Teacher quality — selection over credentials and workshops

Verdict

Within the school walls, nothing you control matters more than who teaches — and nothing you can read on a résumé tells you who that is.

• Magnitude: 1 SD of teacher effectiveness moves same-year achievement ~0.10–0.15 SD (Texas lower bound; CFR ~0.14 math/0.08 ELA) — larger than a 10-student class-size reduction, annually, at no marginal cost once selected.
• Measurement validity: teacher value-added is the only school-quality measure with experimental-adjacent validation — prior-score-controlled VA forecasts causal impacts roughly 1:1 (CFR's teacher-switching test, corroborated by Kane & Staiger's random assignment of teachers to classrooms). Rothstein's critique bounds possible bias at 10–35% of variance (and the counter-replies argue ~0); the unresolved band is "modest, not reversal."
• Long-run stakes: one year with a 1 SD better teacher → +0.82pp college, +1.3% age-28 earnings; replacing a bottom-5% teacher ≈ $185k PV per classroom using realistic 3-year VA ($250k career-adjusted). Test effects fade ~50%/yr while adult outcomes persist — the same fadeout-with-reemergence pattern as class size.
• Credentials are noise: certification ~null, master's degrees ~null (stop paying for them), experience gains concentrated in the first ~5 years. The top-vs-bottom quartile of observed years-1-2 performance predicts a 10-percentile student gap — twice the class-size effect (selection strategy).
• Development: the canonical pro-PD number (+21 percentiles) rests on 9 small pre-2003 developer-run studies out of 1,300+ screened and did not survive scale, while the spending audit finds ~$18k/teacher/yr and ~19 days/yr with no type, amount or combination distinguishing improvers (descriptive, grade D — it prices the spend, it does not prove development cannot work). The causal exceptions: content- specific instructional coaching (+0.10 SD achievement even in ≥100-teacher trials, trim-and-fill 0.14 — reading-dominated, pooled math 0.04 ns) and evaluation-with-feedback (Cincinnati: +0.11 SD math in post-evaluation years, largest for weaker teachers, ~$7.5k/teacher).

Hereditarian-lens assessment

Risk: low, with the confound named precisely: dynamic sorting of (partly heritable) student ability to teachers is exactly what the VA validation designs bound. Long-run earnings estimates are control-sensitive extrapolations — treat magnitudes as approximate, directions as solid. Note teacher effects are on domain skills, behaviors, and attainment; nothing here moves g, and teacher effects on behaviors predict long-run outcomes as strongly as test VA (Jackson) — measure both.

Practical guidance

• Selection over development: open the hiring funnel (certification-blind), try out more teachers than you keep, and make year-2/3 retention the real tenure decision using 2–3 years of multi-measure evidence (single-year VA is noisy, r≈0.3–0.5).
• Kill the master's-degree pay bump; redirect to performance retention bonuses.
• Replace workshops with: content-specific coaching (especially literacy) and a serious observation-evaluation cycle with feedback — the two development levers with causal support.
• Expect the binding constraint at scale to be talent supply (Houston: 300+ interviews for 19 principals) — which is also why class-size reduction that lowers the hiring bar self-defeats.