CoChat
Sign Up
Cited by 1,729School Choice and PerformanceLabor market dynamics and wage inequalityTeacher Education and Assessments

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

Open access
Chetty, Raj, Friedman, John N., Rockoff, Jonah E.·2014·American Economic Review
DOI 10.1257/aer.104.9.2633
Open this paper in CoChat — free accountOpen PDF

Abstract

Are teachers' impacts on students' test scores (value-added) a good measure of their quality? This question has sparked debate partly because of a lack of evidence on whether high value-added (VA) teachers improve students' long-term outcomes. Using school district and tax records for more than one million children, we find that students assigned to high-VA teachers are more likely to attend college, earn higher salaries, and are less likely to have children as teenagers. Replacing a teacher whose VA is in the bottom 5 percent with an average teacher would increase the present value of students' lifetime income by approximately $250,000 per classroom. (JEL H75, I21, J24, J45)

Integrity signals

Registry

Crossref · 10.1257/aer.104.9.2633

Cites

178 of 178 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

4 of 33 cited sources checked · 3 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Conditioning on prior test scores and other observables is adequate to obtain unbiased estimates of teachers' causal impacts on test scores.

    “Our first method of estimating teachers' long-term impacts builds on our finding that conditioning on prior test scores and other observables is adequate to obtain unbiased estimates of teachers' causal impacts on test scores (Chetty, Friedman, and Rockoff 2014).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Chetty, Friedman, and Rockoff 2014 source
  • Findingown data

    Teachers' impacts on test scores fade out rapidly in subsequent grades.

    “We confirm this rapid fade-out in our data”
    Quote in the paperSubstantial claim
  • Findingown data

    Students assigned to high-VA teachers are more likely to attend college.

    “we find that students assigned to high-VA teachers are more likely to attend college, earn higher salaries, and are less likely to have children as teenagers.”
    Quote in the paperSubstantial claim
  • Findingown data

    Students assigned to high-VA teachers earn higher salaries.

    “we find that students assigned to high-VA teachers are more likely to attend college, earn higher salaries, and are less likely to have children as teenagers.”
    Quote in the paperSubstantial claim
  • Findingown data

    At age 28, a one standard deviation increase in teacher quality in a single grade raises annual earnings by 1.3 percent.

    “At age 28, the oldest age at which we currently have a sufficiently large sample size to estimate earnings impacts, a one standard deviation increase in teacher quality in a single grade raises annual earnings by 1.3 percent.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 31
  • May combine two claims. 49
  • The quote only partly states this claim. 3

  • Classified as probably not a substantial claim.

    Premisecitation

    VA measures which control for lagged test scores exhibit little or no bias.

    “We addressed the first question in the previous paper in this volume (Chetty, Friedman, and Rockoff 2014) and concluded that VA measures which control for lagged test scores exhibit little or no bias.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Chetty, Friedman, and Rockoff 2014 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Using value-added measures to evaluate teachers could induce responses such as teaching to the test or cheating, eroding the signal in value-added measures.

    “First, using VA measures to evaluate teachers could induce responses such as teaching to the test or cheating, eroding the signal in VA measures (e.g., Jacob 2005, Neal and Schanzenbach 2010).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Jacob 2005 source

      The Introduction reports that Chicago’s high-stakes testing produced gains largely in test-specific skills rather than comparable gains on a state exam; it also discusses cheating as a possible response, while noting that estimated cheating explained very little of the score gains. The paper supports the concern that testing incentives can distort scores, but it does not specifically examine using value-added measures to evaluate teachers or the signal in those measures. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites Neal and Schanzenbach 2010 source

  • Classified as probably not a substantial claim.

    Premisecitation

    Existing test-score value-added measures are a good proxy for a teacher's ability to raise students' test scores.

    “Our first paper (Chetty, Friedman, and Rockoff 2014) showed that existing testscore value-added measures are a good proxy for a teacher's ability to raise students' test scores.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Chetty, Friedman, and Rockoff 2014 source

  • May combine two claims.

    Findingown data

    Using a quasi-experimental design based on teacher turnover, a one standard deviation improvement in teacher VA raises the probability of college attendance at age 20 by 0.86 percentage points, nearly identical to the estimate from the first research design.

    “Using this design, we find that a one standard deviation improvement in teacher VA raises the probability of college attendance at age 20 by 0.86 percentage points, nearly identical to the estimate obtained from our first research design.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: High-frequency teacher turnover within school-grade cells is uncorrelated with student and school characteristics.

  • May combine two claims.

    Findingown data

    The qualitative conclusion that teacher VA has long-term impacts is not sensitive to trimming, but including teachers in the top 1 percent reduces our estimates of teachers' impacts on long-term outcomes by 10-30 percent.

    “The qualitative conclusion that teacher VA has long-term impacts is not sensitive to trimming, but including teachers in the top 1 percent reduces our estimates of teachers' impacts on long-term outcomes by 10-30 percent.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The mean college quality conditional on attending college is $41,756, while the quality for all those who do not attend college is $17,920.

    “The mean college quality conditional on attending college is $41,756, while the quality for all those who do not attend college is $17,920.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Both parent characteristics and twice-lagged test scores are strong predictors of college attendance rates even conditional on the baseline controls X ct, with F-statistics exceeding 300.

    “Both parent characteristics and twice-lagged test scores are strong predictors of college attendance rates even conditional on the baseline controls X ct , with F-statistics exceeding 300.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Being assigned to a higher value-added teacher has a significant impact on earnings, with the null hypothesis of κ = 0 rejected with p < 0.01.

    “Being assigned to a higher value-added teacher has a significant impact on earnings, with the null hypothesis of κ = 0 rejected with p < 0.01.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    At early ages, the impact of higher VA on earnings is negative and statistically significant.

    “At early ages, the impact of higher VA is negative and statistically significant”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingreasoning

    A one standard deviation increase in teacher value-added raises earnings at age 28 by 1.34 percent.

    “similar to the observed impact of 1.34 percent.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The unconditional relationship between scores and outcomes is S-shaped, while the relationship conditional on prior scores and other covariates is almost perfectly linear.

    “The unconditional relationship between scores and outcomes is S-shaped, while the relationship conditional on prior scores and other covariates is almost perfectly linear.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Both math and English test scores are highly positively correlated with earnings, college attendance, and neighborhood quality, and are negatively correlated with teenage births.

    “Both math and English test scores are highly positively correlated with earnings, college attendance, and neighborhood quality, and are negatively correlated with teenage births.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Mean individual earnings at age 28 is $20,885, while mean total income is $21,272 (in 2010 US$).

    “Mean individual earnings at age 28 is $20,885, while mean total income is $21,272 (in 2010 US$).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    A one standard deviation increase in VA raises college quality by $190 for children whose parents have below-median income, compared with $380 for those whose parents have above-median income.

    “a one standard deviation increase in VA raises college quality by $190 for children whose parents have below-median income, compared with $380 for those whose parents have above-median income.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The t-statistic for college attendance in the cross-cohort regression is 3.78.

    “the t-statistic for college attendance is 3.78”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    A one standard deviation increase in current teacher quality typically raises future teacher quality by about 0.05 standard deviations or less in elementary school and by about 0.25 standard deviations in middle school.

    “A one standard deviation increase in current teacher quality typically raises future teacher quality by about 0.05 standard deviations or less in elementary school and by about 0.25 standard deviations in middle school (online Appendix Table 12).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The gains from deselecting teachers based on estimated value-added grow fairly rapidly with more data in the first three years, but the marginal value of additional information is small.

    “The gains grow fairly rapidly with more data in the first three years, but the marginal value of additional information is small.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The percentage impact of a one standard deviation improvement in teacher value-added on earnings observed at age 28 is constant at b = 1.34 percent over the life cycle.
    • The authors state this as an assumption: Earnings gains are discounted at a 3 percent real annual rate back to age 12.

  • May combine two claims.

    Findingown data

    With three years of test score data, the gain from deselecting teachers based on estimated value-added is $266,000, but the gain increases to only $279,000 after ten years.

    “With three years of test score data, the gain is $266,000, but the gain increases to only $279,000 after ten years.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The percentage impact of a one standard deviation improvement in teacher value-added on earnings observed at age 28 is constant at b = 1.34 percent over the life cycle.
    • The authors state this as an assumption: Earnings gains are discounted at a 3 percent real annual rate back to age 12.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The corresponding t-statistic for test scores is 34.0.

    “The corresponding t-statistic for test scores is 34.0”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Contemporaneous changes in teacher quality have significant effects on college attendance rates, but past or future changes have no impact.

    “contemporaneous changes in teacher quality have significant effects on college attendance rates, but past or future changes have no impact”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The undiscounted cumulative lifetime earnings gains from deselection are 5.5 times larger than the present value gains, at $80,000 per student and $2.25 million per classroom.

    “The undiscounted cumulative lifetime earnings gains from deselection are 5.5 times larger than these present value gains ($80,000 per student and $2.25 million per classroom), as shown in online Appendix Table 13.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    After three years, waiting for one more year would increase the gain by $4,000 but has an expected cost of $266,000.

    “After three years, waiting for one more year would increase the gain by $4,000 but has an expected cost of $266,000.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The percentage impact of a one standard deviation improvement in teacher value-added on earnings observed at age 28 is constant at b = 1.34 percent over the life cycle.
    • The authors state this as an assumption: Earnings gains are discounted at a 3 percent real annual rate back to age 12.

  • May combine two claims.

    Findingreasoning

    With one year of data, the expected gain per class from deselecting teachers based on estimated value-added is $226,000, which is 56 percent of the gain from selecting on true value-added.

    “With one year of data, the expected gain per class is $226,000, 56 percent of the gain from selecting on true VA.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The percentage impact of a one standard deviation improvement in teacher value-added on earnings observed at age 28 is constant at b = 1.34 percent over the life cycle.
    • The authors state this as an assumption: Earnings gains are discounted at a 3 percent real annual rate back to age 12.

  • May combine two claims.

    Findingown data

    The impacts of teacher VA on college quality are larger for higher-income and non-minority households in absolute terms.

    “the impacts are larger for higherincome and non-minority households in absolute terms.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Limiting the sample to cells with no imputed VA increases the point estimate of a one standard deviation increase in teacher quality but widens the confidence interval because the sample size is considerably smaller.

    “the point estimate of a one standard deviation increase in teacher quality increases, but the confidence interval is significantly wider because the sample size is considerably smaller.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The coefficient on contemporaneous changes in teacher quality is positive and highly significant while all other coefficients are near zero and statistically insignificant.

    “κ 0 is positive and highly significant while all the other coefficients are near 0 and statistically insignificant.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The point estimates of the impacts of teacher VA are larger for females than males, although we cannot reject equality of the impacts (p = 0.102).

    “the point estimates of the impacts of teacher VA are larger for females than males, although we cannot reject equality of the impacts ( p = 0.102).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning

    The deselection policy would generate net present value gains of $184,000 per teacher deselected, or $9,250 for all teachers on average.

    “Based on our calculations above, the deselection policy would generate NPV gains of $184,000 per teacher deselected, or $9,250 for all teachers on average (because only 1 out of 20 teachers would actually be deselected).”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The percentage impact of a one standard deviation improvement in teacher value-added on earnings observed at age 28 is constant at b = 1.34 percent over the life cycle.
    • The authors state this as an assumption: Earnings gains are discounted at a 3 percent real annual rate back to age 12.

  • May combine two claims.

    Findingreasoning

    The estimated total undiscounted earnings gains from this policy are approximately $50,000 per child and $1.4 million for the average classroom.

    “The estimated total undiscounted earnings gains from this policy are approximately $50,000 per child and $1.4 million for the average classroom.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: A one standard deviation improvement in teacher VA raises earnings by 1.34 percent in all subjects and grade levels.
    • The authors state this as an assumption: Teachers teach one class per year.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given

    Measuring parent income at other points in time yields very similar results.

    “Measuring parent income at other points in time yields very similar results (not reported).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingcitation

    The attrition rate is unrelated to teacher value-added, consistent with the findings of Boyd et al. (2008).

    “The attrition rate is unrelated to teacher VA, consistent with the findings of Boyd et al. (2008).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Boyd et al. (2008)

  • May combine two claims.

    Findingreasoning

    Drift-unadjusted measures yield rankings of teacher quality that are very highly correlated with the paper's measures and thus produce similar gains.

    “As a result, drift-unadjusted measures yield rankings of teacher quality that are very highly correlated with our measures and thus produce similar gains.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Study design and conductnone given

    The within-grade variation in achievement in the district examined is comparable to the within-grade variation nationwide, so the results can be compared to estimates from other samples.

    “The within-grade variation in achievement in the district we examine is comparable to the within-grade variation nationwide, so our results can be compared to estimates from other samples.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Some teachers may be effective at raising students' earnings even if they are not effective at raising test scores, for instance by directly instilling other skills which have long-term payoffs.

    “Intuitively, some teachers may be effective at raising students' earnings even if they are not effective at raising test scores, for instance by directly instilling other skills which have long-term payoffs (Jackson 2013).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Jackson 2013

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    κ_g does not correspond directly to earnings VA, but it reveals the extent to which the test-score-based VA measures currently used by school districts are informative about teachers' long-term impacts.

    “Although κ g does not correspond directly to earnings VA, it reveals the extent to which the test-score-based VA measures currently used by school districts are informative about teachers' long-term impacts.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The gains from policies which pay bonuses to high-VA teachers to increase retention rates are only slightly larger than their costs because most bonus payments end up going to high-VA teachers who would have stayed even without the additional payment.

    “The gains from such policies are only slightly larger than their costs because most bonus payments end up going to high-VA teachers who would have stayed even without the additional payment.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The reduced-form impact κ_g combines the direct impact of having a higher-VA teacher in grade g and the indirect impact of endogenous changes in other educational inputs.

    “First, the reduced-form impact combines two effects: the direct impact of having a higher-VA teacher in grade g and the indirect impact of endogenous changes in other educational inputs.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    In the core sample, 15.8 percent of women have teenage births, compared with 14.6 percent in the 2003 ACS.

    “In our core sample, 15.8 percent of women have teenage births, compared with 14.6 percent in the 2003 ACS.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    Rankings of college quality are very stable across cohorts and robust to alternative specifications provided that earnings are measured after age 28.

    “We find that rankings of college quality are very stable across cohorts and are robust to alternative specifications provided that earnings are measured after age 28 (online Appendix Figure 1, online Appendix Table 2).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    We estimate β_Y using within-teacher variation to account for potential sorting of students to teachers based on VA.

    “Importantly, we estimate β Y using within-teacher variation to account for the potential sorting of students to teachers based on VA.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The pattern of fade-out and re-emergence echoes the findings of recent studies of early childhood interventions.

    “This pattern of fade-out and re-emergence echoes the findings of recent studies of early childhood interventions (Deming 2009; Heckman et al. 2010b; Chetty et al. 2011).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Deming 2009 source
    • cites Heckman et al. 2010b

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Chetty et al. 2011

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims. The quote only partly states this claim.

    Interpretationreasoning

    Higher-quality teachers increase not just attendance but also college completion rates.

    “While we cannot directly measure college completion in our data, this finding suggests that higher-quality teachers increase not just attendance but also college completion rates.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    The increase in earnings from an additional year of schooling is around 9 percent.

    “A second benchmark is the increase in earnings from an additional year of schooling, which is around 9 percent (Gunderson and Oreopoulos 2010, Oreopoulos and Petronijevic 2013).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Gunderson and Oreopoulos 2010 source
    • cites Oreopoulos and Petronijevic 2013

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    At most $195 of the $299 impact is due to the extensive margin response, confirming that teachers improve the quality of colleges which students attend.

    “This suggests that at most (41, 756 -17, 920) × 0.82 percent = $195 of the $299 impact is due to the extensive margin response, confirming that teachers improve the quality of colleges which students attend.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Those who are induced to attend college attend a college of average quality.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Existing evidence shows that improvements in education raise contemporaneous scores, then fade out in later scores, only to reemerge in adulthood.

    “This result aligns with existing evidence that improvements in education raise contemporaneous scores, then fade out in later scores, only to reemerge in adulthood (Deming 2009; Heckman et al. 2010b; Chetty et al. 2011).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Deming 2009 source
    • cites Heckman et al. 2010b

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Chetty et al. 2011

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • The quote only partly states this claim.

    Study design and conductown data

    Inclusion of parent characteristics and lagged test score gains has only a modest effect on the point estimates for college quality.

    “As with college attendance, the inclusion of these controls has only a modest effect on the point estimates.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationreasoning

    The continued impacts on college attendance in the mid-20s may be due to increased attendance of graduate or professional schools.

    “However, the impacts remain significant even in the mid-20s, perhaps because of increased attendance of graduate or professional schools.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    This increase is most consistent with an intensive margin effect, as students would be unlikely to jump from not going to college at all to attending a high-quality college.

    “This increase is most consistent with an intensive margin effect, as students would be unlikely to jump from not going to college at all to attending a high-quality college.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    Higher income households, whose children have higher earnings on average, should be willing to pay larger amounts for higher teacher VA.

    “Hence, higher income households, whose children have higher earnings on average, should be willing to pay larger amounts for higher teacher VA.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationcitation

    Comparing the long-term outcomes of students assigned to different teachers conditional on the same control vector will yield unbiased estimates of teachers' long-term impacts.

    “Given this result, one may expect that comparing the long-term outcomes of students assigned to different teachers conditional on the same control vector will yield unbiased estimates of teachers' long-term impacts.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    The more conservative approach of clustering by school increases standard errors by 30 percent (for earnings), but does not affect our hypothesis tests at conventional levels of statistical significance.

    “The more conservative approach of clustering by school increases standard errors by 30 percent (for earnings), but does not affect our hypothesis tests at conventional levels of statistical significance (online Appendix Table 7).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    The mean present value of lifetime earnings at age 12 in the US population is approximately $522,000.

    “the mean present value of lifetime earnings at age 12 in the US population is approximately $522,000.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The percentage gain in earnings remains constant at 1.34 percent over the life cycle.
    • The authors state this as an assumption: Earnings are discounted at a 3 percent real rate (5 percent discount rate with 2 percent wage growth) back to age 12.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    We cluster standard errors at the school by cohort level when estimating (8) at the class level, which adjusts for correlated errors across classrooms and repeat student observations within a school.

    “We instead cluster standard errors at the school by cohort level when estimating (8) at the class level, which adjusts for correlated errors across classrooms and repeat student observations within a school.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premiseown data

    In elementary school, the standard deviation of teacher effects on student test scores is 0.124 in English and 0.163 in math.

    “In Table 2 of our companion paper, we estimated that the standard deviation of teacher effects on student test scores in elementary school is 0.124 in English and 0.163 in math.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Under the assumption that teacher value-added is normally distributed, a teacher in the bottom 5 percent of the true value-added distribution is on average 2.063 standard deviations below the mean teacher quality.

    “Under the assumption that m jt is normally distributed, a teacher in the bottom 5 percent of the true VA distribution is on average 2.063 standard deviations below the mean teacher quality.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Teacher value-added is normally distributed.

  • May combine two claims.

    Study design and conductown data

    We estimate one-year persistence using more than 5.6 million student-year-subject observations and four-year persistence using more than 1.3 million student-year-subject observations.

    “We estimate one-year persistence using more than 5.6 million studentyear-subject observations and four-year persistence using more than 1.3 million student-year-subject observations.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    The first research design relies on a stronger identification assumption but is bolstered by generating results similar to the quasi-experimental design in the full sample.

    “which as noted above relies on a stronger identification assumption but is bolstered by generating results which are similar to the quasi-experimental design in the full sample.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Higher income families should be willing to pay more for high-VA teachers.

    “This result implies that higher income families should be willing to pay more for high-VA teachers.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    The quasi-experimental estimate is much less precise because it exploits less variation.

    “the quasi-experimental estimate is much less precise because it exploits less variation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    We have much less power here because college attendance is only observed for the older half of our sample and because college is a much noisier outcome than end-of-grade test scores.

    “We have much less power here both because the college attendance is only observed for the older half of our sample and because college is a much noisier outcome than end-of-grade test scores.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Even though teachers have much smaller impacts on English test scores than math test scores, the small improvements that good teachers generate in English are associated with substantial long-term impacts.

    “Even though teachers have much smaller impacts on English test scores than math test scores, the small improvements that good teachers generate in English are associated with substantial long-term impacts.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The pattern of no systematic trend in teacher impacts across grades is consistent with the cross-sectional correlations between test scores and adult outcomes, which are also relatively stable across grades.

    “This pattern is consistent with the cross-sectional correlations between test scores and adult outcomes, which are also relatively stable across grades (online Appendix Table 6).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The greater degree of tracking in middle school is consistent with the availability of subject-specific honors classes in grades 7 and 8.

    “The greater degree of tracking in middle school is consistent with the availability of subject-specific honors classes in grades 7 and 8.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Including teacher VA in all grades simultaneously is not feasible because the primary research design requires conditioning on lagged test scores and lagged scores are endogenous to teacher quality in the previous grade.

    “Unfortunately, this is not feasible because our primary research design requires conditioning on lagged test scores and lagged scores are endogenous to teacher quality in the previous grade.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Because value-added estimates are not perfect predictors of true value-added, there is still substantial room to use other measures, such as principal evaluations or student surveys, to complement value-added estimates and improve predictions of teacher quality.

    “Nevertheless, because VA estimates are not perfect predictors of m jt , there is still substantial room to use other measures-such as principal evaluations or student surveys-to complement VA estimates and improve predictions of teacher quality.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Replacing ineffective teachers is more cost-effective than attempting to retain high-value-added teachers because most teachers stay for the following school year and are relatively inelastic to salary increases.

    “Replacing ineffective teachers is more cost-effective than attempting to retain high-VA teachers because most teachers stay for the following school year and are relatively inelastic to salary increases.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Based on these estimates, a one-time bonus payment of $1,800 to high-value-added teachers who return for a fourth year would increase retention rates in the next year by 1.5 percentage points and generate an average benefit of $3,180.

    “Based on these estimates, a one-time bonus payment of $1,800 to high-VA teachers who return for a fourth year would increase retention rates in the next year by 1.5 percentage points and generate an average benefit of $3,180.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The percentage impact of a one standard deviation improvement in teacher value-added on earnings observed at age 28 is constant at b = 1.34 percent over the life cycle.
    • The authors state this as an assumption: Earnings gains are discounted at a 3 percent real annual rate back to age 12.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Parents should place great value on having their child in the classroom of a high value-added teacher.

    “Whether or not VA is ultimately used as a policy tool, our results show that parents should place great value on having their child in the classroom of a high valueadded teacher.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Recent work has shown that early childhood education has significant long-term impacts.

    “Recent work has shown that early childhood education has significant long-term impacts (e.g.,Heckman et al. 2010a, 2010b, Chetty et al. 2011)”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Heckman et al. 2010a

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Heckman et al. 2010b

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Chetty et al. 2011

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationown data

    Estimation error and drift do not heavily erode the gains from deselection of low value-added teachers because very few of the teachers rated in the bottom 5 percent turn out to be high-value-added teachers.

    “The reason that estimation error and drift do not heavily erode the gains from deselection of low VA teachers is that very few of the teachers rated in the bottom 5 percent turn out to be high-VA teachers.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    While accounting for drift is important for evaluating out-of-sample forecasts accurately, it may not be critical for practical policy applications for value-added.

    “Hence, while accounting for drift is important for evaluating out-of-sample forecasts accurately, it may not be critical for practical policy applications for VA.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Classroom observations have significant noise and may capture transitory fluctuations in teacher quality.

    “For instance, classroom observations have significant noise and may capture transitory fluctuations in teacher quality(Kane et al. 2013).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Kane et al. 2013

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    The remaining students are likely to have parents who did not file tax returns in the early years of the sample when they could have claimed their child as a dependent, making it impossible to link the children to their parents.

    “The remaining students are likely to have parents who did not file tax returns in the early years of the sample when they could have claimed their child as a dependent, making it impossible to link the children to their parents.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A policy that fires teachers if their estimated value-added after three years falls below the fifth percentile would require a mean salary increase of 1.4 percent to equilibrate the teacher labor market.

    “Rothstein estimates that a policy which fires teachers if their estimated VA after three years falls below the fifth percentile would require a mean salary increase of 1.4 percent to equilibrate the teacher labor market.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Rothstein (2013)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    Other measures will also be affected by drift and estimation error.

    “Of course, these other measures will also be affected by drift and estimation error.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    Together with the preceding results, Rothstein's findings imply that deselecting low-value-added teachers could be a very cost effective policy if the signal quality of value-added does not fall substantially when used for personnel evaluation.

    “Together with the preceding results, Rothstein's (2013) findings imply that deselecting low-VA teachers could be a very cost effective policy if the signal quality of VA does not fall substantially when used for personnel evaluation.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Rothstein (2013)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

76 of 76

  • The authors state this as an assumption.

    Unobserved determinants of outcomes in adulthood such as student ability are unrelated to teacher VA conditional on observable characteristics.

    “The identification assumption underlying this research design is selection on observables: unobserved determinants of outcomes in adulthood such as student ability must be unrelated to teacher VA conditional on the observable characteristics.”

  • The authors state this as an assumption.

    High-frequency teacher turnover within school-grade cells is uncorrelated with student and school characteristics.

    “Our quasi-experimental results rest on the identification assumption that high-frequency teacher turnover within school-grade cells is uncorrelated with student and school characteristics.”

  • The authors state this as an assumption.

    Test-score value-added estimates are forecast unbiased: the covariance between true and estimated VA divided by the variance of estimated VA equals 1.

    “ASSUMPTION 1 (Forecast Unbiasedness of Test-Score VA): Test-score value-added estimates  μ jt are forecast unbiased: Cov ( μ jt ,  μ jt ) _ Var (  μ jt ) = Cov ( m jt ,  m jt ) _ Var (  m jt ) = 1.”

  • The authors state this as an assumption.

    The impact of teacher VA on earnings remains constant at 1.3 percent over the life cycle.

    “If the impact on earnings remains constant at 1.3 percent over the life cycle, students would gain approximately $39,000 on average in cumulative lifetime income from a one standard deviation improvement in teacher VA in a single grade.”

  • The authors state this as an assumption.

    The signal quality of VA for long-term impacts does not fall substantially when it is used to evaluate teachers.

    “This result suggests that VA could potentially be a useful tool for evaluating teacher performance if the signal quality of VA for long-term impacts does not fall substantially when it is used to evaluate teachers.”

  • The authors state this as an assumption.

    Test-score value-added estimates are orthogonal to unobserved determinants of earnings conditional on X.

    “ASSUMPTION 2 (Selection on Observables): Test-score value-added estimates are orthogonal to unobserved determinants of earnings conditional on X it : (7) Cov (  m jt , η it ′ ) = 0.”

  • The authors state this as an assumption.

    Those who are induced to attend college attend a college of average quality.

    “Second, we derive a lower bound on the intensive margin effect by assuming that those who are induced to attend college attend a college of average quality.”

  • The paper relies on this without support.

    Any selection effects would be manifested in all outcomes (test scores, college attendance, earnings, etc.).

    “as any selection effects would presumably be manifested in all of these outcomes.”

  • The authors state this as an assumption.

    The unobserved determinants of students' long-term outcomes are uncorrelated with teacher quality conditional on observables.

    “The estimates in the previous section rely on the assumption that the unobserved determinants of students' long-term outcomes are uncorrelated with teacher quality conditional on observables.”

  • The authors state this as an assumption.

    The distribution of unobserved determinants of students' long-term outcomes does not change between adjacent cohorts; student quality is stable over time.

    “provided that student quality is stable over time.”

  • The authors state this as an assumption.

    Changes in teacher quality across cohorts within a school-grade are orthogonal to changes in other determinants of student outcomes across cohorts.

    “ASSUMPTION 3 (Teacher Switching as a Quasi-Experiment): Changes in teacher quality across cohorts within a school-grade are orthogonal to changes in other determinants of student outcomes Δ η sgt ′ across cohorts: (10) Cov ( Δ Q sgt , Δ η sgt ′ ) = 0.”

  • The authors state this as an assumption.

    The percentage gain in earnings remains constant at 1.34 percent over the life cycle.

    “Assume that the percentage gain in earnings remains constant at 1.34 percent over the life cycle”

  • The authors state this as an assumption.

    Earnings are discounted at a 3 percent real rate (5 percent discount rate with 2 percent wage growth) back to age 12.

    “and that earnings are discounted at a 3 percent real rate (i.e., a 5 percent discount rate with 2 percent wage growth) back to age 12”

  • The authors state this as an assumption.

    The undiscounted lifetime earnings gain assumes a 2 percent growth rate but 0 percent discount rate.

    “assuming a 2 percent growth rate but 0 percent discount rate”

  • The authors state this as an assumption.

    The marginal entrant into the labor market would take a job paying the mean earnings level in the sample ($21,256).

    “If the marginal entrant into the labor market were to take a job that paid the mean earnings level in the sample ($21,256)”

  • The paper relies on this without support.

    The marginal entrant into the labor market most likely has lower earnings than the mean.

    “Since the marginal entrant most likely has lower earnings than the mean”

  • The authors state this as an assumption.

    All tracking to future teachers occurs exclusively via teachers' test-score value-added.

    “First, it assumes that all tracking to future teachers occurs exclusively via teachers' test-score VA.”

  • The authors state this as an assumption.

    Net impacts do not net out potential changes in other factors besides teachers, such as peer quality or parental inputs.

    “Second, ˜ κ g does not net out potential changes in other factors besides teachers, such as peer quality or parental inputs.”

  • The authors state this as an assumption.

    Teacher effects are additive across grades.

    “Finally, our approach assumes that teacher effects are additive across grades.”

  • The authors state this as an assumption.

    The percentage impact of a one standard deviation improvement in teacher value-added on earnings observed at age 28 is constant at b = 1.34 percent over the life cycle.

    “First, we assume that the percentage impact of a one standard deviation improvement in teacher VA on earnings observed at age 28 is constant at b = 1.34 percent (Table 3, column 2) over the life cycle.”

  • The authors state this as an assumption.

    General equilibrium effects which may reduce wage rates if all children are better educated are ignored.

    “Second, we ignore general equilibrium effects which may reduce wage rates if all children are better educated.”

  • The authors state this as an assumption.

    Earnings gains are discounted at a 3 percent real annual rate back to age 12.

    “Third, we follow Krueger (1999) and discount earnings gains at a 3 percent real annual rate (consistent with a 5 percent discount rate and 2 percent wage growth) back to age 12, the average age in our sample.”

  • The authors state this as an assumption.

    Teacher value-added is normally distributed.

    “Finally, we assume that teacher VA m jt is normally distributed.”

  • The authors state this as an assumption.

    Teachers are evaluated based purely on their value-added in one subject, and value-added in the other subject is ignored.

    “For simplicity, we assume that teachers are evaluated based purely on their VA in one subject (say math) and VA in the other subject is ignored.”

  • The authors state this as an assumption.

    The calculations ignore downstream costs that may be required to generate earnings gains, most notably the cost associated with higher college attendance rates.

    “First, they ignore downstream costs that may be required to generate earnings gains, most notably the cost associated with higher college attendance rates.”

  • The authors state this as an assumption.

    The calculations ignore the fact that teachers need to be compensated for the added employment risk they face from such an evaluation system.

    “Second, and more importantly, they ignore the fact that teachers need to be compensated for the added employment risk they face from such an evaluation system.”

  • The paper relies on this without support.

    There is no evidence to date on the long-term impacts of teacher quality as measured by value-added.

    “but there is no evidence to date on the long-term impacts of teacher quality as measured by value-added.”

  • The authors state this as an assumption.

    A one standard deviation improvement in teacher VA raises earnings by 1.34 percent in all subjects and grade levels.

    “We therefore assume that a one standard deviation improvement in teacher VA raises earnings by 1.34 percent in all subjects and grade levels in the calculations that follow.”

  • The authors state this as an assumption.

    Teachers teach one class per year.

    “All calculations assume that teachers teach one class per year and report mean values for math and English teachers, which are calculated separately.”
29 of 29

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • There is substantial variability in teacher job performance, measured in the short run by students' test scores and in the long run by students' economic and social success as adults.

    “And there is substantial variability in teacher job performance: measured both in the short-run with students' test scores (see Jackson, Rockoff, and Staiger 2014 for a review) and the long-run with students' economic and social success years later as adults (Chetty, Friedman, and Rockoff 2014).”

    From Learning Job Skills from Colleagues at Work: Evidence from a Field Experiment Using Teacher Performance Data (2020) · cites this paper as Chetty, Friedman, and Rockoff 2014

  • Chetty et al. (2014) estimate that a 1 standard deviation increase in teacher value-added corresponds with a 0.4 percentage point increase in adult employment.

    “Chetty et al. (2014) estimate that a 1 standard deviation increase in teacher value-added (VA) corresponds with a 0.4 percentage point increase in adult employment.”

    From A More Credible Approach to Parallel Trends (2023) · cites this paper as Chetty et al. (2014)

  • VA measures which control for lagged test scores exhibit little or no bias.

    “We addressed the first question in the previous paper in this volume (Chetty, Friedman, and Rockoff 2014) and concluded that VA measures which control for lagged test scores exhibit little or no bias.”

    From Measuring the Impacts of Teachers II: Teacher Value-Added and Student Outcomes in Adulthood (2014) · cites this paper as Chetty, Friedman, and Rockoff 2014

  • Conditioning on prior test scores and other observables is adequate to obtain unbiased estimates of teachers' causal impacts on test scores.

    “Our first method of estimating teachers' long-term impacts builds on our finding that conditioning on prior test scores and other observables is adequate to obtain unbiased estimates of teachers' causal impacts on test scores (Chetty, Friedman, and Rockoff 2014).”

    From Measuring the Impacts of Teachers II: Teacher Value-Added and Student Outcomes in Adulthood (2014) · cites this paper as Chetty, Friedman, and Rockoff 2014

  • Existing test-score value-added measures are a good proxy for a teacher's ability to raise students' test scores.

    “Our first paper (Chetty, Friedman, and Rockoff 2014) showed that existing testscore value-added measures are a good proxy for a teacher's ability to raise students' test scores.”

    From Measuring the Impacts of Teachers II: Teacher Value-Added and Student Outcomes in Adulthood (2014) · cites this paper as Chetty, Friedman, and Rockoff 2014

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Teacher-quality changes are described as both entirely staff-driven and partly due to drift

    Minor

    The methodology says changes in mean teacher VA are driven purely by staff changes. A footnote instead says drift in predicted VA for continuing teachers accounts for 5.5 percent of their variance.

1 of 1

Cites

Flagged first, then by citations

Research works printed but not linked: 4

  1. Chetty Raj, 2014, American Economic Review. http://dx.doi.org/10.1257/aer., 104, 2633, 10.1257/aer.104.9.2633 · DOI
  2. Gordon Robert, 2006, The Brookings Institution, Hamilton Project Discussion Paper, 2006
  3. Hanushek Eric A, 1971, American Economic Review, 61, 280
  4. 2010, Calendar Year Projections of Information and Withholding Documents for the United States and IRS Campuses, 2010

Cited by

Sorted by citations

Related

Authors
Chetty, R., Friedman, J. N., & Rockoff, J. E.
Affiliations
  • Harvard University (US)
  • Columbia University (US)
ORCID
Chetty, Raj
Venue
American Economic Review
Publisher
American Economic Association
Published
2014-09-01
Volume
104
Issue
9
Pages
2633-2679
ISSN
0002-8282 (ISSN-L)
DOI
10.1257/aer.104.9.2633
OpenAlex
W1976049077
Cited by
1,717
References
55
Citations by year
2026: 116 · 2025: 169 · 2024: 146 · 2023: 180 · 2022: 181 · 2021: 153 · 2020: 142 · 2019: 145 · 2018: 134 · 2017: 92 · 2016: 144 · 2015: 73 · 2014: 32 · 2013: 7 · 2012: 1
FWCI
584.51
Type
journal article
Language
en
Open access
Locations
  • American Economic Review (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • Digital Access to Scholarship at Harvard (DASH) (Harvard University) (submittedVersion, cc-by)
Topics
  • Social Sciences › Education › School Choice and Performance
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Labor market dynamics and wage inequality
  • Social Sciences › Education › Teacher Education and Assessments
Keywords
teacher value-added, teacher quality, student outcomes, college attendance, adult earnings, teenage childbearing, lifetime income, teacher assignment

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-11.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-11.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review