The Effects of Cognitive and Noncognitive Abilities on Labor Market Outcomes and Social Behavior

The Effects of Cognitive and Noncognitive Abilities on Labor Market Outcomes and Social Behavior Heckman Stixrud Urzua

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

The Effects of Cognitive and Noncognitive Abilities on Labor Market Outcomes and Social Behavior 2006 Journal of Labor Economics 10.1086/504455 Heckman2006-the-effects-of-cognitive-and-noncognitive-ab Heckman2006-the-effects-of-cognitive-and-noncognitive-ab.txt, the project's staged copy 623db8115d4ebeb7630dd3c1d06e4f2ebd5ceebc not-redistributable economics

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them existence Quotation withheld because this record is not cleared for redistribution. cognitive and noncognitive skills as determinants of labour-market and behavioural outcomes Quotation withheld because this record is not cleared for redistribution. observational-study observational-study Quotation withheld because this record is not cleared for redistribution. mechanisms A single low-dimensional pair of latent abilities acts as the common cause generating the observed clustering of risky adolescent behaviors together with schooling and wage outcomes. Quotation withheld because this record is not cleared for redistribution. mechanisms Noncognitive skills raise wages along two routes: directly by raising productivity, and indirectly by raising schooling attainment and accumulated work experience. Quotation withheld because this record is not cleared for redistribution. mechanisms The GED wage penalty arises because GED holders' deficit in noncognitive skill lowers employment and work-experience accumulation and removes a trait the market pays for. Quotation withheld because this record is not cleared for redistribution. The paper does not estimate the separate channels (preferences, discount rate, human-capital productivity, market productivity) through which the abilities operate. what this limits: any mechanistic reading of the results: WE DO NOT ESTIMATE RELATIONSHIPS FOR EACH OF THE CHANNELS THROUGH WHICH COGNITIVE AND NONCOGNITIVE ABILITIES MIGHT OPERATE Quotation withheld because this record is not cleared for redistribution. Collapsing noncognitive skill into a single latent factor is unlikely to capture all its aspects. what this limits: every claim stated in terms of noncognitive skill: THERE ARE MANY ASPECTS ... SELF CONTROL, TIME PREFERENCE, SOCIABILITY ... IT IS LESS LIKELY THAT ONE TRAIT CAPTURES ALL ASPECTS - the construct is collapsed to one latent factor Quotation withheld because this record is not cleared for redistribution. Support conditions for semiparametric identification are not met, so results rest on the parametric structure of the model. what this limits: every estimate in the paper: THE EMPIRICAL RESULTS THAT WE GENERATE ARE IDENTIFIED FROM THE PARAMETRIC STRUCTURE OF THE MODEL - the support conditions for semiparametric identification are not met, so the answers come from the assumed form Quotation withheld because this record is not cleared for redistribution. The underlying theoretical model is static and ignores the timing of decisions. what this limits: any conclusion about sequencing or timing: OUR THEORETICAL MODEL IS STATIC AND DOES NOT CONSIDER THE TIMING OF DECISIONS Quotation withheld because this record is not cleared for redistribution. The corroborating CNLSY79 data lack adult outcomes and have small samples. what this limits: the CNLSY79 corroboration only: MANY OF THE CHILDREN ARE STILL YOUNG AND WE LACK INFORMATION ON THEIR WAGES, OCCUPATIONAL STATUS AND EMPLOYMENT AT AGE 30 Quotation withheld because this record is not cleared for redistribution. The noncognitive measures used (Rotter, Rosenberg) are narrower than would be ideal and are not linked to economic preference parameters. what this limits: the interpretation of the noncognitive measures: IT WOULD BE BETTER TO USE A WIDER ARRAY OF PSYCHOLOGICAL MEASUREMENTS AND ... TO CONNECT THEM WITH MORE CONVENTIONAL MEASURES OF PREFERENCE PARAMETERS IN ECONOMICS Quotation withheld because this record is not cleared for redistribution. The scale of the two latent abilities is not measured but set: two normalizations are stated to be required for identification, and they fix the scale of the factors. what this limits: the magnitude of a factor score as a quantity of ability: the paper states two normalizations are required for identification and that they set the scale of the factors Quotation withheld because this record is not cleared for redistribution. The direction of the bias in the conventional estimates is stated to be unknown, because the problems push in different directions. what this limits: treating an OLS estimate as a bound on the effect: the paper states the biases run in different directions so the sign of the difference from its own model cannot be predicted Quotation withheld because this record is not cleared for redistribution. The measures used are declared imperfect: the attitude scales proxying noncognitive ability are affected by family background and are called at best imperfect measures of true ability. what this limits: any test score read as the ability itself: the paper states these scores are affected by family background and are at best imperfect measures of an individual's true abilities Quotation withheld because this record is not cleared for redistribution. What the construct covers is bounded by the instrument available in the survey, and the authors say a wider array of psychological measurement would have been better. what this limits: the noncognitive factor as the full space of noncognitive ability: the authors state a wider array of psychological measurements would ideally have been used Quotation withheld because this record is not cleared for redistribution. The channels through which the abilities act are not separately estimated; the model is a linear-in-the-parameters approximation. what this limits: attributing an effect to a particular channel: the paper states it does not estimate the relationships for each channel separately and that its model approximates the general one Quotation withheld because this record is not cleared for redistribution. An interchangeability the paper asserts in principle is explicitly deferred rather than shown. what this limits: the claim that test scores and behaviours are interchangeable proxies: the paper states this in principle and leaves the task for future work Quotation withheld because this record is not cleared for redistribution. The disputed word in this literature is 'importance', and the paper's claim holds under one sense of it and reverses under another that the paper states in the same sentence: cognitive skills explain much more of the variance of log wages, while the skill gradients of the two abilities are similar. The paper marks its choice ('for the sense of importance adopted in this paper') rather than leaving it implied, which is what makes the reversal checkable at all. Quotation withheld because this record is not cleared for redistribution. The authors take the latent-variable measurement models built in psychometrics and extend them so that a choice variable (schooling) and background variables can enter the measurement equations of an economic model. Quotation withheld because this record is not cleared for redistribution. two psychology instruments - the Rotter Locus of Control Scale and the Rosenberg Self-Esteem Scale - are imported into labour economics to stand for noncognitive skill Quotation withheld because this record is not cleared for redistribution. cognitive skill is measured with five subtests of the Armed Services Vocational Aptitude Battery, an instrument built for military selection and re-used as an economic variable Quotation withheld because this record is not cleared for redistribution. holding cognitive ability constant when comparing GED recipients with high school dropouts eliminates: cognitive ability as the explanation of any wage difference attributed to the credential - and with it removed, the difference reverses rather than shrinking · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. noncognitive skills noncognitive skills Quotation withheld because this record is not cleared for redistribution. cognitive skills cognitive skills Quotation withheld because this record is not cleared for redistribution. normalizations normalizations Quotation withheld because this record is not cleared for redistribution. measurement error measurement error Quotation withheld because this record is not cleared for redistribution. ASVAB ASVAB Quotation withheld because this record is not cleared for redistribution. Rotter Locus of Control Scale Rotter Locus of Control Scale Quotation withheld because this record is not cleared for redistribution. coding speed coding speed about this quote: DAMAGED GLYPHS IN THIS QUOTE: (cid:48). These are raw PDF font glyph indices that leaked out in place of the characters they stand for - usually a mathematical operator. The quote is verbatim to the staged text and its anchor is sound; the STAGED TEXT is what is damaged. NOT repaired: the cid-to-character mapping is defined by the font embedded in that particular PDF, so guessing it would produce a clean and unverifiable quote. Repair needs the source PDF and is a re-staging job. Quotation withheld because this record is not cleared for redistribution. AFQT AFQT Quotation withheld because this record is not cleared for redistribution. schooling schooling Quotation withheld because this record is not cleared for redistribution. ability → schooling THE TWO CANNOT BE SEPARATED IN THIS STUDY - the DESIGN form / It bounds what the paper's own results can mean, and it lives in a methods paragraph where nothing downstream will find it. The confound is between the paper's own explanatory variables, so it bounds every coefficient in the table it is attached to. Quotation withheld because this record is not cleared for redistribution. noncognitive skills - AS MEASURED: scores on attitude scales, which the paper calls at best imperfect measures of the individual's true ability Quotation withheld because this record is not cleared for redistribution. noncognitive skills - AS MODELLED: a latent factor whose scale is fixed by normalization rather than by measurement Quotation withheld because this record is not cleared for redistribution. Bayesian methods - used here for computation only, with the identification argument stated to be strictly classical and the priors asymptotically irrelevant Quotation withheld because this record is not cleared for redistribution. test score - a quantity that depends on schooling completed AT THE TIME OF THE TEST, so the measurement is not independent of the outcome being explained Quotation withheld because this record is not cleared for redistribution. importance - IN THE SENSE THIS PAPER ADOPTS: the size of the skill gradient, how much an outcome moves when a person is moved across the distribution of one ability. ON THE RIVAL MEASURE THE RANKING REVERSES AND THE PAPER SAYS SO IN THE SAME SENTENCE: cognitive skills 'explain much more of the variance of (log) wages', while by skill gradients the two are similar about this quote: PAGE FURNITURE IS SPLICED INTO THE 'evidence' QUOTE - the text of footnote 33 and a page number, between 'their effects on (log)' and 'wages (as measured'. The words on either side of it were never continuous on the page; read the quote with it deleted. Quotation withheld because this record is not cleared for redistribution. importance - A THIRD SENSE THE PAPER NAMES AND EXPLICITLY DOES NOT CLAIM: equal strength in its own sense 'does not translate into equal cost of changing these skills', so the finding is silent on which ability is cheaper to change Quotation withheld because this record is not cleared for redistribution. contradicts → the "g" theory of Herrnstein and Murray (1994) and Jensen (1998) Quotation withheld because this record is not cleared for redistribution. builds-on → Hansen, Heckman, and Mullen (2004) Quotation withheld because this record is not cleared for redistribution. extends → the LISREL model of Jöreskog (1977) and the MIMIC model of Jöreskog and Goldberger (1975) Quotation withheld because this record is not cleared for redistribution. replicates → Neal and Johnson (1996); Hansen, Heckman, and Mullen (2004); Heckman, Larenas, and Urzua (2004) Quotation withheld because this record is not cleared for redistribution. contradicts → Herrnstein and Murray (1994) and Herrnstein and Wilson (1985) on cognitive ability and crime Quotation withheld because this record is not cleared for redistribution. contradicts → the single-skill signalling model of Arrow (1973) and Spence (1973) Quotation withheld because this record is not cleared for redistribution. cognitive skills —operationalised-as→ ASVAB cognitive skills ASVAB strength of the relation: 0.9 · note: Five subtests of a battery built to decide military eligibility - the paper says elsewhere that the AFQT composite 'is a general measure of trainability and a primary…'. An instrument designed for a selection decision is here the measure of the construct. Quotation withheld because this record is not cleared for redistribution. noncognitive skills —operationalised-as→ Rotter Locus of Control Scale noncognitive skills Rotter Locus of Control Scale strength of the relation: 0.85 · note: THE PAPER STATES ITS REASON FOR CHOOSING THEM AND IT IS NOT FITNESS: '…these measures because of their availability in the NLSY…' The construct carrying the paper's headline claim against Herrnstein and Murray is measured by a four-item abbreviated locus-of-control scale and a self-esteem scale, picked for what the survey happened to contain. Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span writes 'noncognitive measures', not 'noncognitive skills'. Quotation withheld because this record is not cleared for redistribution. cognitive skills —operationalised-as→ coding speed cognitive skills coding speed strength of the relation: 0.8 · note: THE UNIT OF THE LATENT ABILITY IS SET BY WHICHEVER INDICATOR WAS GIVEN THE 1. The cognitive factor's loading is normalised to 1 on CODING SPEED (ASVAB 5) and the noncognitive factor's on the Rosenberg Self-Esteem Scale, both 'for individuals in grades 9 to 11 at the time of the test' - so the scale is fixed by an indicator choice AND a schooling subgroup. The record already carries the existence of the normalisation; this names what the units are. NOTE THE HEADLINE COMPARISON DOES NOT DEPEND ON IT: the paper compares the two abilities by moving people between DECILES of each factor's own distribution, which is invariant to the scale. Quotation withheld because this record is not cleared for redistribution. schooling —causes→ AFQT schooling AFQT strength of the relation: 0.9 · note: THE PROXY IS CAUSED BY THE OUTCOME IT IS USED TO EXPLAIN: each additional year of schooling raises measured AFQT by 2 to 4 percentage points on average, and '…THIS CREATES A REVERSE CAUSALITY…' problem. The paper says flatly in its conclusion that schooling '…raises measured cognitive…' and noncognitive ability. Naming this is the reason the paper builds a latent-factor model instead of regressing on test scores. Quotation withheld because this record is not cleared for redistribution. ordinary least squares regression ordinary least squares regression Quotation withheld because this record is not cleared for redistribution. probit / multinomial discrete choice models probit / multinomial discrete choice models Quotation withheld because this record is not cleared for redistribution. Bayesian Markov chain Monte Carlo estimation Bayesian Markov chain Monte Carlo estimation Quotation withheld because this record is not cleared for redistribution. chi-squared test chi-squared test Quotation withheld because this record is not cleared for redistribution. goodness-of-fit test goodness-of-fit test Quotation withheld because this record is not cleared for redistribution. factor analysis factor analysis Quotation withheld because this record is not cleared for redistribution. correlation correlation Quotation withheld because this record is not cleared for redistribution. standard error standard error Quotation withheld because this record is not cleared for redistribution. A low-dimensional vector of cognitive and noncognitive skills explains a variety of labor-market and behavioral outcomes. Quotation withheld because this record is not cleared for redistribution. For many dimensions of social performance, cognitive and noncognitive skills are equally important. Quotation withheld because this record is not cleared for redistribution. Noncognitive skills shape schooling choices and additionally raise wages conditional on schooling. Quotation withheld because this record is not cleared for redistribution. Schooling raises measured test scores. Quotation withheld because this record is not cleared for redistribution. Male GED holders have a worse noncognitive ability distribution than dropouts. Quotation withheld because this record is not cleared for redistribution. Noncognitive rather than cognitive ability is what mainly explains differences in criminal participation. Quotation withheld because this record is not cleared for redistribution. The two latent skills carry different hedonic prices in the labor markets of different schooling levels. Quotation withheld because this record is not cleared for redistribution. Estimated returns to ability shrink when test scores are corrected for the schooling attained at test date, showing that endogenous test scores bias conventional OLS estimates upward. Quotation withheld because this record is not cleared for redistribution. GED GED Quotation withheld because this record is not cleared for redistribution. AFQT AFQT Quotation withheld because this record is not cleared for redistribution. illegal activities illegal activities Quotation withheld because this record is not cleared for redistribution. Rotter Internal-External Locus of Control Scale Rotter Internal-External Locus of Control Scale Quotation withheld because this record is not cleared for redistribution. risky behaviors risky behaviors Quotation withheld because this record is not cleared for redistribution. cognitive skills cognitive skills Quotation withheld because this record is not cleared for redistribution. schooling level at the time of the test schooling level at the time of the test Quotation withheld because this record is not cleared for redistribution. gender gender Quotation withheld because this record is not cleared for redistribution. ethnicity dummies ethnicity dummies Quotation withheld because this record is not cleared for redistribution. cohort dummies cohort dummies Quotation withheld because this record is not cleared for redistribution. local labor market conditions local labor market conditions Quotation withheld because this record is not cleared for redistribution. region of residence dummies region of residence dummies Quotation withheld because this record is not cleared for redistribution. work experience work experience Quotation withheld because this record is not cleared for redistribution. choice of occupation choice of occupation Quotation withheld because this record is not cleared for redistribution. employment employment Quotation withheld because this record is not cleared for redistribution. schooling schooling Quotation withheld because this record is not cleared for redistribution. 9% — share of the variance of log wages explained by the cognitive ability measure when schooling is not controlled for Quotation withheld because this record is not cleared for redistribution. 0.9% — share of the variance of log wages explained by the noncognitive measure for men Quotation withheld because this record is not cleared for redistribution. 77% — proportion of the trace of the male cognitive test-score correlation matrix explained by a single factor Quotation withheld because this record is not cleared for redistribution. 31% — proportion of the trace of the noncognitive correlation matrix explained by a single factor, much lower than for cognitive scores Quotation withheld because this record is not cleared for redistribution. 2 to 4 percentage points — increase in measured AFQT score produced by each additional year of schooling Quotation withheld because this record is not cleared for redistribution. 66% — share of the Perry Preschool treatment group graduating from high school by age 18 Quotation withheld because this record is not cleared for redistribution. 431.65 with 150 degrees of freedom — chi-squared statistic rejecting the null that schooling does not affect measured cognitive tests for males Quotation withheld because this record is not cleared for redistribution. 116.53 with 40 degrees of freedom — chi-squared statistic rejecting the null that schooling does not affect the noncognitive measurement equations Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1086/504455.