II. Human Capital, Returns to Education and...

Fortin – Econ 560 Lecture 2B

II. Human Capital, Returns to Education and Experience

2. Signaling, Screening, and Ability

1. Human Capital Theory vs. Signaling

2. Spence’s Signaling Model

3. Empirical Evidence of Signaling a. Overview b. Compulsory Schooling Laws (CAL) c. Use College Proximity as a constraint to college enrollment d. Variation in GED passing standards across U.S. states


2.1 Human Capital Theory vs. Signaling

According to human capital theory, school and on-the-job training enhance productivity: o This explains why workers with higher levels of education and more

experiences are paid higher wages o But this may not explain all the wage differences associated with schooling

and work history. o Better-educated workers are not a random sample of workers (generally!).

Returns to education reflect not just the productivity enhancing effect of education (HC theory) but an effect on earnings of the underlying ability that education signals (Spence, 1973).

Fundamental difficulty in unraveling the extent to which education is a signal of existing productivity as opposed to enhancing productivity


In both theories the wage equation is a hedonic function o both theories imply that there is a positive correlation between earnings and

education o “This fact makes it virtually impossible to come up with a valid test of the

screening hypothesis …..” (Lazear 1977) o So we have to look into the details to sort it out


2.2 Spence’s Signaling Model in a Nutshell

Signaling takes its root in the fact employers have less information about the

ability of potential workers that the workers themselves, there is asymmetric information.

To simplify, let’s assume that there are only two types of workers: low ability workers and high ability workers

1) High ability people are inherently more productive than low ability people. 2) High ability people have lower cost of attending school than others (Why?). 3) Education does not affect productivity. 4) The employer does not know whether a worker is L or H ability


Spence (1973, 2002) uses the following parameters as an example Group Productivity Population Share Cost of Education L 1 E H 2 )1( 2/E

Low productivity worker has an incentive to lie and pass for a high productivity

worker

In the absence of information, the employer could pay the average productivity of its workforce: 2)1(21 Low productivity workers will like it High productivity workers do not like it Employers do not like it because they cannot properly allocate workers to

jobs

Employers would prefer a separating equilibrium to properly allocate workers to job, they need credible information about the high productivity workers


Education, which is more costly for low productivity people, can serve as a signal under certain conditions

Assume that firms offer the following wage schedule: ][1)( *EEIEw (1)

where I[·] is the indicator function. A worker with ≥ *E years of education is paid 2 and otherwise 1. o If *E perfectly separates the two types of workers, each will be paid its

average productivity

How much education will workers’ obtain? The worker’s problem is:

),()()(max EcEwE aaE

For a type H worker, the cost of *E years

of education is 2/*E and the wage benefit is 1. So type H workers will attend school for *E years if: 2 − 2/*E > 1.

= Ch

E=Cl


For a type L worker, the cost of *E years of education is *E , and the wage benefit is 1 so L type workers will attend school if: 2 − *E > 1.

Consider if the employer sets 1*E , where ε is a very small, positive number. Who will obtain education? o For type H workers, since 2 − hC > 1, they will get education *E o For type L workers, since 2 − lC < 1, they will get 0E

Is the employer’s wage schedule, represented by (1), an equilibrium (separating)? An equilibrium price/wage schedule needs to be internally consistent: employer’s

find it worthwhile to pay the wages offered given the productivity of workers who claim these wages, 훾(푤).

Hence, in equilibrium 피(훾(푤)) = 푤. The expected productivity of workers

accepting the wage given the wage schedule must be equal to the wage.


Here, the function 훾(푤) gives the productivity of workers supplying labor at wage w. o In this case, workers with E = 0 are type L. They have productivity 1 and

wage 1 and so the employer’s wage schedule is rational for these workers: 푤(퐸 = 0) = 피(훾[퐸 = 0]) = 1.

o Moreover, workers with 퐸 = 퐸∗ are type H. They have productivity 2 and wage 2 and so the employer’s wage schedule is also rational for these workers: 푤(퐸 = 퐸∗) = 피(훾[퐸 = 퐸∗]) = 2.

So, this is a separating

equilibrium: o high ability workers will

obtain 퐸 = 퐸∗education, o low worker will obtain 퐸 = 0

education, o employers will break even, o neither H or L workers or

employers will have incentive to deviate from the pay scheme.


By obtaining education, H type workers ‘signal’ that they deserve a high wage —

but because education is completely unproductive, this is a pure private benefit.

From a social perspective, this signaling is wasteful; it could lead to overinvesting in education

Are there some pooling equilibria?

o Yes, in cases where the reward to high ability workers or the cost of low ability workers was too low. Then neither low and high ability workers gets any education or obtain a low level of schooling.

For example, suppose that employers offer the wage schedule

]3[)2()( EIEw Not even the H type workers will buy education, since

)2(2/31)2(

Here the employer’s beliefs are self-confirming since the pool of (uneducated) workers has have productivity 2-훼 is equal to the wage: (훾[푤]) = 2 − 훼 .


Does a pooling equilibrium make sense? o It is supported by the belief that the worker who gets education is no better

than a worker who does not o But education is more costly for low ability workers, so they should be less

likely to deviate to obtaining education o If there exists a type who will never benefit from taking a particular

deviation, then the uninformed parties (e.g. firms) should deduce that this deviation is very unlikely to come from this type

o The overall conclusion is that if education is a valuable signal, then the separating equilibrium may be more likely than pooling equilibrium

Policy Implications o In a case of pure signaling, where education has positive private returns

but does not make workers more productive, it may turn out to be a relatively expensive sorting mechanism, and the public funding of education, especially higher education, is questioned. The debate is not so much about whether there should be any public

funding at all, but what the correct level of funding should be.


2.3 Empirical Evidence of Signaling a. Overview

Human capital theory and signaling models share many of the same empirical

implications o People who attend additional years of schooling are more productive

(through human capital acquisition in HC, through sorting in signaling). o People who attend additional years of schooling receive higher wages. o People will attend school while they are young, i.e., before they enter the

workforce.

The hypothesis that employers have little information about the ability of their workers make more sense for new workers, thus opens up a different set of important questions about employer learning (Farber and Gibbons (1996), Altonji and Pierret (2001), Lange (2007)) o How fast do employers “learn” about the ability of their workers? o If the signal is different for different groups, how is this related to

discrimination?


o Altonji and Pierret argue that as employers learn about the productivity of workers, schooling S will get less of the credit for an association with productivity that arises because S is correlated with ability Z, that unobserved by the employer, but available to the econometrician!

Considering new workers, how can we empirically distinguish the human capital and signaling models? o Keeping in mind that the ideal test of signaling would find people of

identical ability and randomly assign them a diploma and see if the ones with the diploma earn more.

Different types of evidence

o Some of these are more or less informative in distinguishing the signaling model from human capital

o Note that education can be both a signal and productivity enhancing o If both models are operating, may be hard to identify which is dominant all

the time.

Earlier empirical strategies

TABLE ITHE EFFECTS OF STANDARDIZED AFQT AND SCHOOLING ON WAGES

Dependent Variable: Log Wage; OLS estimates (standard errors).

Panel 1—Experience measure: potential experience

Model: (1) (2) (3) (4)

(a) Education 0.0586 0.0829 0.0638 0.0785(0.0118) (0.0150) (0.0120) (0.0153)

(b) Black �0.1565 �0.1553 0.0001 �0.0565(0.0256) (0.0256) (0.0621) (0.0723)

(c) Standardized AFQT 0.0834 �0.0060 0.0831 0.0221(0.0144) (0.0360) (0.0144) (0.0421)

(d) Education � �0.0032 �0.0234 �0.0068 �0.0193experience/10 (0.0094) (0.0123) (0.0095) (0.0127)

(e) Standardized AFQT � 0.0752 0.0515experience/10 (0.0286) (0.0343)

(f) Black � experience/10 �0.1315 �0.0834(0.0482) (0.0581)

R2 0.2861 0.2870 0.2870 0.2873

Panel 2—Experience measure: actual experience instrumentedby potential experience

Model: (1) (2) (3) (4)

(a) Education 0.0836 0.1218 0.0969 0.1170(0.0208) (0.0243) (0.0206) (0.0248)

(b) Black �0.1310 �0.1306 0.0972 0.0178(0.0261) (0.0260) (0.0851) (0.1029)

(c) Standardized AFQT 0.0925 �0.0361 0.0881 0.0062(0.0143) (0.0482) (0.0143) (0.0572)

(d) Education � �0.0539 �0.0952 �0.0665 �0.0889experience/10 (0.0235) (0.0276) (0.0234) (0.0283)

(e) Standardized AFQT � 0.1407 0.0913experience/10 (0.0514) (0.0627)

(f) Black � experience/10 �0.2670 �0.1739(0.0968) (0.1184)

R2 0.3056 0.3063 0.3061 0.3064

Experience is modeled with a cubic polynomial. All equations control for year effects, education inter-acted with a cubic time trend, Black interacted with a cubic time trend, AFQT interacted with a cubic timetrend, two-digit occupation at first job, and urban residence. For these time trends, the base year is 1992. Forthe model in Panel 1 column (1) the coefficient on AFQT and Black are .0312 and �.1006, respectively, whenevaluated for 1983. In Panel 2 the instrumental variables are the corresponding terms involving potentialexperience and the other variables in the model. Standard errors are White/Huber standard errors computedaccounting for the fact that there are multiple observations for each worker. The sample size is 21,058observations from 2976 individuals.

330 QUARTERLY JOURNAL OF ECONOMICSSource: Altonji and Pierret (2001)

Nicole

Rectangle

Nicole

Rectangle

Nicole

Rectangle

Nicole

Rectangle

Nicole

Rectangle

Nicole

Rectangle

Nicole

Line


1) Sheepskin effects The increase in wages associated with obtaining a higher credential is sometimes

referred to as the “Sheepskin Effect”, since in medieval times diplomas were written on parchment made of sheepskin

Human Capital => no wage premia associated with credential years unless more productive

Signaling => large wage premia associated with credential years For example, Jaeger and Page (1996 ) after controlling for actual years of

schooling find sizeable premia for High School, Associate and Bachelor’s degrees, as well as Professional degrees, but no significant premia for Master’s and Ph.D.’s

Some high-school and college drop-outs may be lower ability and knowledge may itself be discrete

2) Curriculum effects Signaling => curriculum should not matter Human Capital => “useful” curriculum should matter Graduates earn more in jobs that use their subject specific training

NOTES 735

TABLE 2.-EsTIMATED DIPLOMA EFFECTs FOR WHITE MEN USING DmmRENT SPECIFICATIONS

Model

Coefficient (1) (2) (3) (4)

COmPleted Years of Education (Spline) Years of Education (S) 0.076 0.076

(0.018) (0.018) S 8 -0.141 -0.112

(0.080) (0.078) (S2 8) * (S - 8) 0.002 -0.022

(0.027) (0.023) S 2 12 0.034

(0.053) (S 2 12) (S - 12) -0.006 -0.019

(0.022) (0.017) S 16 0.114

(0.035) S= 17 -0.055

(0.042) S= 18 -0.006

(0.031) Completed Years of Education (Dummy)

9 -0.227 -0.109 (0.049) (0.061)

10 -0.164 -0.046 (0.040) (0.054)

11 -0.137 -0.044 (0.043) (0.051)

12 ref. ref.

13 0.089 0.020 (0.027) (0.033)

14 0.167 0.073 (0.022) (0.031)

15 0.166 0.052 (0.038) (0.044)

16 0.406 0.178 (0.019) (0.045)

17 0.422 0.164 (0.039) (0.057)

18 or more 0.544 0.224 (0.023) (0.054)

Diploma Effects High School 0.106 0.123

(0.037) (0.041) Marginal Effect Over High School

Some College, No Degree 0.074 0.083 (0.022) (0.027)

Occupational Associate's 0.074 0.076 (0.039) (0.043)

Academic Associate's 0.188 0.191 (0.042) (0.046)

Bachelor's 0.273 0.245 (0.038) (0.045)

Marginal Effect Over Bachelor's Master's 0.032 0.050

(0.030) (0.041) Professional 0.271 0.286

(0.050) (0.059) Doctoral 0.052 0.067

(0.058) (0.067) R 2 0.145 0.153 0.147 0.154 Adjusted R2 0.144 0.151 0.145 0.151 Mean Square Error 0.372 0.369 0.372 0.369

Note: Dependent variable is log hourly wages. Estimated using ordinary least squares. Standard errors are in parentheses. Calculated from a matched sample of individuals 25 to 64 years old from the 1991 and 1992 March Current Population Survey. Model also includes Potential Experience and Potential Experience Squared as covariates. Columns (3) and (4) also include dummy variables for zero through eight completed years of education. Sample size is 8,957.

Source: Jaeger and Page (1996)

Nicole

Rectangle


Compensating differentials and endogenous subject choice Returns to additional courses in academic subjects in high school is small

(Altonji, 1995) Large math effects on earnings of high school graduates (Rose and Betts, 2001) Kane and Rouse (1995) showed that, among those who failed to earn the degree,

the number of credits (i.e., partial completion of a Two-Year College's degree, for instance) does matter.

3) Speed of completion of education Signaling => slow/fast completion penalty/bonus But slow/fast completers are unobservably different, especially using grade

repetitions Brodaty, Gary-Bobo and Prietoz (2008) find that a year of delay causes a 9%

decrease in the student's wage using an instrumental variables strategy based on school openings in France where grade repetitions is frequent (45% among boys)


Empirical strategies using an exogenous source of variation in education B) Compulsory Schooling Laws (CAL) Lang and Kropp (1986) exploits cross state changes in minimum school leaving

age laws

Compulsory schooling should have different effects on educational attendance depending on whether education is a tool for human capital accumulation or if it is a form of signaling

Human Capital => a change in the law will not change the return to education o Unless there are important supply effects not corrected by factor price

equalization across states, but excess supply more likely on education level affected by CAL

Workers for whom the return to schooling sufficiently low as to choose dropping out will be bound to stay in school if they are affected by the law

Individuals who are not bound by the law should be largely unaffected


min CAL rise only increases schooling of workers constrained by the minimum

Signaling => the lowest ability types increase their educational attainment because of law, some higher ability types must increase schooling to distinguish themselves from the lowest type.

If laws impose a certain minimum signal for the lowest type, then the signaling hypothesis suggests some individuals not bound by the law will still increase their educational attainment.

min CAL rise increases schooling of late (high ability) leavers as well as early leavers: significant "ripple" effects

Underlying assumption: Individuals who choose the level of compulsory schooling (Sc) as their optimal level of schooling will be observationally equivalent to those who would choose S < SC but are forced to stay

Table I gives the distribution of minimum school leaving ages over our sample. Somewhat surprisingly, there is not a steady increase in CALs over our sample period.

618 QUARTERLY JOURNAL OF ECONOMICS

IV. EMPIRICAL METHODS

To test the sorting and human capital models against each other, we measure the effect of compulsory attendance laws on educational enrollments for different age groups. We test the hypothesis that a CAL increases the enrollment rate of age groups not directly affected by it against the hypothesis that it does not affect this rate.

Enrollment rates for sixteen- and seventeen-year-olds were obtained from the 1910, 1920, 1930, 1950, 1960, and 1970 censuses. The 1940 census was not included because differences in data gathering methods make the enrollment figures not com- parable with other censuses.

Data on state CALs proved surprisingly difficult to obtain. In the end, CALs for 1908, 1918, 1928, 1945, 1958, and 1965 had to be used. The fact that the CALs precede the sample date by two to five years may actually be advantageous if their effect operates with a lag. CALs differ considerably across states. Table I gives the distribution of minimum school leaving ages over our sample. Somewhat surprisingly, there is not a steady increase in CALs over our sample period. In 1918, all states had CALs. By 1965, Mississippi and South Carolina had abolished their laws. Moreover, the number of states with a minimum school leaving age of 18 declined from four in 1945 to three in 1965.

The coverage of CALs varies among states. All have some exceptions to universal coverage, generally for those who are "mentally unable to benefit from education" or for individuals whose work is needed for the support of the household. Some laws extend coverage through the entire school year so that individuals reaching the minimum school leaving age in the middle of the

TABLE I

THE DISTRIBUTION OF COMPULSORY ATTENDANCE LAWS

Year No law 12-14 15 16 17 18

1908 0.28 0.45 0.13 0.15 - - 1918 - 0.19 0.15 0.66 - -

1928 - 0.13 0.04 0.62 0.11 0.11 1945 - 0.06 0.02 0.70 0.09 0.13 1958 0.04 - 0.02 0.74 0.11 0.09 1965 0.04 - - 0.72 0.15 0.09 Average 0.06 0.14 0.06 0.60 0.07 0.07

Source: Lang and Kropp (1986)

Nicole

Text Box

Percentage of states with laws


TABLE II

DETERMINANTS OF THE LOG ODDS ENROLLMENT RATE

16-Year-olds 17-Year-olds

CAL12/14 0.126 0.028 (0.124) (0.115)

CAL15 0.272 0.168 (0.149) (0.139)

CAL16 0.258 0.143 (0.123) (0.115)

CAL17/18 0.497 - (0.139)

CAL17 0.335 (0.131)

CAL18 - 0.308 (0.131)

1920 -0.093 -0.078 (0.097) (0.091)

1930 0.423 0.399 (0.102) (0.095)

1950 1.144 1.183 (0.105) (0.098)

1960 1.587 1.660 (0.105) (0.098)

1970 2.113 2.254 (0.106) (0.099)

Constant 0.041 - 0.553 (0.103) (0.096)

Standard errors are in parentheses.

dren must remain in school increases.4 In fact, the point estimates for the effect of CAL16 are less than those for CAL15 in both equations. However, in both cases the 95 percent confidence in- terval for the effect of CAL16 minus the effect of CAL15 contains a large range of positive numbers. The results are therefore consistent with the sorting hypothesis.

VI. SUMMARY AND CONCLUSIONS

This paper showed that if sorting gives rise to a separating equilibrium and there is a continuum of ability classes, a com-

4. Note that whether a law requiring attendance until age fourteen has a bigger effect on enrollment rates for sixteen-year-olds or seventeen-year-olds de- pends on the distribution of school leaving ages in the absence of the law and cannot be determined a priori.

Source: Lang and Kropp (1986)

Nicole

Rectangle

Nicole

Rectangle


Lang and Kropp (1986) use a SUR with time and state fixed effects and compare the effects of CAL on the Treated Group (16 year olds, affected by the law) vs. the Control Group ( 17-18 year olds, unaffected by the law).

They find that all of the crucial parameters (CAL12/14 and CAL15 for sixteen-year-olds and CAL12/14, CAL15, and CAL16 for seventeen-year-olds) are positive. o Individually all the critical parameters except CAL15 for sixteen-year-olds are

insignificant, their joint distribution may be unlikely. But forcing people to go to school longer could teach them that the benefits from

higher level schooling

This test is pretty weak because they are testing second order predictions of the signaling model without testing if there is an actual value for a "signal" which is the first order prediction of the model.


C) Using College Proximity as a constraint to college enrollment Bedard (2001) suggested that low ability people reduce education when a

constraint on the high ability people is relaxed,

Under the educational sorting hypothesis, increasing university access, by expanding the university system and thereby lowering the cost of postsecondary education, may increase the high school dropout rate.

Human Capital => Greater access raises university enrolment rates an ambiguous change in the mean skill level for university enrollees, that is,

access has no impact on the university skill mean if the probability of constraint is independent of ability.

decrease in the mean skill level of high school graduates no change for high school dropouts


Signaling => Higher university enrolment rates and higher high-school drop-out rates.

Some previously constrained but relatively high-ability students leave the high-school graduate (HSG) group to become university enrollees o the skill pool of HSG is reduced o the signaling value of HSG is reduced and thus the incentive to obtain the

degree is diminished o higher high school drop out rates (HSD) and skill pool of HSD increase.

College proximity would imply higher college participation and higher drop-out

rates, this generates two testable implications 1) higher high school dropout rate in regions that contain a university while the

human capital model predicts no difference 2) signaling model predicts a higher skill pool among dropouts in regions with a

university and the human capital model does not.

Empirical Strategy:

a

f(O)

Oh Ou 0

b

f(O)

Oh Oh ' Ol ,, 0

FIG. 1.-Uniformly distributed ability: a, before constraints are eased; b, after constraints are eased.

..1.

-

U ,U

Source: Bedard (2001)

Nicole

Text Box

constrained

Nicole

Line

Nicole

Line


o Uses data from the National Longitudinal Survey of Young Men (NLSYM) and Young Women (NLSYW) for men to investigate the role that university access plays in schooling decisions.

o Uses variation in presence of local university across regions in 1966-1968.

Variables of interest: o Access to university recorded in base year (1966 for men, 1968 for women) o Schooling decision of group 14-19 years old (in 1966-1968) o Measure of ability: Knowledge of the World (KWW) and IQ tests

Identification Strategy:

o The variation in university access across regions is exogenous when controlled for family characteristics and individual elements.

o Controls for city/suburb, individual characteristics, family characteristics and local labour market characteristics.

Findings

o Areas with higher university access have o higher postsecondary participation: 10-15 percent higher o higher high school drop out rates: 4-31 percent higher

JOURNAL OF POLITICAL ECONOMY

TABLE 3 PREDICTED EDUCATIONAL GROUP SIZES

MEN WOMEN

Regions Regions without Regions with without Regions with

EDUCATION GROUP Access Access Access Access

Access Is Defined as a Public 2- or 4-Year Degree-Granting Institution

High school dropouts 18.5** 19.8 15.8*** 18.5 High school graduates 34.2 28.3 51.5 45.0 University enrollees 47.3 51.9 32.7 36.5

Access Is Defined as a Public 4-Year Degree-Granting Institution

High school dropouts 18.9** 19.7 17.0*** 17.8 High school graduates 33.8 27.5 50.3 44.9 University enrollees 47.3 52.8 32.7 37.3

Access Is Defined as a 2- or 4-Year Degree-Granting Institution

High school dropouts 18.0 19.6 14.0*** 18.4 High school graduates 36.5 29.2 54.6 45.7 University enrollees 45.5 51.2 31.4 35.9

Access Is Defined as a 4-Year Degree-Granting Institution

High school dropouts 18.6* 19.6 17.2 17.5 High school graduates 35.9 28.3 51.6 45.8 University enrollees 45.5 52.1 31.2 36.7

* The access measure in the dropout/graduate cut point is statistically significant at the 10 percent level. ** The access measure in the dropout/graduate cut point is statistically significant at the 5 percent level. *** The access measure in the dropout/graduate cut point is statistically significant at the 1 percent level.

side in areas with a university, these estimates understate the increase of high school dropouts associated with university access.

The impact and statistical significance of access on the dropout/ graduate cut point differ across access measures for men and women. This likely reflects differences in program/degree preferences between men and women during the late 1960s and early 1970s. Training for

"good" female jobs, such as nursing, teaching, and more technical office

jobs, was more likely available at two-year colleges. It is not, therefore, surprising that the female estimates are more sensitive to the definition of access. Further notice that the estimates are more precise when access is defined as a local public university. This is exactly as one would expect. It is more likely that constrained individuals can gain access to a local

public institution than to a private university. One might also wish to control for ability. Adding the KWW score to

the independent variable list does not substantially alter any of the results, and they are therefore not reported. The statistical significance

766

Source: Bedard (2001)

Nicole

Rectangle

Nicole

Rectangle

Nicole

Rectangle

Nicole

Rectangle


o Increased university access might increase education dispersion and result in lower earning power for the less able.

Issues

o Ranges of effects is quite large o Exogeneity of university access not as compelling as other source of

variation as despite controls for parents's SES

D)Variation in GED passing standards across U.S. states

What is a GED? The General Educational Development certificate (GED) is an High School equivalent credential in the United States, about 10% of High School Diploma holders have a GED.

The GED exam is in five parts (writing, reading, social studies, science, and mathematics) and the average person spends 20 hours studying for it.

Why can’t we simply compare wages of GED versus non-GED holders to measure the signaling effect of the GED?

Self-selection (endogenous choice):


o GED holders probably would have earned less than HS Diploma holders regardless. These are not typically the cream of your HS class.

o GED holders probably would have earned more than other HS dropouts regardless.

o Relative to other dropouts, GED holders have: More years of schooling prior to dropout. Higher measured levels of cognitive skills. Their parents have more education.

Simple comparisons of earnings among dropouts/ GED holders/ HS diploma

holders tell us nothing about the causal effect of a GED for a person who obtains it.

Tyler, Murnane, Willett (2000) [TMW] use the variation in GED passing standards by U.S. states to generate random variation o Some test takers who would receive a GED in Texas with a passing score of

40 − 44 would not receive a GED in New York, Florida, Oregon or Connecticut with the identical scores.


o If GED score is a good measure of a person’s ability/productivity, then people with same ‘ability’ (40 − 44) are assigned a GED in Texas but not in New York.

This seems very close to a signal which has no impact on human capital

production and is as if randomly assigned to people with otherwise identical expected productivity.

Human Capital => If the score in the exam is an unbiased measure of human capital, and there is no signaling, these two individuals should get the same wages.

Signaling => If the GED is a signal, and employers do not know where the individual took the GED exam, these two individuals should get different wages.

TMW use 3 of the 7 possible different standards in GED passing standards by

U.S. states to generate random variation. 1) a minimum score of at least 40 or a mean score of at least 45 2) a minimum score of at least 35 and a mean score of at least 45 3) a minimum score of at least 40 and a mean score of at least 45

LABOR MARKET SIGNALING VALUE OF THE GED 439

TABLE IIIGED SCORE GROUPS FORMED BY COMBINING MINIMUM AND MEAN SCORES

(OUTLINED CELLS = VARIATION IN GED-STATUS BY STATE, DARK SHADING = ALL

POSSESS GED, No SHADING = NONE HAVE GED.)

Mean score

Minimum score

20-3435-3940-4445-4647-4849-5051-5253+

<45

Score group 1Score group 2IScore group 31

> = 45

variation in GED status by the state where the GED exams weretaken in 1990. In each of these two scoring ranges individuals instates with the lower passing standard-the "treatment states"have a GED, while individuals in "comparison states" do not, dueto higher passing standards.

Thus, we have these three natural experiments to test thesignaling hypothesis as it pertains to the GED. We denote theseexperiments by the affected score group in each experiment.

-Experiment 4, where variation in GED status by state is inscore group 4, the treatment states are those states that award aGED in score groups 4 and higher, and the comparison states arethose that award a GED in score groups 5 and higher.

-Experiment 3, where variation in GED status by state is inscore group 3, the treatment states are those states that award aGED in score groups 3 and higher, and the comparison states arethose that award a GED in score groups 5 and higher.

-Experiment 3*, where variation in GED status by state isin score group 3, the treatment states are those states that awarda GED in score groups 3 and higher, and the comparison states arethose that award a GED in score groups 4 and higher.

Since the GED passing standards are set by the individualstate education departments, and since these agencies are embedded in the state political system, the interstate variation in passing standards on which we rely is not transparently exogogenous.The fact that GED passing standards in general, and the ones weexploit in particular, have been relatively stable over time,

Source: Tyler, Murname, and Willett (2000)

Nicole

Line

Nicole

Line

Nicole

Text Box

Pass in some states not others


Using these groups they can define 3 distinct, treatment and control group? (see

Table 3)

With a treatment (those who get the GED) and control group (those who do not obtain the GED), a differences-in-differences econometric strategy would estimate

T = E [Y1|η = k] − E [Y0|η = k] , where Y1, Y0 are earnings of GED and non-GED holders respectively, η is ‘ability,’

and k is a constant, which in our application equals 40 to 44. The variable that randomizes assignment of the GED is location: Texas vs. New

York. So, we could potentially estimate the treatment effect as:

푇 = E [Y |η = k, NY ] − E [Y |η = k, TX] .

However, we might be concerned that there is also a direct effect of being in NY vs. TX that operates independently of the GED at any level of ability. For example

E [Y |η, NY ] − E [Y |η, TX] = δ.


In this case, 푇 from our previous equation would estimate T + δ, i.e., the treatment effect plus the location effect.

To surmount this problem, TMW select a control group of GED test takers with

scores just about the cutoff for both groups of states. Hence, the GED treatment works as follows: Low Passing Standard High Passing Standard Low Score (treatment group)

GED NO GED

High Score (control group) GED GED The outcome variable will be earnings for each of these four groups:

Low Passing Standard High Passing Standard Low Score (treatment group)

E [Y |η = k, TX] E [Y |η = k, NY ]

High Score (control group) E [Y |η = k + 5, TX] E [Y |η = k +5, NY ]


Thus, the DID estimate is: E[푇] = E [Y |η = k, N Y ] − E [Y |η = k, T X]

−E [Y |η = k + 5, N Y ] − E [Y |η = k + 5, T X] = T + δ – δ = T The results are in Table 5 where in experiment 3 and 4, the control group are the

comparison states are those that award a GED in score groups 5 and higher (in 3’, only score group 4 and higher) o Large signaling effects for whites, estimated at 20% earnings gain after 5 years.

The study shows that the GED is taken as a positive signal from white men by

employers, but why does it take five years to show up.

It must be that 1. GED holders are on average more productive than non-GED holders. 2. The GED is in some sense more expensive for less productive than more productive workers to obtain. This probably has to do with maturity, intellect, etc. 3. Employers are unable to perfectly distinguish productivity directly and hence use GED status as one signal of expected productivity.


TABLE V DIFFERENCE-IN-DIFFERENCES ESTIMATES OF THE IMPACT OF THE GED ON 1995

EARNINGS OF DROPOUTS WHO TESTED IN 1990 (STANDARD ERRORS ARE IN PARENTHESES.)

Experiment 4 Experiment 3 Experiment 3*

State passing State passing State passing standard is Low-High standard is Low-High standard is Low-High

standard standard standard Low High contrast Low High contrast Low High contrast

Panel A: Whites Test score is

Low 9628 7849 1779 9362 7843 1509 9362 8616 746 (361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 -23 9143 9304 -162 (80) (65) (103) (135) (63) (149) (135) (135) (150)

Difference-in-differences 1473* 1531** 907- for whites (678) (529) (481)

Panel B: Minorities Test score is

Low 6436 8687 -2252 7005 7367 -363 7005 6858 147 (549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 -894 7782 8375 -593 7782 7568 214 (184) (96) (207) (214) (93) (233) (214) (133) (252)

Difference-in-differences -1357 231 -67

for minorities (906) (548) (518)

** = significant at the 0.01 level, * = significant at the 0.05 level, - = significant at the 0.10 level. Experiment 4: Test Score Low: score group = 4; Test Score High score groups = 5-10. Passing Standard Low: 35 minimum score and 45 mean score; Passing Standard High: 40 minimum score

and 45 mean score. Low Passing Standard states: All states except for TX, LA, MS, NE, FL, NY, CA, WA, and CT; High

Passing Standard states: NY and FL. Experiment 3: Test Score Low: score group = 3; Test Score High score groups = 5-10. Passing Standard Low: 40 minimum score or 45 mean score; Passing Standard High: 40 minimum score

and 45 mean score. Low Passing Standard states: TX, LA, MS, and NE; High Passing Standard states: NY and FL. Experiment 3*: Test Score Low: score group = 3; Test Score High: score groups = 5-10. Passing Standard Low: 40 minimum score or 45 mean score; Passing Standard High: 35 minimum score

and 45 mean score. Low Passing Standard states: TX, LA, MS, and NE; High Passing Standard states: all states except TX,

LA, MS, NE, NY, FL, and CT.

The results for nonwhite dropouts differ sharply from the results for white dropouts. The three experiments yield no statistically significant evidence that acquisition of a GED results in higher earnings for minority dropouts. We return to the minority results later. Based on the results from experiments 4, 3, and 3*, our estimates are robust to the use of different treatment and comparison groups.

B. timing of GED Treatment Effects

To avoid underestimating the impact of the GED by measur- ing earnings too close to receipt of the credential, we have

Source: Tyler, Murname, and Willett (2000)

Nicole

Rectangle

Nicole

Rectangle

Nicole

Oval

Nicole

Line

Nicole

Line


Figure 1: Experiment 4 1473* (678)

1600 756

(635) 01100

E 600 -277 -325 -798- 664 82

wU (291) (364) (474) (56 (51

00

E -4001 -2 -1 7: P2l 3 4 5

-900

Year from Test

Figure II: Experiment 3 1578* 53j**

1600 ~~~~~~~~~~~~~~~~(508) (529)

1100 639- (406)I

E 600 o 542** -593*

0 ~~~~~~ ~~~~ ~ 1 2 3 4 5 -400

-900

Year from Test

Figure III: Experiment 3* j497**

1600 -95* (459) 889** 847* (413) 1 907

0 1100 (329) (373) (48)

il p

i9 lX E- 600 WD 58 gn ~

(152) (~11 0

-400 -2 -1 1 2 3 4 5

-900

Year from Test

FIGURES I-III Pretreatment and first through fifth year Difference-in-Differences Estimates for Young White Dropouts. (** = Significant at the 0.01 Level; * = Significant at the

0.05 Level; - = Significant at the 0.10 Level.)

Source: Tyler, Murname and Willett (2000)


But why do the results for minority men differ? o Many minority men take GED while incarcerated; the stigma of incarceration

may depress the post-prison earnings of dropouts, and eliminate any positive signaling value of the GED credential.

o In some social programs, getting a GED is ‘‘quasi-compulsory’’ for those seeking benefits

Other concerns about identification come from those who get GEDs do so to qualify for post-secondary school training programs

o They may actually acquire other skills (not observed)

Individuals may respond to the different standards by migrating from high to low standard states

o Not much evidence of this


Basic readings: Spence, A.M., "Job Market Signaling," Quarterly Journal of Economics, Vol. 87 (August 1973)

355-374. Tyler, John H., Richard J. Murnane and John B. Willett. "Estimating the Labor Market Signaling

Value of the GED" Quarterly Journal of Economics, Vol. 115 (May 2000) 431- 468.

II. Human Capital, Returns to Education and...

Documents

Transcript of II. Human Capital, Returns to Education and...