The Evidence on Teaching

Measuring the Impacts of Teachers II: Teacher Value-Added and Student Outcomes in Adulthood

Chetty R, Friedman JN, Rockoff JE · 2014

grade Bnatural-experimentindependentreplicated
Sample
~1,000,000 students linked to tax records into their late twenties
Population
US grades 3-8 in one large urban district, followed into adulthood via tax data.
Design
Links the value-added measures validated in Part I to adult outcomes. The long-run estimates rest on selection-on-observables rather than on the switching design, and are control-sensitive — treat the magnitudes as approximate.
Key findings
One year with a better teacher moves college attendance and adult earnings, even though the test-score effect itself decays ~50% per year. For a small school the practical reading is the noise: single-year VA has reliability of only r ~ 0.3-0.5, so use 2-3 years plus observation plus behavioural indicators.
Genetic confound
The adult-outcome estimates are selection-on-observables, so heritable family traits correlated with teacher assignment are not fully differenced out. The fadeout-with-persistence pattern is the interesting part and does not depend on the exact magnitude.
Replication notes
Jackson 2018 shows teacher effects on BEHAVIOURS predict long-run outcomes as strongly as test-score VA, which is the natural mechanism for fadeout-with-persistence.

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
One year with a teacher 1 SD betteroutcomes+0.82pp college attendance at 20 (base 37%); +1.3% earnings at 28standardizedlong-runnoneadulthoodattainment
Replacing a bottom-5% teacherPV/classroom$185k using noisy 3-year VA; $250k career/drift-adjusted (the widely misquoted ~$1.4M is the UNDISCOUNTED lifetime total)standardizedlifetime projectionnoneadulthoodattainment
Fadeout patternpatterntest-score effects decay ~50% per year, yet adult outcomes persist (fadeout-with-reemergence)standardizedlong-runnoneadulthoodattainment

Cited by