When managers trust the kids: U21 substitutions by game state

Managers talk about giving youth a chance as a club philosophy, but in-game substitution choices treat age as a function of the scoreboard and the part of the pitch where mistakes are fatal.

Measuring raw under-21 substitute minutes by scoreline confounds manager preference with bench composition: weaker or injury-hit squads carry more teenagers, and teams that spend more time winning or losing have different benches to begin with. Framing each substitution as a discrete choice over the outfield players actually sitting unused on the bench at that second isolates the decision itself. If two of the six unused outfield players on the bench are under 21, an age-blind pick chooses an under-21 player a third of the time. Comparing the observed picks with that equal-share expectation (1.00 = picked at the same rate as everyone else on the bench) across 25,100 tactical outfield substitutions in the Premier League from 2015/16 to 2026/27 separates where young players are trusted from where they are merely making up the numbers.

25,100 tactical outfield substitutions (15,394 with at least one U21 on the bench; 3,087 U21 sub-ons). Ratios divide observed U21 picks by uniform bench-share expectation (1.00 = no preference); the final column adjusts for each player’s overall pick rate when eligible.

Key findings

1. The one-goal lead is the trough

Under-21 players are picked below an equal share of the bench in every score state because squad pecking order correlates with age and prior minutes. Even so, the curve across game states is a pronounced U-shape. A one-goal lead is the most risk-averse state in football: the U21 pick ratio drops to 0.53 (95% match-bootstrap CI 0.50 to 0.57), or 0.61 relative to each player’s own baseline rate. Once a lead reaches three goals or more, the ratio climbs to 0.83 (0.76 to 0.90), and 0.96 against player baseline - effectively parity with senior teammates. Permuting score states across substitutions within each season places the null 95% band between 0.59 and 0.73 in every state; winning by one sits well below it, while winning by three or more and losing by two or more (0.74) sit above it. Restricting to changes from the 60th minute onwards leaves the numbers unchanged (0.52 at one up, 0.85 at three up).

2. Chasing a game opens the door for young attackers

Splitting by position group shows where the losing-state bump comes from. Under-21 attacking midfielders and forwards are picked above an equal share of the bench when their side is losing by one (1.16, CI 1.08 to 1.24), losing by two or more (1.15, CI 1.05 to 1.26), or winning by three or more (1.18, CI 1.05 to 1.35). When the same young attackers are on the bench with a one-goal lead to protect, their ratio falls to 0.67 (0.59 to 0.75). Against their own individual baseline pick rates, young attackers sit at 1.00 to 1.04 when losing and 0.61 when one goal ahead: when a team needs a goal, a young forward is treated like any other attacker on the bench, whereas protecting a lead removes them from consideration.

3. Young defenders are held back regardless of the score

A natural intuition is that managers specifically avoid bringing on young defenders or defensive midfielders when nursing a narrow lead. The data says the reluctance is positional, not situational. Under-21 defenders and defensive midfielders are picked at 0.27 to 0.31 of expectation when losing or level, 0.39 when winning by one, and 0.48 to 0.50 when winning by two or more. In a conditional logit over the eligible bench, the three-way interaction between U21 status, defensive position, and a one-goal lead is indistinguishable from zero (+0.02 log-odds, SE 0.11). Young defenders are the last option off the bench in every undecided state; a lead is the only thing that nudges their rate upwards.

4. An older like-for-like alternative cuts the rate by nearly a third

When an unused outfield player aged 23 or older in the same position group is also sitting on the bench, the U21 pick ratio is 0.53 (0.68 against player baseline), compared with 0.77 (0.83 against player baseline) when no older same-position option is present. That gap holds within losing states (0.55 vs 0.79 when losing by one) and within three-goal leads (0.66 vs 1.01, where having no senior positional alternative brings U21 picks to exact parity). The penalty remains statistically clear in the conditional logit after controlling for continuous age, log prior Premier League minutes, and like-for-like positional replacement (-0.28 log-odds, SE 0.04), with the caveat that a manager who names a teenager as the sole positional cover on the bench has already made part of the decision before kick-off.

5. U23s follow the same shape, and bigger benches diluted the U21 rate

Moving the threshold to under-23 (present in 20,723 of the 25,100 choice sets) produces the same curve with less steepness: 0.90 when losing by two or more, 0.85 when losing by one, 0.80 when level, a trough of 0.71 when winning by one, 0.81 when winning by two, and 0.92 when winning by three or more. U23 attackers are picked at 1.33 to 1.41 of uniform expectation when trailing vs 0.89 when one goal up.

Across seasons, the one-goal-lead trough appears in all eleven completed campaigns for both U21 and U23 players. The overall U21 pick ratio, however, stepped down after 2020: between 2015/16 and 2019/20 it sat between 0.69 and 0.83 (0.54 to 0.76 when winning by one), whereas from 2020/21 to 2025/26 it ranged between 0.55 and 0.64 (0.40 to 0.53 when winning by one). The five-substitute rule introduced permanently in 2022/23 raised the U21 share of total substitutions from roughly 11% to 14%, but the pick ratio relative to bench availability fell (0.64 in 2022/23, 0.59 in 2025/26). Nine-man benches added academy slots at the end of the teamsheet faster than they added trust on the pitch.

Methodology and sources

• Source: Opta Premier League match and squad records covering 2015/16 to 2026/27 (through 12 October 2026; league fixtures only, cup ties excluded).
• Choice sets: Every substitution coded as tactical where both the player leaving and the player entering are outfield players (n = 25,100 substitutions; 146,374 bench-option rows). Injury substitutions and goalkeeper changes are excluded. At each substitution, the choice set consists of all outfield substitutes on that team’s bench who have not yet entered the match. Substitutions with no U21 on the bench contribute zero to both observed and expected counts and do not affect the ratio.
• Game state and positions: Scoreline and red-card state are reconstructed at the exact second of the substitution (with own goals credited to the beneficiary team). Each player’s position group (DEF, DM, CM, AM, FW) is assigned from their modal starting position in that Premier League season, falling back to their listed substitute position where they made no starts. Age is exact age in years on the match date.
• Expectation and controls: Uniform expectation assigns each eligible bench player 1 / n_alt. Player-baseline expectation weights each eligible option by that player’s overall pick rate across the dataset (shrunk towards 1.0 with a one-pick prior) normalised within the choice set. A placebo test flagging players aged 24 to 25 produces a flat ratio across score states (1.18 losing by one, 1.14 level, 1.15 winning by one, 1.06 winning by three or more), confirming that the state curve is specific to youth rather than a mechanical artefact of lower-rotation squad members.
• Conditional logit: Estimated across all 25,100 choice sets with match-clustered bootstrap standard errors (60 replications), controlling for continuous age (-0.27 per 5 years, SE 0.01), log prior Premier League minutes since 2015/16 (+0.89, SE 0.03), and same-position-group replacement as the player leaving (+1.38, SE 0.02). The main U21 terms are -0.42 (SE 0.05) at baseline, +0.53 (SE 0.10) interacted with winning by 3+, -0.25 (SE 0.07) interacted with winning by 1, -0.33 (SE 0.07) interacted with losing for non-attackers, +1.11 (SE 0.07) interacted with losing for attackers, and -0.28 (SE 0.04) when an eligible 23+ player in the same position group is also on the bench.

SubstitutionsYouth developmentGame statePremier League

← All footnotes and notes