently), while FantasySharks (+30.7 ME) also showed significant positive bias. FFToday was again the most conservative source (+14.2 ME historically, +13.6 ME recently), while FFA Average (+21.8 ME historically, +20.2 ME recently) and FFA Weighted (+22.2 ME historically, +20.6 ME recently) tracked closely with the league average. CBS posted one of the lowest biases in recent years (+17.4 ME), consistent with their strong RB accuracy.

Wide Receiver Bias

Wide receivers show consistent positive bias across all sources, with a historical average WR ME of +24.4 that increased slightly to +26.8 in recent seasons. The calibration plot confirms the familiar pattern, the highest-projected WRs tend to underperform their projections, while lower-projected WRs in the top 40 come closer to meeting or slightly exceeding their expected output. The calibration slope at WR (0.85) is the closest to 1.0 of any position, suggesting that WR projections are the most well-calibrated in terms of spread, even though the overall directional bias remains positive. WalterFootball showed the highest historical WR bias (+42.2 ME), while RTSports (+37.8 ME historically, +49.0 ME recently) continued to be one of the most aggressively over projecting sources. FFToday (+15.1 ME historically, +16.3 ME recently) was again the most conservative, while FFA Average (+22.2 ME historically, +26.4 ME recently) and FFA Weighted (+23.0 ME historically, +26.8 ME recently) were near the middle. NFL posted the second-lowest recent bias (+22.3 ME), pairing well with its strong recent WR accuracy.

Tight End Bias

Tight end shows the lowest overall bias of any position, with a historical average ME of +17.6 and a recent average of +15.7. This is somewhat counterintuitive given how volatile TE scoring can be but likely reflects the smaller scoring range for the position. The calibration slope at TE (0.72) is the second lowest of any position, behind only QB (0.67), indicating a heavily exaggerated spread between projected and actual outcomes. RTSports (+27.2 ME historically, +28.6 ME recently) and FantasySharks (+22.2 ME historically, +21.3 ME recently) showed the most over projection. FFToday was the least biased TE source (+11.5 ME historically, +8.7 ME recently), followed by NumberFire (+12.5 ME historically, +15.1 ME recently). FFA Average posted a moderate bias at +14.6 ME historically and +14.3 ME recently, closely tracking with FFA Weighted (+14.5 ME and +14.4 ME). NFL stood out in recent years with the second-lowest TE bias (+11.1 ME).

The Value of FFA Projections

If one theme runs through this entire analysis, it’s that aggregation works. The two aggregation-based approaches in our study, FFA and FantasyPros, consistently outperformed individual sources across positions, time periods, and metrics. This isn’t a coincidence. By combining projections from multiple sources, aggregation smooths out the idiosyncratic errors that plague any single model. Individual sources may occasionally claim the top spot for a specific position or year, but they often experience sharp declines in accuracy that make them unpredictable from season to season. CBS, for example, ranked first for QB projections in 2019 (35.0 MAE) but fell to eighth by the last three seasons (82.1 MAE). ESPN produced the best QB projections in 2016 and 2017 but finished dead last in recent years. This kind of volatility is the norm, not the exception, among individual sources.

FFA’s edge lies in what it builds on top of the aggregation foundation. While FantasyPros offers a consensus approach, FFA provides a full analytical toolkit designed for competitive fantasy managers. The FFA Weighted Average lets you leverage historical accuracy to give more weight to sources that have been right more often and while our analysis shows the simple average has a slight edge historically, the weighted approach offers a distinct advantage in seasons where source accuracy does persist. And if you prefer to choose your own weighting, the web app allows you to set each source’s weight manually, upweighting or downweighting any source as you see fit. Beyond projections, FFA’s web app enables league-specific scoring adjustments so projections reflect your league’s actual settings, not generic defaults. Combined with tools for auction values, snake draft optimization, and the accuracy analysis shown in this article, FFA provides the most comprehensive platform for turning projections into actionable draft strategy.

Both FFA Average and FFA Weighted deliver top-tier accuracy without the catastrophic drops that plague individual sources. FFA Average ranked in the top six at every position historically, placing in the top five at QB, RB, and WR, and both FFA projections finished in the top five for average MAE across positions in recent seasons. Among individual sources, no single source matches this level of cross-positional reliability. FFToday excels at WR and TE but is less dominant at QB. CBS leads at RB but has dropped off significantly at QB and WR in recent years. NumberFire performs well at TE but struggles at QB and RB. Only the aggregation-based approaches deliver consistently competitive accuracy across the board without a significant weak spot.

In any given season, you can expect the FFA projections to perform at or near the top for all positions. No other platform combines this level of consistent accuracy with the analytical depth to help you act on it.

Interesting Observations

In a 2015 article on this site, we highlighted several key findings about fantasy football projections. With twelve seasons of data now in hand, it’s worth revisiting those observations to see which have held up and which have evolved.

The average of sources is more accurate than individual sources. This remains true and is perhaps the most robust finding in our analysis, consistent with the principle of the wisdom of the crowd. FFA Average outperformed individual sources in 69% of head-to-head comparisons across all positions and seasons, finishing with an average rank of 4.2 out of 11 sources. Both FFA and FantasyPros, the two aggregation-based approaches, consistently outperformed individual sources and FantasyPros posted the most top-three finishes (28 out of 44), while FFA Average posted 16 out of 48. The aggregation principle continues to be one of the most reliable strategies for improving projection accuracy.

The weighted average and simple average are essentially interchangeable. The original 2015 analysis suggested the weighted average was slightly more accurate than the simple average. Over our twelve-season sample, this finding has not held up. FFA Average (47.4 MAE) outperformed FFA Weighted (47.8 MAE), and FFA Average won 64% of head-to-head comparisons. This suggests that source accuracy does not persist reliably from year to year where a source that performed well last season is not guaranteed to do so again. Because the weighted average gives more influence to historically accurate sources, it can be penalized when those sources regress. The simple average, by treating all sources equally, provides more robust protection against this kind of volatility. In a world where individual source accuracy fluctuates unpredictably, equal weighting may be the safer and simpler bet. This finding is consistent with what we observe in our DFS analysis, where the two approaches also perform nearly identically.

Projections explain a meaningful but limited share of variance. Our 2015 analysis reported that the weighted average explained roughly 60% of the variance in players’ actual performance. In our current analysis, focusing on the top projected players at each position, FFA Average explains about 14–26% of within-position variance depending on position (highest at RB, lowest at QB). The difference from the 2015 figure is largely methodological as our current sample restricts to the top 20 QBs/TEs and top 40 RBs/WRs, which reduces variability and naturally lowers R² through range restriction. The 2015 figure included all projected players, where the spread between high and low scorers is much wider. The takeaway remains, projections provide meaningful signal, but a substantial portion of fantasy outcomes remains unpredictable, which is precisely why aggregation is so valuable.

Projections are more accurate for some positions than others. This continues to hold, though the details have shifted. In terms of raw MAE, tight ends have the lowest error (31.4 for the best source), and quarterbacks have the highest (61.0), but this largely reflects differences in scoring scale. When expressed as relative error (MAE divided by average projected points), QBs show the lowest relative error at 24%, followed by WRs (30%), TEs (33%), and RBs (34%). Running backs show the highest relative error, consistent with their reputation as the most volatile position. In terms of R², running backs show the highest within-position explanatory power (25.9% for FFA Average), likely because their production is volume-dependent and workload is somewhat predictable. The 2015 article noted that QBs and WRs were most accurately projected and this still holds when using relative error, as both positions benefit from more stable roles and target/snap distributions.

Projections overestimate players’ performance. The 2015 article reported about 5–6 points of overestimation on average across positions. Our current analysis shows this bias has grown substantially with the average ME across all sources and positions now +21.6 points. However, this increase is largely explained by sample composition. Our analysis focuses exclusively on the top projected players, who are inherently more likely to fall short of expectations than to exceed them. The 2015 analysis included all projected players, where the lower projected tiers include more players who outperform their modest expectations, pulling the overall ME closer to zero. Nevertheless, the direction of the bias is unchanged and projections remain systematically optimistic for the players fantasy managers are most actively drafting. This is particularly pronounced at quarterback, where recent over projection has reached an average of +47 points in the last three seasons.

Wrapping Up

The most reliable fantasy football projection sources are those that consistently produce accurate projections across all positions and seasons. While some sources may shine in a particular year or at a specific position, the FFA projections stand out for their steady top-tier accuracy year after year, no matter the position. This consistency makes them an excellent tool for fantasy managers seeking dependable projections that minimize risk and avoid the drastic fluctuations that come with relying on a single source.

Our analysis also reveals important structural patterns in how projections miss. All sources exhibit exaggerated spread, overstating the gap between their top and bottom projected players, and all sources systematically over project the top tier at every position. Understanding these patterns helps fantasy managers interpret projections with appropriate context that the difference between the first and fifth projected player at a position is almost certainly smaller than projections suggest, and late-round options carry more upside than their modest projections imply. The R² analysis further reinforces this, showing that even the best projections explain only a fraction of within-position variance, which is precisely why aggregating sources, as FFA does, is a valuable strategy for reducing noise.

But don’t take our word for it. You can use the web app to evaluate historical accuracy for yourself. Looking for an analysis of weekly DFS accuracy? Check out our companion article, Which DFS Projections Are Most Accurate? For a deeper dive into positional bias and how it affects your draft strategy, see our projection bias article.

Stay Connected with Fantasy Football Analytics
For more data-driven fantasy football analysis and projections, follow us on social media:

Visit FantasyFootballAnalytics.net for our latest tools and articles.