Working hard for the money? Eciency wages and worker
e?ort
Arthur H. Goldsmith a,*, Jonathan R. Veum b,1, William Darity, Jr. c,2
a Department of Economics, Washington and Lee University, Lexington, VA 24450, USA
b Freddie Mac, 8200 Jones Branch Drive-Mailstop 289, McLean, VA 22102, USA
c Department of Economics, University of North Carolina at Chapel Hill, Chapel Hill, NC 27514, USA
Received 19 September 1998; received in revised form 26 January 2000; accepted 24 May 2000
Abstract
This paper o?ers a test of the relative wage version of the eciency wage hypothesis ± that
®rms are able to improve worker productivity by paying workers a wage premium. Psychologists
believe work e?ort re¯ects motivation that is governed by a feature of personality referred
to as locus of control. Measures of locus of control are available in the National
Longitudinal Survey of Youth, Using data drawn from the NLSY in 1992 we simultaneously
estimate structural real wage and e?ort equations. We ®nd that receiving an eciency wage
enhances a person’s e?ort and that person’s providing greater e?ort earn higher wages.
Ó 2000 Elsevier Science B.V. All rights reserved.
PsycINFO classi®cation: 3000; 3630
JEL classi®cation: E24; J6
Keywords: Locus of control; Employee motivation; Salaries; Employee bene®ts
Journal of Economic Psychology 21 (2000) 351±385
www.elsevier.com/locate/joep
* Corresponding author. Tel.: +1-540-463-8970; fax: +1-540-463-8639.
E-mail address: GoldsmithA@wlu.edu (A.H. Goldsmith).
1 Tel.: +1-703-903-3274; fax: +1-703-903-2814.
2 Tel.: +1-919-966-2156; fax: +1-919-966-4986.
0167-4870/00/$ - see front matter Ó 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 7 - 4 8 7 0 ( 0 0 ) 0 0 0 0 8 - 8
1. Introduction and statement of the problem
This paper o?ers a test of the relative wage version of the eciency wage
hypothesis. This form of the eciency wage hypothesis states that ®rms are
able to improve worker productivity by paying workers a wage premium ± a
wage that is above the wage paid by other ®rms for comparable labor. A link
between wage premiums and productivity might arise for a number of distinct
reasons. A wage premium may enhance productivity by improving
nutrition (Leibenstein, 1957), boosting morale (Solow, 1979), encouraging
greater commitment to ®rm goals (Akerlof, 1982), reducing quits and the
disruption caused by turnover (Stiglitz, 1974), attracting higher quality
workers (Stiglitz, 1996; Weiss, 1980), and inspiring workers to put forth
greater e?ort (Shapiro & Stiglitz, 1984).
Much attention (Krueger & Summers, 1988; Dickens & Katz, 1987) has
focused on whether ®rms pay eciency wages. 3 Another line of inquiry
(Leonard, 1987; Groshen & Krueger, 1990) has explored whether ®rms that
pay wage premiums recoup some of the costs by allocating less resources for
employee supervision. 4 Economists have taken the position that e?ort is not
only imperfectly observed by the employer, but that it also is unobserved for
the investigator or econometrician. Thus, economists have been unable to
examine the impact of wage premiums on e?ort and hence directly test the
eciency wage hypothesis. 5
As a result, Allen (1984) opted to probe indirectly the eciency wage
hypothesis by investigating the impact of wage premiums on an observable,
3 These researchers report that workers with similar skills and job characteristics earn substantially
di?erent wages. The standard competitive labor market model does not provide a straightforward
explanation of the persistence of such di?erentials for comparable labor. They interpret their ®nding as
evidence that ®rms pay eciency wages.
4 Leonard (1987) ®nds no signi®cant evidence of a trade-o? between supervisory intensity and wage
premiums. Groshen and Krueger (1990) report that enhanced supervision leads to lower wages for nurses,
but in three other occupations (e.g. food service employees, radiographers, and physical therapists) pay is
found to be statistically independent of the level of supervision.
5 A rare exception is an unpublished exploratory study of the relation between wage premiums and selfreported
work e?ort conducted by Krueger and Summers (1986) using data from the 1977 Quality of
Employment Survey. They use OLS to estimate an equation where self-reported work e?ort is the
dependent variable; it is a qualitative limited dependent variable ranging from 1±4. The wage premium is
speci®ed to be exogenous and a limited set of control variables is included in their e?ort equation. They
®nd that a greater wage premium has a positive, but statistically insigni®cant, impact on self-reported
work e?ort.
352 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
absenteeism, that is likely to be related to productivity. 6 An alternative tactic
has been to concentrate on testing the predictions of the labor turnover
(Campbell, 1993; Leonard, 1987; Krueger & Summers, 1988) and shirking
(Cappelli & Chauvin, 1991) versions of the eciency wage theory, since quit
behavior is readily observed and data are available on disciplinary dismissals
± a potential measure of shirking.
Psychologists believe work e?ort re¯ects motivation, which is governed by
a feature of personality, referred to as locus of control. In their view, locus of
control can be detected by employers, can be measured by investigators, and
can be used as a measure of e?ort. Measures of locus of control are construed
by psychologists as an index of e?ort and are available in the National
Longitudinal Survey of Youth (NLSY).
In this paper, we use data drawn from the NLSY to advance two
questions germane to the eciency wage literature that economists have
yet to explore. First, are workers who receive an eciency wage likely to
exhibit greater e?ort? Second, are wages enhanced by improved e?ort? A
real wage equation is estimated to identify the contribution of e?ort to
hourly compensation. We estimate an individual e?ort equation also to
determine if earning an eciency wage, and other factors that a?ect the
perceived cost of job loss, in¯uence e?ort. We introduce a new method of
measuring a person’s eciency wage into the eciency wage literature. In
our empirical work a person earns an eciency wage when they earn
more than they expect to earn given their personal characteristics rather
than earning more than a typical worker in their industry or occupation
does.
This paper is organized as follows: In Section 2 we present a brief review of
the relative wage version of the eciency wage hypothesis. The model guides
our subsequent empirical work. In Section 3 we discuss the literature from
the ®eld of psychology that advances a relationship between personality and
e?ort. Also, based on this knowledge, we describe and evaluate the measurement
of e?ort. Section 4 contains a description of our empirical procedures,
including data, model speci®cation, and estimation technique. An
alternative paradigm for explaining the relation between economic outcomes
and e?ort, based on stress process theory, is discussed in this section. We
6 In a similar line of inquiry, Hamermesh (1977) found that high wages enhance job satisfaction ± which
he believes is measurable ± that, in turn may promote productivity.
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 353
present estimates of the impact on e?ort of receiving an eciency wage,
unemployment and other factors that in¯uence the perceived cost of job loss
in Section 5. This section also contains our estimates of the determinants of
wages, including the contribution of e?ort. The implications of our ®ndings
and concluding remarks appear in Section 6.
2. Monitoring, e?ort, and wages
The basic tenet of eciency wage theory is that e?ort, e, depends on
compensation. Shapiro and Stiglitz (1984), founders of the shirking variant
of the ef®ciency wage theory, contend that effort must be elicited from
workers through either external monitoring, ME, or internal monitoring, MI.
External monitoring occurs when ®rms utilize supervisors and equipment to
oversee work effort. Shapiro and Stiglitz (1984) and Krueger and Summers
(1988) claim that external monitoring is costly and impractical in some industries
and occupations. Due to technology and the manner in which work
is organized it may be dif®cult to observe an individual employee’s contribution
to output. Under these conditions, how can employers elicit greater
effort from their employees?
The fundamental insight in shirking models is that more e?ort can be
obtained by providing incentives for workers to “internally monitor” (or selfmonitor).
Workers self-monitor when they view their job as relatively attractive.
Therefore, workers receiving a wage, w, above what they could
command if employed elsewhere, w, or those earning a wage premium,
?w ÿ w > 0?, are expected to internally monitor.
The extent to which workers self-monitor is a?ected by factors that in-
¯uence the perceived cost of job loss besides the wage rate. These factors
include items such as the odds of being exposed to job loss, availability and
generosity of unemployment insurance, transferability of skills, household
wealth, earnings of other family members, and the perceived psychological
e?ect of exposure to joblessness. In addition, early childhood socialization
establishes attitudes toward work intensity and self-governance, which in-
¯uence the propensity to self-monitor.
In work environments where it is easy for managers to observe and evaluate
workers, external monitoring is accurate and cost e?ective. When external
monitoring is dicult there is an incentive for management to establish
policies that foster internal monitoring.
354 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
3. Personality and e?ort
Individuals who exert higher levels of e?ort on the job are expected to
exhibit greater productivity. Psychologists treat e?ort as the response to an
underlying motivation. Thus, theories of motivation can be viewed as theories
of e?ort. Economists (Kim & Polachek, 1994) recognize that motivation
di?ers across individuals and is likely to in¯uence their productivity. How
else can we explain the hardworking individual with modest skills who
consistently outperforms other more gifted persons? The motivated are
generally characterized as contributing an abnormally strong commitment to
the tasks they face.
The founders of motivational theory (Atkinson, 1964; Vroom, 1964) hypothesized
that motivation depends upon motives and expectancies. Motives
are best thought of as an orientation, disposition, or taste to seek or to avoid
various behaviors. Psychologists believe motives are established early in life
and remain stable over the life cycle (Atkinson, 1964, p. 242). Expectancies
entail an individual’s assessment of the likelihood that their actions will result
in attainment of a desired outcome. Bandura (1986), the founder of social
learning theory, refers to a person’s expectancy in a speci®c domain as selfef
®cacy. According to Bandura (1986) motivation to initiate action is governed
by motives, which are time-invariant, and self-ef®cacy that responds to
salient events including labor market outcomes such as unemployment. 7
Economists Summers (1988), Shapiro and Stiglitz (1984), and Yellen (1984)
have argued that compensation and “fear of unemployment” induce motivation
at the workplace.
Currently, among psychologists, expectancy theory is the most widely
accepted and empirically supported theory of motivation (Robbins, 1993;
Muchinsky, 1977). Expectancy theory has its roots in the motivation theory
developed by Atkinson (1964) and Vroom (1964). According to expectancy
theory, the strength of a person’s motivation depends on the extent to which
they believe that “exertion, performance, and reward” are linked tightly. 8
7 Psychologists also have asserted that motivation depends upon satisfaction of needs (Maslow, 1954),
goal-setting (Locke, 1968), and equity (Adams, 1965).
8 This theory posits that a person’s motivation is directly related to their belief that: (1) e?ort will lead to
performance ± like achievement of the attempted task; (2) performance will be rewarded by compensation,
opportunity to use skills, security, and the chance to develop professional relations; and (3) the rewards
contribute to the realization of individual goals ± like self-respect, status, recognition, friendship, and
security.
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 355
Attribution theorists (Heider, 1958; Rotter, 1966) have proposed that an
aspect of personality ± locus of control ± governs a person’s perception of the
relation between exertion, performance, and reward.
Rotter (1966) classi®ed individuals who believe they are masters of their
own fates, and hence bear personal responsibility for what happens to them,
as “internalizers”. Internalizers see control of their lives as coming from
within themselves. On the other hand, many people believe that they are
pawns of fate, that they are controlled by outside forces over which they have
little, if any, in¯uence. Such people feel that their locus of personal control is
external rather than internal, and they bear little or no responsibility for what
happens to them. Rotter referred to the latter group as “externalizers”.
Expectancy theory predicts that a person with a more internal locus of
control will be more motivated than a comparable individual whose locus of
control is external because internalizers see themselves as “in-control”, i.e.
able to produce desired outcomes (Skinner, Chapman & Baltes, 1988).
Skinner (1995, pp. 69,70) asserts that the primary psychological mechanism
by which perceived control in¯uences outcomes is through its e?ects on action
or motivation. 9 According to Bandura (1989, p. 1176) a person’s beliefs
about their capabilities to exercise control over events ± locus of control ±
“determines their level of motivation, as re¯ected in how much e?ort they
will exert in an endeavor and how long they will persevere in the face of
obstacles”.
Dunifon and Duncan (1998, p. 34) claim that
Because of the importance attached to motivation by personality psychologists
motivational measures were included in both the National
Longitudinal Surveys (NLS) and the National Longitudinal Survey of
Youth (NLSY) labor-market panels, and the early waves of the Panel
Study of Income Dynamics (PSID). For the original NLS cohorts all
23 items from Rotter (1966) `locus of control’ scale were included as a
measure of expectancy; in the NLSY, a four-question subset of these
was included. . . The PSID-based expectancy items are essentially equivalent
to this subset of Rotter’s scale. . .
9 Skinner (1996) found that researchers use a large number of terms to describe control. Some constructs
include the term “control” in their name ± locus of control, personal control, sense of control ± while
others do not explicitly use the term ± self-ecacy, mastery, helplessness ± but nevertheless are closely
related. Thus, terms like locus of control and self-ecacy are comparable and used interchangeably.
356 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
Thus, the designers of the NLS and PSID anticipated that measures of
expectancy would be used as indexes of motivation or e?ort. A number of
investigators, including Goldsmith, Veum and Darity (1999), Duncan and
Dunifon (1998), Dunifon and Duncan (1998) and Hill et al. (1985) have
adopted this means of measuring motivation. 10
Direct evidence that locus of control in¯uences motivation comes from a
number of sources. Studies by Harter (1978) and Kuhl (1981) reveal that
when perceived control is high, a person tends to embrace challenges, construct
more e?ective action plans and exert more sustained e?ort in their
enactment. Heckhausen (1991) and Kuhl (1984) reach a similar conclusion.
They ®nd that people with high control are better able to concentrate completely
on tasks, enhancing access to their working memory and boosting
their persistence in the face of obstacles. Bandura and Cervone (1983) found
individuals with a stronger belief that they are in control exert greater e?ort
to master a challenge and are more persistent in their e?orts. In addition,
when actions do not initially succeed, people with high control are more
likely to increase their e?ort exertion and continue to try to achieve their goal
(Bandura, 1989; Dweck, 1990; Jacobs, Prentice-Dunn & Rogers, 1977; Baum,
Fleming & Reddy, 1986). These ®ndings corroborate the earlier ®ndings of
Seligman (1975) that repeated exposure to uncontrollable events, leading to
feelings of helplessness and an external outlook, reduces motivation to engage
in goal-directed behavior. 11
Bandura (1989) and Dweck (1990) believe that persons with a greater sense
of control are more productive because they exhibit a pattern of more effective
strategy selection, hypothesis testing, problem-solving, and general
analytic thinking. In summarizing her review of the literature on the relationship
between locus of control and action, Skinner (1996, p. 556) stated
“when people perceive that they have a high degree of control, they exert
10 Psychologists Skinner et al. (1988) assert that perceived control depends on three conceptually
independent sets of beliefs; control beliefs, expectancies about the extent to which a person can obtains
desired outcomes, means±ends beliefs, expectations about what factors produce outcomes; and agency
beliefs, opinions about the possession of various means. They provide evidence that effort is most closely
associated with means±ends beliefs. However, they also report a positive and statistically signi®cant
relation between effort and control beliefs. Thus, using control beliefs as a proxy for motivation is viable.
11 Maier and Seligman (1976) argue that once events or socialization lead an individual to hold a
particular locus of control or e?ort level, their view of the link between action and outcome, and hence
motivation, is transferred to all other situations they encounter. Thus, if a person ®nds that attempts to
succeed in school or to succeed socially are unsuccessful, they are not only likely to become apathetic
students and seek the company of others less often, but also would be less motivated workers.
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 357
e?ort, try hard, initiate action, and persist in the face of failures and setbacks;
they evince interest, optimism, sustained attention, problem solving, and an
action orientation”. In short, persons with a more internal locus of control
are both more motivated and productive.
Psychologists have designed and validated survey instruments capable of
measuring locus of control, and hence, motivation or e?ort. This makes it
possible for economists to explore the reciprocal in¯uences of real wages and
e?ort. The following section discusses the empirical procedures we adopt to
perform such an examination.
4. Empirical procedures
4.1. Data
The data used in this study is from the NLSY. The NLSY is a sample of
12,686 males and females who were between the ages of 14 and 22 in 1979
and who have been interviewed annually since then. The NLSY is a data set
rich in economic and demographic information, including data on wages and
multiple aspects of human capital. It also contains information on motivation.
Motivation or e?ort is expected to depend upon motives and self-ef®cacy.
Motives, a disposition to pursue or evade various behaviors, are established
early in life remain stable and are heavily in¯uenced by socialization. Selfef
®cacy is a variable feature of personality that is likely to respond to salient
experiences, such as occurrences in the labor market. 12 Therefore, holding
motives constant, ¯uctuations in effort can be attributed to variations in selfef
®cacy.
Families and signi®cant others socialize youths and are thereby largely
responsible for the establishment of a person’s motives early in life. The
NLSY contains information describing a person’s adolescent home environment,
which can be used to represent their motives.
The Mastery Scale was developed by Pearlin, Lieberman, Menaghan, and
Mullan (1981) to measure a person’s locus of control or self-ef®cacy. The
NLSY contains each person’s score on the Mastery Scale in 1992. Mastery
12 Gorman (1968), McArthur (1970), and Smith (1970) o?er evidence that contemporary events
in¯uence individuals perceptions of causality and hence control.
358 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
Scale scores range in value from 0 to 7 (an internal response to each question).
Individuals with a high score ± those with a more internal locus of
control ± are expected to be more motivated than a comparable persons with
lower scores on the Mastery Scale. 13
If Mastery Scale scores are used to measure motivation, because they
gauge self-ecacy, then Pearlin et al.’s (1981) Stress Process Theory, like the
economists eciency wage theory, predicts a direct relation between work
place e?ort and unexpected wages. However, Pearlin’s explanation is
grounded in psychological theory rather than a conjecture about how individuals
respond to economic incentives such as the cost of job loss. Stress
Process Theory links life event with stress and stress with self-ecacy, and
hence, motivation.
Following the seminal work of Cannon (1935) and Selye (1956), Pearlin
et al. (1981) argue that humans are fundamentally intolerant of change. In
their view salient life events either foster or curtail stress. They believe stresses
directly alter aspects of self-concept including “mastery” or self-ecacy.
Thus, earning an eciency wage provides a person with concrete evidence of
their success and proof they are able to alter circumstances of their lives, both
of which reduce life strains and contribute to mastery. Disappointing life
events such as bouts of unemployment would provoke erosion of self-ecacy
and motivation.
Social support and coping behavior are expected to in¯uence the amount
of stress that people experience. Pearlin et al. (1981) believe these elements
are important components of the stress process and in¯uence the motivation
level people exhibit. Pearlin and his colleagues claim, and o?er evidence, that
13 Many economists are sceptical that psychological constructs such as locus of control can be measured
accurately by scales constructed from self-reported evaluations collected in the form of responses to survey
questions. Psychologists assess the usefulness of scales developed to measure a psychological construct
such as locus of control by examining three features of the scale: convergent validity, reliability, and
stability. Convergent validity is concerned with whether an alternative scale seeking to measure the same
construct yields a similar assessment. A scale is reliable when the questions that comprise the scale are all
probing similar or related features of the individual’s make-up. A scale is only considered stable if
administering the same scale a short time in the future generates a similar assessment. Pearlin et al. (1981)
found the Mastery Scale correlated well with other scales used to measure to locus of control. In addition
to meeting the criteria for convergent validity, they discovered the scale was internally consistent, and
stable over time, For a detailed discussion of Mastery Scale Validity, see Seeman (1991, pp. 304±306).
Economists also have an aversion to making inter-personal comparisons using self-reported evaluations
(Easterlin, 1974). For a detailed discussion of both the measurement and comparison issues raised by
economists, and the procedures adopted by psychologists that address these concerns, see Darity and
Goldsmith (1996) and Goldsmith, Veum and Darity (1996a).
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 359
emotional support characterized by “qualities of trust and intimacy. . .commonly
properties of marital relation”, reduce life strains and thereby contribute
to self-ecacy.
Coping behaviors also are likely to alter the stress levels people experience.
Coping may entail modi®cation of a stressful situation, altering the meaning
associated with undesirable life events, and management of stress symptoms.
People often seek assistance from family members, friends, professional
councillors, and clergy in developing and applying coping skills and strategies.
The NLSY provides a means of measuring motives and self-ef®cacy as well
as social support and coping. Moreover, information on labor market outcomes
and demographic factors are available in the NLSY. Thus, the NLSY
is an ideal data set for an investigation of the relation between effort and
unanticipated wages, which economists refer to as the shirking version of the
ef®ciency wage hypothesis.
4.2. Model speci®cation and hypotheses
Following the convention initiated by Mincer (1962), the productivity, and
hence wage, of a worker is expected to depend on their personal attributes,
such as skills and e?ort, as well as the characteristics of their workplace.
According to the eciency wage hypothesis, a worker’s e?ort depends upon
both external monitoring ± the extent of direct supervision ± and internal
monitoring. Internal monitoring re¯ects early childhood socialization and the
perceived costs of job loss, including the wage a person receives relative to
their expected wage. Therefore, both wages and e?ort should be viewed as
endogenous and determined simultaneously. In order to account for the joint
determination of wages and e?ort, and to allow for the impact of life events
on stress and e?ort, the following two equation structural model is speci®ed:
EFFORTi ? /?WAGEi ÿ EXPECTED WAGEi? ? ?Ci?w
? ?Si?k ? ?Ai?d ? li; ?4:1?
WAGEi ? a?EFFORTi? ? ?Hi?b ? ?Xi?c ? ei: ?4:2?
Table 1
Variable names, de®nitions, means, and (standard deviations): Wage and e?ort equations
Variable name Variable de®nition All Male Female White Black Hispanic
WAGE Natural log of hourly wage in 1992 2.17
(0.57)
2.26
(0.56)
2.06
(0.57)
2.25
(0.58)
2.01
(0.53)
2.16
(0.57)
EFFORT Sum of the response to the seven Pearlin
questions used to measure locus of control
6.11
(1.23)
6.14
(1.22)
6.07
(1.25)
6.21
(1.16)
5.95
(1.31)
6.04
(1.29)
EDUCATION Years of education completed at 1992
interview date
13.12
(2.38)
12.92
(2.44)
13.35
(2.30)
13.50
(2.41)
12.86
(2.07)
12.41
(2.49)
EXPERIENCE Weeks of work experience at 1992 interview
date
529
(159)
548
(146)
507
(170)
558
(148)
475
(166)
521
(162)
TENURE Weeks with current employer at 1992
interview date
191
(177)
193
(179)
189
(174)
203
(180)
172
(168)
186
(177)
JOB TRAINING 1 if received company training from 1992
employer since 1991 interview date,
0 otherwise
0.90eÿ01
(0.29)
0.86eÿ01
(0.28)
0.95eÿ01
(0.29)
0.95eÿ01
(0.29)
0.82eÿ01
(0.27)
0.89eÿ01
(0.28)
AFQT Percentile score on the Armed Forces
Qualifying Test
42.08
(28.61)
41.69
(29.94)
42.51
(27.05)
54.22
(26.88)
24.68
(22.03)
32.30
(24.90)
AGE Age 30.79
(2.23)
30.70
(2.22)
30.90
(2.23)
30.81
(2.23)
30.81
(2.23)
30.73
(2.22)
UNEMPLOYMENT Local unemployment rate 0.13
(0.34)
0.13
(0.33)
0.13
(0.34)
0.13
(0.33)
0.43
(0.20)
0.26
(0.44)
UI BENEFITS Average weekly unemployment insurance
bene®t in state of residence in 1992 dollars
165
(26)
165
(27)
165
(27)
169
(27)
161
(27)
160
(22)
SMSA 1 if live in an SMSA, 0 otherwise 0.75
(0.43)
0.74
(0.44)
0.75
(0.43)
0.70
(0.46)
0.79
(0.41)
0.83
(0.37)
UNEMPLOYMENT
BOUTS
Number of spells of unemployment since
January 1, 1978
4.92
(3.73)
5.24
(4.02)
4.56
(3.33)
4.41
(3.62)
5.91
(3.67)
4.95
(3.83)
UNEMPLOYMENT
DURATION
Duration of longest unemployment spell
since January 1, 1978
22.50
(24.01)
24.24
(25.15)
20.59
(22.53)
17.99
(20.24)
32.33
(29.21)
21.36
(21.47)
MARRIED 1 if married, 0 otherwise 0.55
(0.50)
0.54
(0.50)
0.56
(0.50)
0.63
(0.48)
0.37
(0.48)
0.57
(0.50)
SPOUSE EARNINGS Earnings of spouse in 1992 dollars,
0 if single
11030
(18324)
6877
(11590)
15634
(22779)
13657
(19679)
6469
(15730)
10054
(16289)
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 361
Table 1 (Continued)
Variable name Variable de®nition All Male Female White Black Hispanic
CHILDREN Number of children in household 1.10
(1.20)
0.91
(1.16)
1.30
(1.20)
1.01
(1.11)
1.13
(1.26)
1.32
(1.29)
PART-TIME 1 if usually work less than 30 hours
per week, 0 otherwise
0.10
(0.30)
0.50eÿ01
(0.22)
0.17
(0.37)
0.11
(0.31)
0.93eÿ01
(0.29)
0.94eÿ01
(0.29)
ASSETS total value of ®nancial assets in 1992 12751
(37975)
12257
(35172)
13298
(40858)
17218
(48925)
5810
(12791)
9923
(20554)
MALE 1 if male, 0 otherwise 0.53
(0.50)
0.52
(0.50)
0.52
(0.50)
0.54
(0.50)
BLACK 1 if black, 0 otherwise 0.27
(0.44)
0.27
(0.44)
0.27
(0.44)
HISPANIC 1 if Hispanic, 0 otherwise 0.19
(0.39)
0.19
(0.39)
0.19
(0.39)
PROFESSIONAL
PARENT
1 if occupation of either parents was
professional or manager when respondent
was 14, 0 otherwise
0.25
(0.43)
0.24
(0.43)
0.26
(0.44)
0.35
(0.48)
0.14
(0.34)
0.14
(0.35)
BOTH PARENTS 1 if both parents lived in household when
respondent was 14, 0 otherwise
0.78
(0.42)
0.78
(0.42)
0.78
(0.41)
0.88
(0.32)
0.58
(0.49)
0.77
(0.42)
PARENT
EDUCATION
Average highest grade completed by
respondent’s parents
10.86
(3.06)
10.88
(3.09)
10.85
(3.02)
11.91
(2.52)
10.39
(2.40)
8.54
(3.75)
RELIGION 1 if aliated with any religious group,
0 otherwise
0.96
(0.20)
0.95
(0.21)
0.96
(0.18)
0.96
(0.20)
0.95
(0.22)
0.98
(0.14)
362 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
ESTABLISHMENT
SIZE
Number of employees at establishment 538
(2240)
491
(2130)
591
(2356)
543
(2290)
589
(2357)
454
(1900)
MULTIPLE
LOCATIONS
1 if company has employees at another
location, 0 otherwise
0.63
(0.48)
0.61
(0.49)
0.65
(0.48)
0.62
(0.49)
0.64
(0.48)
0.65
(0.48)
LARGE MULTIPLE
LOCATIONS
1 if employer has 1000 or more employees
at other locations
0.37
(0.48)
0.35
(0.48)
0.39
(0.49)
0.34
(0.47)
0.43
(0.50)
0.39
(0.49)
UNION 1 if member of a union, 0 otherwise 0.14
(0.34)
0.16
(0.36)
0.12
(0.32)
0.12
(0.32)
0.16
(0.37)
0.16
(0.37)
NORTHEAST 1 if lived in Northeast region, 0 otherwise 0.16
(0.37)
0.17
(0.37)
0.16
(0.37)
0.19
(0.39)
0.14
(0.35)
0.14
(0.34)
NORTH-CENTRAL 1 if lived in North Central region,
0 otherwise
0.23
(0.42)
0.24
(0.43)
0.22
(0.42)
0.32
(0.47)
0.17
(0.37)
0.73eÿ01
(0.26)
WEST 1 if lived in Western region, 0 otherwise 0.21
(0.41)
0.21
(0.41)
0.20
(0.40)
0.18
(0.38)
0.79eÿ01
(0.27)
0.47
(0.50)
IMILLS Selection correction term 0.20
(0.25)
0.15
(0.17)
0.26
(0.30)
0.17
(0.22)
0.25
(0.27)
0.21
(0.28)
n Number of observations 5579 2933 2646 3013 1509 1057
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 363
4.2.1. E?ort equation
A person’s level of EFFORTi, the dependent variable in Eq. (4.1), is
measured by their 1992 score on the “Mastery Scale” ± a gauge of self-ef®-
cacy ± since measures of an individual’s motives are included as explanatory
variables in the effort equation. It is interesting to note that Mastery Scale
scores are surprisingly high with 49% of the sample providing self-reports
placing them in the highest motivation category. However, there is substantial
variability in the remaining responses with 44% of all scores ranging
between 4 and 6.
The vector Si contains a cluster of variables describing an individual’s
adolescent home environment of age 14 to account for the in¯uence of socialization
on the formation of motives. Measures of PARENT EDUCATION,
whether a PROFESSIONAL PARENT resides in the home, and the
presence of BOTH PARENTS are included in Si.
Self-ef®cacy, later in life, is likely to be enhanced by an adolescence where
BOTH PARENTS are present, a PROFESSIONAL PARENT resides in the
home, and PARENT EDUCATION is greater. Thus, including Si as an
explanatory variable in the effort equation serves two purposes; it captures
the contribution of motives to subsequent self-ef®cacy, and accounts for the
“trait-like” component of motivation. Thus, holding constant a person’s
motives, ¯uctuations in self-ef®cacy correspond with movements in effort.
The frequency distribution for Mastery Scale scores in 1992 is presented in
Table 2.
Table 2
Frequency distribution: E?ort scalea
Mastery scale
Score Frequency Percent
0 10 0
1 53 1
2 144 2
3 263 4
4 462 7
5 859 13
6 1635 24
7 3321 49
n 6747 100
a E?ort, e, is measured by a person’s score on the Pearlin et al. (1992) Mastery Scale. The distribution
presented is for all persons in the sample in 1992 whether or not they were working ± the sample used to
estimate the reduced form effort and wage equations. The distribution is similar to the distribution for
those who were employed at the time of the 1992 survey.
364 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
In our view a worker receives an eciency wage when they are earning a
WAGEi greater than the wage they expect to earn, EXPECTED WAGEi. In
prior studies (Leonard, 1987; Krueger & Summers, 1988) the wage premium
expected to induce greater e?ort is measured by the di?erence between what
an individual earns and the average wage in their occupation. However, an
individual is likely to believe they are earning a wage premium only when
they earn more than what they expect to earn based upon their personal
characteristics ± which may di?er from those of the average person in their
occupation. A person earning an eciency wage would ®nd job loss to be
especially costly. Thus, individuals who receive an EFFICIENCY WAGE,
?WAGEi ÿ EXPECTED WAGEi? > 0, are expected to monitor internally to
a greater extent and to o?er their employer greater e?ort.
The vector Ci is composed of the remaining factors that are likely to
determine the perceived cost of job loss. Workers may fear long, and hence
costly, bouts of unemployment. Thus, a rise in the local UNEMPLOYMENT
rate, which portends longer spells for those who become jobless, will
prompt greater e?ort to reduce the likelihood of discharge for inadequate
performance. In contrast, the bigger the local labor market the easier it is to
®nd a desirable job. Thus, we might expect that individuals who live in a
larger SMSA will be inclined to provide less e?ort on the job. Residents of
states with more generous unemployment insurance, greater UI BENEFITS,
face a smaller cost of job loss and are presumed to extend less e?ort
at work.
Unemployment generates ®nancial and psychological hardships (Goldsmith,
Veum & Darity, 1996b). These consequences of unemployment are
likely to be more vivid or salient for persons who in the past have been exposed
to UNEMPLOYMENT BOUTS more often and have experienced
greater UNEMPLOYMENT DURATION. Therefore, greater personal exposure
to joblessness may enlarge the perceived costs of unemployment
leading to more e?ort in an attempt to prevent experiencing unemployment
again. Alternatively, individual’s with unemployment in their past may become
helpless and fatalistic, believing that the likelihood of experiencing
unemployment in the future is independent of their current level of e?ort on
the job. If this were the case, workers with more and longer bouts of unemployment
in their past may choose to give less e?ort than comparable
employees with better labor market histories. Hence, the impact of prior
unemployment on current e?ort levels is ambiguous.
People who have accumulated more transferable human capital are
likely to be less fearful of unemployment and, therefore, more prone to
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 365
o?er their employers less e?ort. It is possible also that workers with more
human capital secure jobs they enjoy and are attached to leading them to
o?er their employers greater e?ort. Measures capturing these di?erent aspects
of general human capital are contained in the vector Ci. Broad-based
formal skills are captured by EDUCATION. An individual’s verbal and
mathematical skills developed while attending school and at home are
measured by scores on the Armed Forces Qualifying Exam, AFQT (see
Fischer et al., 1996, pp. 55±69). General workplace skills are represented
by EXPERIENCE.
Job loss is costly for workers who possess non-transferable or ®rm speci®c
skills, leading those with non-transferable skills to give greater e?ort on the
job to avoid losing the skills they have required. Following Becker (1962),
TENURE and JOB TRAINING, which are included in Ci are often described
as forms of ®rm speci®c human capital. However, TENURE and
formal training received on the job may provide workers with both general
and ®rm-speci®c skills (Neal, 1995). Thus, the impact of longer TENURE
and JOB TRAINING on e?ort is ambiguous, depending on the composition
of the skills acquired.
More mature young workers (those of greater AGE), with a given set of
skills and experiences, are likely to have learned the employer’s minimally
acceptable standard of e?ort. Younger workers who have yet to discover this
level may provide more e?ort, to be viewed as o?ering an adequate level of
job performance. 14
Membership in a UNION reduces the probable costs of job loss by providing
®nancial bene®ts and job location assistance. Part-time jobs are usually
available but are unlikely to be viewed as career positions. Thus, losing a
part-time position is perceived to be less damaging than losing a full-time
appointment, leading PART-TIME employees to provide less e?ort. On the
other hand, PART-TIME employees may provide extraordinary e?ort to
enhance their likelihood of being o?ered a full-time position when one becomes
available.
Job loss is probably viewed as particularly burdensome to people with
more CHILDREN. The responsibilities associated with child rearing are
expected to inspire greater e?ort. As SPOUSE EARNINGS rise the per-
14 Because we are controlling for tenure and general work experience, age is a biological or a real time
variable here. However, the age spread is so small across our sample that it cannot really capture
important life-cycle tissues. It is best interpreted as a learning variable.
366 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
ceived costs of job loss fall and, most likely, e?ort. Similarly, individuals with
greater ®nancial ASSETS will be less fearful of job loss and, ceteris paribus,
will offer less effort on the job.
Women and minorities may believe that discrimination makes it dicult
to secure comparable employment if they are discharged. If so, they face a
higher perceived cost of job loss. Thus, BLACK and HISPANIC workers
are expected to give greater e?ort than otherwise equivalent white employees
do, while MALE workers are expected to exert less e?ort relative to
women.
Persons who are MARRIED are expected to bene®t from superior social
support, relative to comparable individuals who are not married, leading to
a greater sense of self-ecacy and motivation. Individuals who grew up in
households that were aliated with a RELIGION are likely to have
developed coping skills and strategies that contribute to self-ecacy or
e?ort.
Firms can use external monitoring to extract greater e?ort from their work
force. However, as the number of employees at a work site expands, it becomes
more dicult to detect a worker’s intensity on the job. Therefore,
greater ESTABLISHMENT SIZE may diminish worker e?ort. On the other
hand, larger ®rms provide more opportunities for advancement, which may
motivate workers. Thus, it is unclear how ESTABLISHMENT SIZE will
in¯uence worker e?ort. Firms with MULTIPLE LOCATIONS or work sites,
particularly if they are LARGE MULTIPLE LOCATIONS, o?er more
opportunities for professional advancement. Workers identi®ed as giving
greater e?ort are more likely to be granted transfer promotions. Thus, individuals
employed by such ®rms are expected to engage in more internal
monitoring and to extend greater e?ort on the job. The vector Ai contains
three variables representing ®rm characteristics that may in¯uence the extent
of external monitoring workers face, as well as likely employee commitment
to internal monitoring.
Jobs that are challenging and provide workers a high degree of autonomy
are expected to induce greater e?ort from workers controlling for the level of
external monitoring. MANAGEMENT, PROFESSIONAL and CRAFT
positions may o?er these desirable work characteristics relative to LABORER
jobs. Thus, the e?ort equation includes dummy variables that
identify occupation of employment. To account for the possibility that
worker e?ort varies systematically across industries, ceteris paribus, dummy
variables for industry of employment also are included in the effort equation
(Eq. (4.1)).
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 367
4.2.2. Wage equation
Eq. (4.2) stipulates that individuals who expend greater e?ort, ei, and
possess more human capital, Hi, command a higher real wage. The vector Xi
contains a standard set of demographic (e.g. race, gender, marital status,
dependents) and work place (e.g. occupation, industry, local unemployment
rate, ®rm size, union) wage equation regressors.
The wage a person receives also may be a?ected by the region of the US in
which they are employed. Controlling for personal characteristics and labor
market factors Kiefer and Smith (1977) and Sahling and Smith (1983) o?er
evidence that signi®cant regional wage di?erentials exist for otherwise comparable
workers. These pay di?erences may re¯ect cultural and institutional
variation in setting pay scales in internal labor markets, and incomplete responses
to regional labor market shocks. To account for the in¯uence of
region of employment on wages, Xi contains dummy variables to identify
employment in the WEST, NORTHEASTS, and NORTHCENTRAL regions
of the US.
A person’s WAGEi relative to their expected wage, EXPECTED WAGEi,
appears in the e?ort equation (4.1), and EFFORTi is included in the wage
equation (4.2). This accounts for the joint determination of both WAGEi and
EFFORTi. EFFORTi is independent of the region of the country where an
individual is employed (WEST, NORTHEAST, NORTHCENTRAL) which
is expected to affect a person’s WAGEi. As a result, these regional dummy
variables are used to identify the effort equation, Eq. (4.1). Variables re-
¯ecting early childhood socialization (BOTH PARENTS, PARENT EDUCATION,
PROFESSIONAL PARENT), and household ®nancial factors
(SPOUSE EARNINGS, ASSETS) are expected to exert a direct in¯uence on
EFFORTi while only indirectly effecting WAGEi, through their impact on
EFFORTi. Because these variables are included in the effort equation but are
excluded from the wage Eq. (4.2), they identify the wage equation. 15
15 Frantz (1982) estimates a similar model to explore the relation between wages and changes in
attitudes. Using data from the National Longitudinal Survey of Young Men he jointly estimates a wage
equation and a change in attitude equation, where attitudes are measured by locus of control scores. In
contrast, we jointly estimate wages and locus of control. In addition, the equation we specify to explain
locus of control (e?ort) di?ers from the equation Frantz uses to explain locus of control (self-con®dence),
since we are estimating a model to test the eciency wage hypothesis. Thus, in our model e?ort depends on
factors in¯uencing the cost of job loss such as; earning an eciency wage (ie. a wage greater than
expected), educational accumulation, personal unemployment history, and the generosity of unemployment
bene®ts, which are not included in the attitude change equation estimated by Frantz.
368 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
4.3. Estimation technique
Two-stage least squares (2SLS) is used to estimate Eqs. (4.1) and (4.2). In
Stage I each endogenous variable is regressed on all of the exogenous variables
in the system by OLS. Using the coecient estimates from these reduced
form equations, we create estimated values of the endogenous variables or
instruments. 16 The estimated values of WAGEi and EFFORTi, are denoted
as PREDICTED WAGEi and PREDICTED EFFORTi respectively.
In Stage II, PREDICTED EFFORTi, which is uncorrelated with ei, the
wage equation error term, replaces EFFORTi ± which is correlated with ei ±
in Eq. (4.2). A person’s PREDICTED WAGEi, controlling for whether the
person is participating currently in the labor force, is likely to be equivalent
to their EXPECTED WAGEi. Therefore, a person’s EFFICIENCY WAGEi
± the di?erence between WAGEi and PREDICTED WAGEi is ei, the error
term in Eq. (4.2). In Stage II, ei ± a person’s unexpected wages ± is used as a
measure of this individual’s eciency wage in Eq. (4.1). A standard assumption
when estimating equations simultaneously is that cross equation
error terms are uncorrelated. Thus, we assume that ei is uncorrelated with li,
the e?ort equation error term. The structural equations are then estimated by
ordered probit and OLS, respectively. 17
Wages are observed only for those individuals working for pay. Heckman
(1979a,b) has suggested that unobservable features of an individual both
govern a person’s decision on whether or not to participate in the labor force
and their productivity, if they opt to work. If these factors are omitted from
the estimated equations, then the coecients will su?er from selectivity bias.
Following Heckman, a selection±correction variable (IMILLS) is included in
Eq. (4.2), the wage equation. 18 Since the unobservables that inspire a person
16 It might be argued that using a nonlinear estimation technique is more appropriate given that
EFFORTi as measured by a person’s score on the Mastery Scale is a non-continuous dependent variable.
However, predicted means and actual means can vary substantially using nonlinear methods. Fortunately,
the coecients from a OLS estimation, which are used to create the predicted values, are consistent; only
the standard errors are inconsistent. See Heckman (1979a,b) for a detailed discussion of these points.
17 Ordered probit is an appropriate procedure when the dependent variable is categorical and
sequential, such as our Mastery Scale measure of locus of control, and when errors are assumed normally
distributed (Maddala, 1983).
18 As suggested by Heckman (1979a,b) a preliminary regression is run to explain the probability of
working for pay. This equation is estimated as a Probit model and the resulting coecients are used to
construct (IMILLS),, the inverse Mills ratio. A table with the results of the probability of working for pay
equation is available from the authors upon request.
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 369
to participate in the labor force are factors that are likely to also improve
e?ort, (IMILLS) is included in Eq. (4.1), the e?ort equation.
5. Results
The system of equations describing the joint determination of EFFORT
and WAGES, Eqs. (4.1) and (4.2), was estimated separately by gender, race,
and ethnicity using data drawn from the NLSY in 1992. For each of these
data sets, the results for the structural e?ort equation appear in Table 3.
Table 4 presents our estimates of the structural wage equation. 19
5.1. E?ort
The results in Table 3 indicate that receiving a greater EFFICIENCY
WAGE signi®cantly enhances a worker’s e?ort for each of the data sets. Thus,
we ®nd evidence consistent with the eciency wage hypothesis. 20 This ®nding
is also consistent with stress process theory ± unexpectedly high earning reduce
life stresses and enhances self-ecacy and hence e?ort. To explore
whether the impact of earning an eciency wage on e?ort varies by industry
Table 3
Structural ordered probit e?ort estimatesa
Variable name
(expected sign)
All Male Female White Black Hispanic
EFFICIENCY
WAGE (+)
0.15 (4.26) 0.17 (3.62) 0.12 (2.16) 0.14 (2.93) 0.15 (2.01) 0.17 (2.20)
UNEMPLOYMENT
(+)
0.09 (1.92) 0.12 (1.76) 0.59 (0.85) 0.40eÿ02 (0.06) 0.35 (2.24) 0.12 (1.44)
UI BENEFITS (ÿ) 0.90eÿ04 (0.15) ÿ0.19eÿ03 (0.23) 0.44eÿ03 (0.52) ÿ0.79eÿ03 (0.96) 0.23eÿ02 (1.96) ÿ0.16eÿ03 (0.10)
SMSA (ÿ) 0.25eÿ01 (0.67) 0.16eÿ01 (0.31) 0.35eÿ01 (0.63) 0.68eÿ02 (0.13) 0.22eÿ01 (0.29) ÿ0.95eÿ01 (0.97)
UNEMPLOYMENT
BOUTS (?)
ÿ0.14eÿ01 (2.50) ÿ0.13eÿ01 (1.60) ÿ0.15eÿ01 (1.67) ÿ0.18eÿ01 (2.16) ÿ0.20eÿ01 (1.73) ÿ0.32eÿ02 (0.25)
UNEMPLOYMENT
DURATION (?)
ÿ0.19eÿ02
(2.63) ÿ0.11eÿ02 (1.13) ÿ0.28eÿ02 (2.47) ÿ0.18eÿ02 (1.53) ÿ0.25eÿ02 (2.23) ÿ0.47eÿ03 (0.25)
EDUCATION (?) 0.29eÿ01 (2.81) 0.25eÿ01 (1.74) 0.30eÿ01 (2.00) 0.16eÿ01 (1.08) 0.46eÿ01 (2.20) 0.44eÿ01 (2.04)
EXPERIENCE (?) 0.25eÿ03 (0.81) 0.37eÿ03 (0.80) 0.11eÿ03 (0.23) 0.10eÿ03 (0.23) 0.12eÿ03 (0.19) 0.83eÿ03 (1.24)
AFQT (?) 0.58eÿ02 (6.97) 0.66eÿ02 (5.70) 0.49eÿ02 (3.96) 0.41eÿ02 (3.68) 0.11eÿ01 (5.76) 0.74eÿ02 (3.74)
TENURE (?) ÿ0.19eÿ03 (1.85) ÿ0.98eÿ04 (0.70) ÿ0.29eÿ03 (1.94) ÿ0.29eÿ03 (2.11) ÿ0.71eÿ04 (0.34) ÿ0.17eÿ03 (0.68)
JOB TRAINING (?) 0.45eÿ01 (0.80) 0.26eÿ01 (0.32) 0.47eÿ01 (0.59) 0.23eÿ02 (0.03) 0.16eÿ01 (0.14) 0.17 (1.26)
AGE (ÿ) ÿ0.45eÿ01
(4.96) ÿ0.56eÿ01
(4.33) ÿ0.34eÿ01
(2.56) ÿ0.45eÿ01
(3.59) ÿ0.45eÿ01
(2.46) ÿ0.49eÿ01
(2.37)
UNION (ÿ) 0.65eÿ01 (1.37) 0.81eÿ01 (1.27) 0.29eÿ01 (0.40) 0.62eÿ01 (0.88) ÿ0.12eÿ01 (0.14) 0.13 (1.23)
PART-TIME (?) ÿ0.98eÿ01 (1.86) ÿ0.21 (2.19) ÿ0.32eÿ01 (0.49) ÿ0.11 (1.48) ÿ0.11 (0.10) ÿ0.14 (1.10)
CHILDREN (+) ÿ0.12eÿ01 (0.42) ÿ0.25eÿ01 (1.06) ÿ0.13eÿ01 (0.58) ÿ0.10eÿ01 (0.43) ÿ0.67eÿ02 (0.24) ÿ0.26eÿ01 (0.77)
SPOUSE
EARNINGS (ÿ)
0.35eÿ05 (2.81) 0.63eÿ05 (2.56) 0.30eÿ05 (1.89) 0.37eÿ05 (2.30) 0.17eÿ05 (0.66) 0.33eÿ05 (0.99)
ASSETS (ÿ) 0.18eÿ05 (2.48) 0.15eÿ05 (1.73) 0.25eÿ05 (1.94) 0.13eÿ05 (1.73) 0.59eÿ05 (2.10) 0.66eÿ05 (2.43)
MALE (ÿ) 0.13 (3.31) 0.15 (2.90) 0.16 (2.18) 0.50eÿ01 (0.53)
BLACK (+) 0.89eÿ01 (1.96) 0.82eÿ01 (1.31) 0.10 (1.55)
HISPANIC (+) 0.72eÿ01 (1.50) 0.30eÿ03 (0.46) 0.99eÿ01 (1.38)
MARRIED (+) 0.81eÿ01 (1.85) 0.11 (1.73) 0.20eÿ01 (0.31) 0.13 (2.11) 0.23eÿ01 (0.28) 0.47eÿ01 (0.46)
RELIGION (+) ÿ0.62eÿ01 (0.80) ÿ0.11 (1.10) 0.45eÿ02 (0.04) 0.37eÿ02 (0.04) ÿ0.12eÿ01 (0.89) ÿ0.33eÿ01 (1.14)
ESTABLISHMENT
SIZE (ÿ)
0.65eÿ05 (0.89) 0.77eÿ05 (0.71) 0.67eÿ05 (0.67) 0.70eÿ05 (0.71) 0.79eÿ05 (0.58) 0.12eÿ04 (0.62)
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 371
Table 3 (Continued)
Variable name
(expected sign)
All Male Female White Black Hispanic
MULTIPLE
LOCATIONS (+)
0.19eÿ03 (1.00) 0.72eÿ01 (1.31) ÿ0.91eÿ01 (1.54) ÿ0.69eÿ01 (1.27) 0.88eÿ01 (0.11) 0.17eÿ01 (1.77)
LARGE MULTIPLE
LOCATIONS (+)
ÿ0.28eÿ01 (0.68) ÿ0.80eÿ01 (1.38) 0.40eÿ01 (0.68) ÿ0.35eÿ01 (0.61) 0.19eÿ01 (0.24) ÿ0.89eÿ01 (0.94)
PROFESSIONAL
PARENT (+)
ÿ0.67eÿ01 (1.61) ÿ0.36eÿ01 (0.61) ÿ0.11 (1.86) ÿ0.54eÿ01 (1.04) ÿ0.95eÿ01 (0.97) ÿ0.45eÿ01 (0.39)
BOTH PARENTS
(+)
0.18eÿ02 (0.48) ÿ0.63eÿ02 (0.12) 0.47eÿ01 (0.85) ÿ0.25eÿ01 (0.39) 0.49eÿ01 (0.83) 0.39eÿ01 (0.45)
PARENT
EDUCATION (+)
0.11eÿ01 (1.68) 0.48eÿ02 (0.53) 0.16eÿ01 (1.71) 0.17eÿ01 (1.56) 0.46eÿ02 (0.33) 0.42eÿ02 (0.38)
IMILLS (?) 0.90eÿ01 (0.50) 0.21 (0.62) ÿ0.20eÿ01 (0.08) 0.28eÿ01 (0.10) 0.38eÿ01 (0.11) 0.39 (1.05)
n 5579 2933 2646 3013 1509 1057
Log Likelihood ÿ7008 ÿ3578 ÿ3406 ÿ3570 ÿ2029 ÿ1354
Chi square [D.F.] 536 [45] 356 [44] 218 [44] 243 [43] 205 [43] 136 [43]
a All equations include INDUSTRY and OCCUPATION dummy variables (t-statistics in parentheses).
* Statistically signi®cantly di?erent from zero at the 0.1 con®dence level.
** Statistically signi®cantly di?erent from zero at the 0.05 con®dence level.
*** Statistically signi®cantly di?erent from zero at the 0.01 con®dence level.
372 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
Table 4
Structural OLS log wage estimatesa
Variable name
(expected sign)
All Male Female White Black Hispanic
PREDICTED
EFFORT (+)
0.58 (6.86) 0.48 (4.04) 0.35 (4.18) 0.78 (6.59) 0.71 (6.22) 0.13 (4.43)
EDUCATION (+) 0.21eÿ01 (4.50) 0.17eÿ01 (2.67) 0.34eÿ01 (5.56) 0.25eÿ01 (4.22) ÿ0:57 eÿ02 (0.62) 15eÿ01 (1.44)
EXPERIENCE (+) 0.62eÿ03 (5.48) 0.51eÿ03 (2.72) 0.78eÿ03 (5.11) 0.70eÿ03 (4.54) 0.49eÿ03 (2.37) 0.33eÿ03 (1.11)
TENURE (+) 0.44eÿ03 (10.13) 0.36eÿ03 (6.29) 0.44eÿ03 (7.25) 0.54eÿ03 (8.47) 0.52eÿ03 (7.01) 0.19eÿ03 (1.96)
JOB TRAINING (+) 0.14eÿ01 (0.63) 0.63eÿ01 (2.05) ÿ0.88eÿ02 (0.30) 0.51eÿ01 (1.74) ÿ0.60eÿ01 (1.51) 0.22eÿ01 (0.39)
AFQT (+) ÿ0.84eÿ03 (1.44) ÿ0.37eÿ03 (0.44) 0.75eÿ03 (1.18) ÿ0.81eÿ03 (1.25) ÿ0.30eÿ02 (2.41) ÿ0.32eÿ02 (2.52)
AGE (+) 0.13eÿ01 (2.53) 0.16eÿ01 (2.01) ÿ0.13eÿ02 (0.22) 0.18eÿ01 (2.45) 0.24eÿ01 (3.15) 0.18eÿ01 (1.69)
UNION (+) 0.12 (6.35) 0.15 (5.65) 0.10 (3.70) 0.11 (4.08) 0.14 (4.70) 0.13 (2.99)
ESTABLISHMENT
SIZE (+)
0.79eÿ05 (2.84) 0.90eÿ05 (2.18) 0.91eÿ05 (2.48) 0.19eÿ05 (0.49) 0.72eÿ05 (1.61) 0.18eÿ04 (2.25)
MULTIPLE
LOCATIONS (+)
0.31eÿ01 (1.98) ÿ0.93eÿ02 (0.38) 0.71eÿ01 (3.00) 0.95eÿ01 (4.25) 0.31eÿ02 (0.11) ÿ0.10 (2.37)
LARGE MULTIPLE
LOCATIONS (+)
0.54eÿ01 (3.34) 0.75eÿ01 (3.06) 0.24eÿ01 (1.10) 0.75eÿ01 (3.30) 0.34eÿ01 (1.19) 0.50eÿ01 (1.31)
PART-TIME (ÿ) 0.16eÿ01 (0.72) 0.15 (2.79) ÿ0.44eÿ01 (1.82) 0.44eÿ01 (1.43) ÿ0.17eÿ01 (0.46) 0.17eÿ01 (0.29)
SMSA (+) 0.88eÿ01 (5.96) 0.89eÿ01 (4.32) 0.95 (4.54) 0.14 (6.81) 0.22eÿ01 (0.79) 0.85eÿ01 (2.04)
UNEMPLOYMENT
(ÿ)
ÿ0.49eÿ01 (2.44) ÿ0.30eÿ01 (1.02) ÿ0.50eÿ01 (1.88) 0.83eÿ01 (3.07) ÿ0.21 (3.45) ÿ0.16 (4.20)
UNEMPLOYMENT
BOUTS (ÿ)
0.18eÿ02 (0.68) ÿ0.32eÿ02 (0.90) ÿ0.75eÿ03 (0.20) 0.11eÿ01 (2.88) 0.67eÿ02 (1.33) 0.13eÿ01 (2.42)
UNEMPLOYMENT
DURATION (ÿ)
0.14eÿ02 (3.88) 0.49eÿ03 (1.14) 0.98eÿ03 (1.88) 0.15eÿ02 (2.82) 0.22eÿ02 (3.94) 0.43eÿ03 (0.55)
MALE (+) 0.10 (6.60) 0.91eÿ01 (3.96) 0.19eÿ01 (0.64) 0.14 (3.99)
BLACK (ÿ) ÿ0.85eÿ01
(4.21)
0.10eÿ01 (3.74) ÿ0.28eÿ01 (1.01)
HISPANIC (ÿ) 0.46eÿ03 (0.03) ÿ0.46eÿ02 (0.18) 0.25eÿ01 (0.94)
MARRIED (ÿ) ÿ0.36eÿ01 (1.82) ÿ0.21eÿ01 (0.63) ÿ0.17eÿ01 (0.82) ÿ0.10 (3.16) ÿ0.16eÿ01 (0.61) ÿ0.29 (0.75)
CHILDREN (ÿ) 0.11eÿ02 (0.17) 0.21eÿ01 (2.14) ÿ0.22eÿ02
(2.57)
0.26eÿ02 (0.29) 0.23eÿ02 (0.23) 0.20eÿ01 (1.33)
NORTHEAST (?) 0.14 (7.19) 0.86 (2.82) 0.19 (7.17) 0.12 (4.63) 0.88eÿ01 (2.37) 0.20 (4.12)
A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385 373
Table 4 (Continued)
Variable name
(expected sign)
All Male Female White Black Hispanic
NORTHCENTRAL
(?)
ÿ0.60eÿ01
(3.01) ÿ0.28eÿ01 (1.01) 0.48eÿ01 (1.84) ÿ0.83eÿ01
(3.47) ÿ0.69eÿ01 (1.95) ÿ1.5 (1.97)
WEST (?) 0.45eÿ01 (2.19) 0.55 (1.97) 0.79eÿ01 (2.81) ÿ0.45eÿ01 (1.45) 0.15 (5.62) 0.65eÿ01 (1.79)
IMILLS (?) 0.16 (2.42) 0.16 (1.19) 0.24 (3.04) 0.13 (1.40) 0.14 (1.19) 0.55eÿ02 (0.04)
CONSTANT (?) ÿ2.65 (4.82) ÿ1.92 (2.39) ÿ1.16 (2.12) ÿ4.12 (5.16) ÿ3.43 (4.57) ÿ2.38 (2.85)
n 5579 2933 2646 3013 1509 1057
F 98.75 [41, 5537] 45.38 [40, 2892] 52.17 [40, 2605] 52.63 [39, 2973] 33.80 [39, 1469] 16.37 [39, 1017]
R2 0.42 0.39 0.44 0.41 0.47 0.39
a All equations include one-digit industry and one-digit occupation dummy variables (t-statistics in parentheses).
* Statistically signi®cantly di?erent from zero at the 0.1 con®dence level.
** Statistically signi®cantly di?erent from zero at the 0.05 con®dence level.
*** Statistically signi®cantly di?erent from zero at the 0.01 con®dence level.
374 A.H. Goldsmith et al. / Journal of Economic Psychology 21 (2000) 351±385
Table 5
Estimated e?ect of eciency wage on e?ort, and e?ort on the wage: by industrya
(1) (2) (3) (4) (5)
INDUSTRY (Sample size) o?EFFORT?
o?EFFICIENCY WAGE?
o?Wage?
o?Effort?
Wage equation industry
dummy variables: Krueger
±Summers
Wage equation industry
dummy variables: Goldsmith
±Veum±Darity
Agriculture & Mining (185) ÿ0.75eÿ01 (0.42) 0.64 (0.82) 0.22 (2.96) ÿ0.60eÿ01 (1.58)
CONSTRUCTION (384) 0.14eÿ01 (0.10) 1.62 (3.66) 0.11 (3.18) 0.13 (4.65)
MANUFACTURING (1057) 0.43 (2.16) 0.67 (4.18) 0.91eÿ01 (2.84) Control
TRANSPORTATION (373) 0.35 (2.17) 0.84 (2.22) 0.15 (4.26) 0.83eÿ01 (3.05)
WHOLESALE & RETAIL TRADE (952) 0.19 (2.10) 0.32 (1.44) ÿ0.11 (3.36) ÿ0.22 (10.22)
FINANCE (346) 0.21 (1.21) 0.66 (2.03) 0.55eÿ01 (1.62) 0.95eÿ02 (0.33)
BUSINESS & REPAIR SERVICES (440) 0.18 (1.39) 0.96 (3.02) ÿ78eÿ01 (2.43) ÿ0.10eÿ01 (0.37)
PERSONAL SERVICES & ENTERTAINMENT (298) ÿ0.66eÿ01 (0.54) 0.82 (1.80) ÿ0.22 (6.83)
PROFESSIONAL SERVICES (1179) 0.80eÿ01 (1.02) 0.50 (2.76) ÿ0.82eÿ01 (3.70)
PUBLIC ADMINISTRATION (365) 0.36 (1.91) 0.46 (1.55) ÿ0.13eÿ01 (0.46)
a The coecients reported in columns 2 and 3 are extracted from estimates of Eqs. (4.1) and (4.2) using data on a particular industry ± industry is
identi®ed by row. The remaining coecient estimates are suppressed. Columns 4 and 5 report coecients estimated on industry dummy var
