College football rosters are drawing from a wider geographic market, especially through transfers and the very top of high-school recruiting. But regional recruiting still matters.
Published
August 29, 2026
Roster geography
2009–25
Elite cohort
2009–24
532 vs 211 mi
Median hometown distance for transfers versus non-transfers across the 25 notable programs, 2021–25.
0.7% → 17.0%
Detected transfer share across the 25 notable programs, 2014 versus 2025.
203 → 571 mi
Median hometown-to-college distance for national Top-3 recruits, 2009–14 versus 2021–24.
Choose a program. The left map shows its 2014 roster and the right shows 2025. Gold marks players without a detected prior college. Blue marks detected transfers.
LSU · 2014 and 2025
2014Before the modern portal era
2025Portal and NIL era roster
Roster profile
Clusters use distance, local dependence, long-distance share, transfers and geographic diversity.
Long-run example / LSU
LSU's local pipeline, 1995–2026
LSU is useful because the long-run file starts in 1995. Louisiana players made up 72.9% of the roster that year. The 2026 roster is 37.5% Louisiana, while 47.3% of players have a previous school. The full series shows how gradual the first shift was and how quickly the transfer share changed later.
Louisiana hometown shareRoster with a previous school
Y-axis: share of LSU roster (%). The two series answer different questions and intentionally use the same 0–100% scale.
Recruiting footprints / 2009–25
Recruiting routes, 2009–25
Each arc connects a player’s hometown to the selected campus. Filter transfers, non-transfers or positions, then play the seasons.
2025
TransferNon-transfer
Lower-48 hometowns shown. Campus endpoints use city-level coordinates. Hover an artery for player details.
The transfer market
Transfers come from a much wider geographic market
Across the 25 notable programs in 2021–25, the median transfer's hometown was 532 miles from his current campus. The median non-transfer was 211 miles away. Transfers were also about twice as likely to come from 500+ miles away.
Non-transfers · 2021–25
211 mi
45% in-state · 27% from 500+ miles
VS
Transfers · 2021–25
532 mi
24% in-state · 50% from 500+ miles
2025
Non-transferTransfer
How the transfer gap changed over time
Transfer value minus non-transfer value for the selected program and measure. Hover a point for the exact value.
The very top recruits
Top-3 recruit distance increased
The Top-10 cohort is intentionally narrow: ten recruits per class from the 247Sports Composite, 2009–24. Among the national Top 3, median hometown-to-college distance rose from 203 miles in 2009–14 to 571 miles in 2021–24. Ranks 4–10 do not show the same clean jump.
2.8×
Top-3 median distance · 2021–24 vs 2009–14
The increase is concentrated among the Top 3; ranks 4–10 change less.
Elite mobility by recruiting class
Each class contains only 10 prospects, so annual movement is noisy. The pooled era comparison reduces that year-to-year variation.
Pooled era comparison
exact pooled eras
Talent source. National ranks are 247Sports Composite Top-10 classes. Hometown and college-distance fields are matched to the roster source, so “local” means roster hometown relative to the first observed college destination.
Top-10 signees by school
The rank-weighted total assigns 10 points to the national #1 recruit, 9 to #2, … and 1 to #10, so a school landing the very top of the board gets more credit than one accumulating the bottom of the Top 10.
In-state retention by home state
retention = college in home state
Performance context
Prior-year performance and next-class Top-10 signing rate
Top-10 signing is more strongly associated with prior-year final Elo than with a one-year residual above preseason expectations. In 2010–24, the top quintile of teams by prior-year final Elo signed a Top-10 recruit in 18.6% of team-seasons. The bottom quintile did so in 0.3%. Ranking teams by wins above expectation produces a much weaker gradient.
Prior-year performance and Top-10 signing rate
next recruiting class
Team-seasons are split into quintiles on the selected prior-season performance signal. Outcome = share signing at least one national Top-10 recruit in the following class.
Conference capture of national Top 10 · 2019–24
Regional patterns
Home-state advantages still show up clearly
Click a state to see which programs have the largest roster share of players from that state, then move through time. “State recruiting share” means roster share inside the selected comparison universe. Small states are pooled across multiple seasons when the annual sample is thin.
2025
Scroll horizontally to see every state →
Louisiana
How the selected state's top programs changed
Regional patterns / Cross-state recruiting
Cross-state recruiting from states with a baseline leader
The cross-state recruiting score gives more weight to players from states that historically had a clear major-program leader. “Current reach” shows the current score; “expansion” shows the change since 2014–18.
Selected program
Roster-building strategies
Programs are not converging on one roster-building formula
The clusters use recruiting distance, in-state dependence, long-distance shares, transfer share and geographic diversity. The labels describe roster-building style, not quality. The movement view shows changes in cluster assignment between the two periods.
Coordinates are a two-dimensional PCA projection of standardized roster features. Distance on this chart means similarity of roster-building profile - not literal geography. Arrows show movement from the 2014–18 profile to the 2021–25 profile.
Performance
Broader recruiting reach does not reliably predict overperformance
Teams that expanded their recruiting footprint did not consistently beat preseason expectations more often. The pooled relationship is close to zero, and reversing the timing produces a similarly weak result.
Selected sample
−1correlation+1
Do not read this as causal. The reach index is standardized within each season, and both program quality and recruiting geography have persistent causes we have not fully controlled for. Neither design identifies a causal effect.
Specification comparison
Transfer volume looks different when a program is compared with itself
Pooled models compare programs with one another. Team and season fixed effects compare each program primarily with its own history. The all-FBS estimate is positive; uncertainty is larger in the 25-program sample.
Estimated effect of +10 percentage points of transfer share
95% confidence intervals
Selected specification
1
Raw pooledCompares different programs directly.
2
Season adjustedRemoves sport-wide year effects.
3
Team + season fixed effectsCompares a program mainly with itself across seasons.
Still observational. More transfer usage can arrive with coaching changes, roster holes, NIL resources and other changes the model does not identify separately.
Conclusion
The roster market is broader. Home turf still matters.
Transfers come from substantially farther away than non-transfers, and detected transfer usage increased sharply across the 25-program comparison group. The national Top-3 recruiting cohort also traveled farther in 2021–24 than in 2009–14.
At the same time, state-level recruiting advantages remain visible and programs continue to use different roster-building strategies.
NIL, the transfer portal, conference realignment and other structural changes overlap in time. These analyses describe the geography and timing of the period; they do not separately identify the causal effect of NIL or the portal.
Explore the data
Additional views
Program, conference, expectations, event-study and full-field views.
Compare one or two programs by season, with 2014–18 and 2021–25 averages shown below.
Explore / Conferences
Conference performance and roster geography
For each season, rank teams inside the conference they belonged to that year, then compare the top N with the rest. Conference membership is taken from each season, so realignment does not reassign past seasons.
2025
Top group minus the rest
Positive means the conference leaders averaged more of the selected characteristic than the rest of that conference that season; negative means less.
Performance vs roster strategy
How every conference compares
Conference mean for the selected roster characteristic in the selected season. Click a conference to make it the focus above.
Performance layer: completed cfbfastR schedules joined to roster geography. The performance layer uses a preseason-only expected-wins model: each team’s opening Elo and its opponents’ opening Elo are converted to game probabilities using a logistic calibration fit only on earlier seasons. FBS regular-season games only, so the expectation does not peek at bowl opponents.
Explore / Expectations
Expected and actual FBS regular-season wins
Expected wins use preseason Elo, the FBS regular-season schedule, and home or neutral location. The residual is actual minus expected wins.
2025
Expected wins vs actual wins
Above the diagonal = more FBS regular-season wins than the preseason model expected. The expectation model is calibrated on prior seasons only.
Largest positive and negative residuals
actual − expected
Beat expectation
Missed expectation
What “expected” means. For each season, the model uses only opening Elo, the known FBS regular-season opponents, and home/neutral location. Its calibration is fit on completed games from earlier seasons, never the season being predicted. FCS games and bowls are excluded from both expected and actual win totals in this metric.
Explore / Event studies
Roster and performance around selected events
Teams are aligned by season around either an unexpected winning season or a threshold increase in national roster reach.
Recruiting geography around unexpected seasons
Year 0 is the event season. Shaded bars are ±1 standard error around the mean. Recruiting outcomes are shown relative to each program’s year −1 baseline; performance outcomes are shown on their natural scale.
Example events in this sample
Descriptive, not causal. Events are separated by a four-year cooldown so adjacent qualifying seasons are not counted as separate events. Coaching, NIL resources, schedule changes, and roster quality can move at the same time.
Explore / Full field
All 25 programs by season
Programs are sorted by the change from the 2014–18 average to the 2021–25 average.
Scroll horizontally to see every season →
Methods and sources
Methods and sources
Coverage
Roster geography covers 2009–25 across FBS programs, with a 25-program comparison universe used for major-brand views. The LSU long-run case study covers 1995–2026.
Transfer detection comes from prior-school roster history and should be read as a conservative indicator rather than a complete portal census.
Recruiting and performance
Elite recruiting uses the national Top 10 from the 247Sports Composite for 2009–24. Hometown and college-distance fields are matched to the roster source, so “local” means roster hometown relative to the first observed college destination.
Performance uses completed cfbfastR schedules. Preseason Elo and known FBS schedules are converted to expected wins with a calibration fit only on earlier seasons. FCS games and bowls are excluded from both expected and actual win totals.
Interpretation
NIL and the transfer portal arrived together with conference realignment, coaching changes, program resources and other structural changes. These analyses document the geography and timing of the era; they do not separately identify NIL’s or the portal’s causal effect.
Acknowledgement
Thanks to Andrew Dunn for helping pull the data for this analysis. I also recommend checking out more of his work at dunn.us.