High School Recruiting Analytics

Recruiting Trust & Geography Intelligence

Which recruiting star ratings actually predict college outcome, and which states are genuinely under- or over-recruited, tested against five recruiting classes and 20,350 recruits, not assumed.

Open the Dashboard →

Summary

Every recruiting service publishes a star rating, and most evaluation treats a 4-star at one position the same way it treats a 4-star at another. This project tests that assumption directly, two ways: which positions’ ratings actually predict college outcome, and which states produce more (or less) talent than their recruiting attention would suggest, once coverage-depth artifacts are controlled for.

Every finding here had to survive being split across five individual recruiting classes and a bootstrapped confidence interval that genuinely excludes zero. Most candidate findings didn’t survive that bar. The two that did are below.

Findings

Position Trust

QB and DB are the only two positions that stayed in the same reliability tier across all five class years, out of twelve ranked. QB never left the top five, with the tightest correlation band of any position. DB was consistently bottom-tier, every year. DE looked like the single most predictable position when years were pooled together, ranked #1, but swung between #1 and #6 depending on the class, so it’s reported as a finding that didn’t survive, not a headline. CB and DT showed the same instability. OL and LS are reported in a separate, lower-confidence tier, since the underlying outcome data for those positions is measurably weaker than for every other group.

Geographic Inefficiency

Maryland is the one state that’s both under-recruited and rank-stable across all five classes, the clearest geographic signal in the dataset. California, Mississippi, and Washington show an over-recruited signal when pooled, but the exact ranking moves too much year to year to call settled. Every other state with sufficient sample, including Texas, Georgia, and Florida, comes back neutral once a confidence interval is required to exclude zero. That neutral result for most of the map is a real finding, not a gap. Three tested hypotheses (proximity to power programs, in-state program strength, roster crowding) failed to explain why Maryland specifically is under-recruited — reported as an open question, not papered over.

Method & Data Quality

Outcome is measured as a Spearman rank correlation between star rating and an outcome-proxy score, computed per position (or per state, using 3-star-and-above recruits as the headline population) with a bootstrapped confidence interval. Every result is stress-tested by re-running it on each individual class year separately, a finding only counts as stable if it holds across all five. Every category label is asserted programmatically from that data, not hand-picked, so a future data refresh can’t silently ship a stale narrative.

One data-quality catch worth noting directly: Washington’s raw recruit count initially looked like a real signal, more total recruits than Ohio, Alabama, or Georgia. It wasn’t. 64% of that population was 2-star recruits, and the state’s blue-chip rate sat at 3.8% against a 12%+ national average, a scouting-depth artifact, not a talent signal. That’s the specific class of error this project is built to catch before it becomes a headline.

Data: CollegeFootballData.com  ·  2019–2023 Recruiting Classes  ·  20,350 Recruits Tracked