Students randomly assigned to higher-quality kindergarten classrooms in Tennessee's Project STAR earn more as adults even though their test-score advantages fade by grade 4 — the fade-out and re-emergence puzzle documented by Chetty, Friedman, Hilger, Saez, Schanzenbach, and Yagan (2011). The measure of class quality they use, a leave-out mean of classmates' end-of-kindergarten test scores, is an omnibus index that bundles teacher practices, peer interactions, and transitory class-level shocks, so the finding establishes that early classrooms matter without identifying what within them is worth manipulating. Fifteen years of follow-up work has established that teachers, peers, and noncognitive-skill formation can each affect adult outcomes in other settings, but no study has decomposed kindergarten class-quality variance into persistent teacher quality, peer composition, and transitory shock components with long-run earnings outcomes, causally tested the noncognitive-skills mediation of re-emergence in early classrooms, or settled whether teacher experience per se (rather than selection of skilled people into experience) generates adult gains. The problem is to identify the manipulable causal channels from early-childhood classroom quality to adult outcomes and the mechanism by which test-score effects fade while earnings effects re-emerge.

Background

Project STAR (Student/Teacher Achievement Ratio, Tennessee, 1985–1989) randomly assigned roughly 11,600 children, upon kindergarten entry, to small or regular-size classes and, within schools, to classrooms; teachers were also randomly assigned to classrooms. Because assignment to classrooms was random within schools, differences in outcomes across kindergarten classrooms can be given a causal interpretation as the combined effect of everything that differed across classrooms. Chetty, Friedman, Hilger, Saez, Schanzenbach, and Yagan (2011) linked STAR participants to U.S. federal tax records and showed that this combined effect is large in adulthood: students assigned to a kindergarten classroom one within-school standard deviation higher in "class quality" earned about 9.6% more at age 27, were more likely to attend college, saved more, and lived in better neighborhoods, even though the same students' test-score advantages had largely faded by grade 4. This coexistence of rapid test-score fade-out with re-emerging adult impacts is known as the fade-out and re-emergence puzzle, and it motivated a broad shift toward judging classroom interventions by adult outcomes rather than by later-grade test scores.

The paper's measure of class quality is deliberately omnibus. Because each STAR teacher is observed in the data only once, teacher quality cannot be separated from peers or luck; the authors instead proxy class quality by a leave-out mean (jackknife) of classmates' end-of-kindergarten test scores — the average score of a student's classmates excluding the student's own score, relative to schoolmates' average. This measure captures teacher practices, classroom management, peer interactions, and any idiosyncratic class-level shock (such as a disruptive event during the year) that systematically affected the class. The paper's own cost-benefit calculations therefore state that the implied value of improving classroom quality is "useful primarily for understanding the stakes" because "one cannot implement interventions that directly improve classroom quality." Noncognitive skills — attributes such as effort, initiative, and self-discipline measured by teacher ratings for a subset of STAR students in grades 4 and 8 — are elevated as one plausible mediator of re-emergence: kindergarten class quality raises these measures while leaving later test scores unaffected, and the measures predict earnings conditional on test scores. But the authors emphasize that this evidence is suggestive, available only for 28–30% of the matched sample, subject to attrition concerns, and not a demonstration that manipulating noncognitive skills causes better adult outcomes.

Three identification obstacles frame the open question. First, decomposing class-level variance requires separating persistent teacher quality from transitory classroom shocks and from peer composition; the reflection problem (Manski 1993) — the difficulty of distinguishing genuine peer influence from the fact that a group's average outcome mechanically contains each member's own characteristics — further contaminates peer estimates in linear-in-means regressions. Second, the teacher experience association is confounded: teachers were randomized to classrooms, but experience was not randomly assigned to teachers, so experience correlates with unobserved attributes such as attachment to the profession. Third, the teacher value-added (VA) literature that followed — VA being a teacher-specific estimate of the teacher's average contribution to student test-score growth, identified from repeated observations of the same teacher across cohorts — established that teachers have causal effects on adult earnings in grades 3–8 settings (Chetty, Friedman, and Rockoff 2014), but this does not decompose the kindergarten bundle, and STAR's design (each teacher seen once) cannot support VA estimation. Adjacent literatures on class size, peers, and fade-out mechanisms supply partial evidence but leave the decomposition and the mediation question for kindergarten classrooms open.

Problem Statement

Identify which manipulable components of kindergarten classroom quality causally generate the long-run earnings gains documented in Project STAR, and the mechanism by which the test-score effects of those components fade by the middle grades while adult outcomes improve. Following the source paper's own research agenda, the open problem has two parts: (1) the decomposition of the omnibus class-quality effect — the leave-out mean of classmates' end-of-kindergarten test scores used by the source paper, an index that aggregates teacher, peer, and classroom-shock influences — into persistent teacher quality, peer composition, and transitory class-level shocks, quantifying each component's causal contribution to adult earnings (and related adult outcomes such as college attendance), since policy cannot act on "classroom quality" as a whole; and (2) whether noncognitive skills (persistent behavioral attributes such as effort, initiative, and self-discipline, as distinct from test-measured cognitive skills) are the mediating channel of the fade-out and re-emergence pattern — established causally, not merely predicted cross-sectionally, by evidence that targeted manipulation of noncognitive skills in early classrooms raises long-run outcomes, or evidence that another channel is responsible. The source paper also explicitly acknowledges a closely related open question: whether increasing a teacher's own experience, or substituting more-experienced teachers, produces student gains comparable to the observed experience-earnings association, given that experience was not randomly assigned to teachers and correlates with unobserved attributes such as attachment to the profession. These questions serve one objective — converting the demonstration that kindergarten classrooms have large long-run causal effects into knowledge of which specific, actionable inputs produce those effects and why they re-emerge in adulthood. The two parts are jointly posed: (1) the decomposition of the omnibus class-quality effect into persistent teacher quality, peer composition, and transitory class-level shocks — quantifying each of these three components' causal contribution to adult earnings within the kindergarten bundle to the extent the evidence identifies it — and (2) the mediation question — establishing causally whether noncognitive skills are the channel of the fade-out and re-emergence, or that another channel is responsible. The teacher-experience question (3) is a closely related open question on the same agenda, one that does not substitute for resolving (1) and (2). Establishing the causal channels requires decomposing the kindergarten class-quality bundle itself; evidence from non-kindergarten settings alone does not identify the kindergarten channels. Part (1) concerns the causal shares of the kindergarten class-quality effect on adult earnings that the available evidence identifies and, for any component it cannot separately identify, the bounds or partial shares that evidence supports, together with a rigorous identification and decomposition strategy — its assumptions and whatever designs or datasets would identify the remaining shares. The decomposition is not identified by STAR's original design, which observes each teacher only once.

References

  1. Chetty, Raj, John N. Friedman, Nathaniel Hilger, Emmanuel Saez, Diane Whitmore Schanzenbach, and Danny Yagan (2011). "How Does Your Kindergarten Classroom Affect Your Earnings? Evidence from Project Star." Quarterly Journal of Economics 126(4):1593–1660. DOI: 10.1093/qje/qjr041. https://doi.org/10.1093/qje/qjr041
  2. Chetty, Raj, John N. Friedman, and Jonah E. Rockoff (2014). "Measuring the Impacts of Teachers II: Teacher Value-Added and Student Outcomes in Adulthood." American Economic Review 104(9):2633–2679. DOI: 10.1257/aer.104.9.2633. https://doi.org/10.1257/aer.104.9.2633
  3. Rose, Evan K., Jonathan T. Schellenberg, and Yotam Shem-Tov (2022). "The Effects of Teacher Quality on Adult Criminal Justice Contact." NBER Working Paper 30274. DOI: 10.3386/w30274. https://www.nber.org/papers/w30274
  4. Carrell, Scott E., Mark Hoekstra, and Elira Kuka (2018). "The Long-Run Effects of Disruptive Peers." American Economic Review 108(11):3377–3415. DOI: 10.1257/aer.20160763. https://doi.org/10.1257/aer.20160763
  5. Bietenbeck, Jan (2019). "The Long-Term Impacts of Low-Achieving Childhood Peers: Evidence from Project STAR." Journal of the European Economic Association 18(1):392–426. DOI: 10.1093/jeea/jvy049. https://doi.org/10.1093/jeea/jvy049
  6. Jackson, C. Kirabo (2018). "What Do Test Scores Miss? The Importance of Teacher Effects on Non–Test Score Outcomes." Journal of Political Economy 126(5):2072–2107. DOI: 10.1086/699018. https://doi.org/10.1086/699018
  7. Dynarski, Susan, Joshua Hyman, and Diane Whitmore Schanzenbach (2013). "Experimental Evidence on the Effect of Childhood Investments on Postsecondary Attainment and Degree Completion." Journal of Policy Analysis and Management 32(4):692–717. DOI: 10.1002/pam.21715. https://doi.org/10.1002/pam.21715
  8. Cascio, Elizabeth U., and Douglas O. Staiger (2012). "Knowledge, Tests, and Fadeout in Educational Interventions." NBER Working Paper 18038. DOI: 10.3386/w18038. https://www.nber.org/papers/w18038
  9. Bailey, Drew, Greg J. Duncan, Candice L. Odgers, and Winnie Yu (2017). "Persistence and Fadeout in the Impacts of Child and Adolescent Interventions." Journal of Research on Educational Effectiveness 10(1):7–39. DOI: 10.1080/19345747.2016.1232459. https://doi.org/10.1080/19345747.2016.1232459
  10. Kaestner, Robert, and Cuiping Schiman (2024). "Do preschool investments depreciate and fadeout or result in dynamic complementarities? An assessment using the Head Start Impact Study." Review of Economics of the Household 24(1):61–97. DOI: 10.1007/s11150-024-09735-4. https://doi.org/10.1007/s11150-024-09735-4
  11. List, John A., and Haruka Uchida (2024, revised July 2026). "Here Today, Gone Tomorrow? Toward an Understanding of Fade-out in Early Childhood Education Programs." NBER Working Paper 33027. DOI: 10.3386/w33027. https://www.nber.org/papers/w33027
  12. Ladd, Helen F., and Lucy C. Sorensen (2017). "Returns to Teacher Experience: Student Achievement and Motivation in Middle School." Education Finance and Policy 12(2):241–279. DOI: 10.1162/edfp_a_00194. https://doi.org/10.1162/edfp_a_00194
  13. Fredriksson, Peter, Björn Öckert, and Hessel Oosterbeek (2013). "Long-Term Effects of Class Size." Quarterly Journal of Economics 128(1):249–285. DOI: 10.1093/qje/qjs048. https://doi.org/10.1093/qje/qjs048
  14. Sojourner, Aaron (2013). "Identification of Peer Effects with Missing Peer Data: Evidence from Project STAR." Economic Journal 123(569):574–605. DOI: 10.1111/j.1468-0297.2012.02559.x. https://doi.org/10.1111/j.1468-0297.2012.02559.x
  15. Manski, Charles F. (1993). "Identification of Endogenous Social Effects: The Reflection Problem." Review of Economic Studies 60(3):531–542. DOI: 10.2307/2298123. https://doi.org/10.2307/2298123
  16. Belfield, Christopher, and Imran Rasul (2020). "Cognitive and Non-Cognitive Impacts of High-Ability Peers in Early Years." Fiscal Studies 41(1):65–100. DOI: 10.1111/1475-5890.12216. https://doi.org/10.1111/1475-5890.12216
  17. Wan, Sirui, Timothy N. Bond, Kevin Lang, Douglas H. Clements, Julie Sarama, and Drew H. Bailey (2021). "Is intervention fadeout a scaling artefact?" Economics of Education Review 82:102090. DOI: 10.1016/j.econedurev.2021.102090. https://doi.org/10.1016/j.econedurev.2021.102090