The current issue and full text archive of this journal is available at

www.emeraldinsight.com/0143-7720.htm

IJM
31,6 Fringe benefits and job
satisfaction
Benjamin Artz
626 Department of Economics, American University of Sharjah, Sharjah,
United Arab Emirates

Abstract
Purpose – The paper seeks to empirically identify the theoretically ambiguous relationship between
employer fringe benefit provision and worker job satisfaction.
Design/methodology/approach – Using the five most recent waves of the National Longitudinal
Survey of Youth, both pooled cross-section and fixed effects estimates explain the relationship
between fringe benefits and job satisfaction. The potential endogenous relationship is also tested using
a recursive bivariate probit procedure.
Findings – Fringe benefits are significant and positive determinants of job satisfaction. The potential
endogeneity between fringe benefits and job satisfaction is not shown in this dataset while controlling
for fixed effects does not remove the significant impact of fringe benefits.
Research limitations/implications – A limitation is the inability to control for total compensation
within the estimations and control for wage changes as a result of fringe benefit provision.
Practical implications – Higher levels of worker job satisfaction, potentially resulting from fringe
benefit provisions, have been linked to important productivity measures such as lower quit rates and
absenteeism.
Originality/value – The paper is the first to study the relationship between fringe benefits and job
satisfaction in detail while additionally testing for the endogeneity of the relationship and controlling
for fixed effects.
Keywords Job satisfaction, Benefits, Compensation, Employee attitudes
Paper type Research paper

International Journal of Manpower
Vol. 31 No. 6, 2010
pp. 626-644
q Emerald Group Publishing Limited
0143-7720
DOI 10.1108/01437721011073346
1. Introduction
Establishing the determinants of job satisfaction remains at the forefront of empirical
testing in using measures of on-the-job utility. At first consideration, desirable job
attributes such as fringe benefits should increase job satisfaction. However, the past
evidence is mixed at best and contradictory at worst. While a valuable form of
compensation, employer provided benefits may lower earnings or reduce job mobility.
This is particularly evident in the USA where many benefits are employer provided and
not transferable between jobs. Thus, the theoretical impact of fringe benefits on job
satisfaction is not immediately clear.
Fringe benefits can impact job satisfaction in several ways. First, fringe benefits stand
as an important component of worker compensation. The National Compensation Survey
conducted by the Bureau of Labor Statistics estimated that benefits made up 30 per cent
of total compensation for all US civilian workers in 2006[1] Some benefits such as
Social Security and Medicare, the country’s publicly provided elderly pension and
The author would like to thanks John Heywood and the participants of the Labor Seminar Series
at the University of Wisconsin – Milwaukee for comments and guidance as well as two
anonymous referees for their insight and suggestions.
health insurance, are legally required and make up roughly 27 per cent of all benefit Fringe benefits
compensation. The remaining 73 per cent of benefit compensation is comprised mostly of
paid leave, insurance plans and retirement and savings plans. These benefits are often not
subject to taxation and are, therefore, cheaper to gain through an employer than through
the market (Alpert, 1987). Consequently, cheaper benefits should increase worker job
satisfaction.
Second, fringe benefits can act as valuable substitutes for wages. Employers may 627
choose to offer fringe benefits since workers can have strong preferences for fringe
benefits thus decreasing the prevalence of turnover as effectively as an equivalently
valuable increase in wages (Dale-Olsen, 2006). In addition, Baughman et al. (2003)
examined employer survey data and found that employers decreased wages once
several benefits had been offered to employees after a few years. Woodbury (1983) found
that workers also view benefits and wages as substitutes, willing to give up wages in
exchange for more benefits. This substitution can increase job satisfaction if the
worker’s marginal income tax rate decreases after giving up wages for fringe benefits.
The less taxed fringe benefits can be substituted for wages and increase job satisfaction
by saving the worker from increased tax burden.
Third, the substitution between wages and benefits can have a negative impact on job
satisfaction if workers find they must sacrifice wages and accept provision of a fringe
benefit they do not necessarily desire. For instance, workers’ spouses may already have
provision of a particular fringe benefit, so a second provision of that fringe benefit may
be viewed as wasteful and can therefore decrease job satisfaction, especially if wages are
lower as a result. On the other hand, workers may find a particular fringe benefit as
essential. Workers may have a feeling of job-lock to a particular employer or job if they
are uncertain about the provision of the necessary fringe benefit at a different place of
work. For example, pre-existing health conditions are often not covered under
employer-provided health insurance, providing a need for employees to remain at jobs
that they may not be satisfied with (Adams, 2004). The combination of uncertainty and
job-lock can decrease job satisfaction.
Since the expected impact of fringe benefits on job satisfaction is unclear, it is not
surprising that past research is inconclusive. When included in typical estimates, the
impact of fringe benefits on job satisfaction is rarely significant. In addition, the evidence
mainly depends on cross-sectional comparisons, raising questions about potential
biases. First, the impact of a particular fringe benefit on job satisfaction can be
misleading if the worker has unmeasured individual specific determinants of job
satisfaction. Indeed, we cannot assume that workers are randomly sorted into jobs but
rather that they sort themselves into the jobs that suit their preferences. In addition, job
satisfaction and fringe benefits may be simultaneously determined such that fringe
benefits are endogenous in determining job satisfaction.
Thus, the relationship between fringe benefits and job satisfaction has not been
appropriately tested. Very little past research has isolated and examined fringe benefits
as a primary determinant of job satisfaction, few studies have included as many fringe
benefits as are available in the US National Longitudinal Survey of Youth (NLSY) and
none have studied the relationship between fringe benefits and job satisfaction in detail,
controlling for fixed effects and testing for endogeneity[2].
In distinction to past results, a pooled cross-section of five NLSY waves suggests that
fringe benefits are important in determining job satisfaction. In addition several
IJM sub-sample estimations suggest fringe benefits are more significant in determining the
31,6 job satisfaction of particular groups such as workers with children at home or non-union
workers. Then, in order to test for the simultaneous determination of fringe benefits and
job satisfaction, a recursive bivariate probit model is used to test for the possible
correlation between the disturbances in job satisfaction and fringe benefit structural
equations. The cross-equation correlation is not significantly different from zero,
628 implying that fringe benefits can be treated as exogenous in an estimation of job
satisfaction and can be properly estimated within the ordered probit framework used in
the pooled cross-section. Finally, after worker sorting is controlled for in fixed effects
estimates, the results are not considerably different from those of the pooled
cross-section estimates, further confirming the importance of fringe benefits in
determining worker job satisfaction.
The following section discusses previous research, the uncertain relationship
between fringe benefits and job satisfaction and the need for a detailed study on the
topic. Section 3 outlines the data that is used and the primary empirical methodology
used to test the relationship. Section 4 discusses the results while Section 5 outlines tests
for endogeneity and unobservable worker characteristics as possible biases of the
primary results. The final section concludes.

2. Past research and ambiguous results

Over the past four decades, economists have given job satisfaction increasing
attention. Job satisfaction is negatively related to job turnover (Freeman, 1978; McEvoy
and Cascio, 1985; Akerlof et al., 1988; Weiss, 1984), absenteeism (Clegg, 1983) and
positively related to productivity (Mangione and Quinn, 1975). Therefore, it is useful to
understand which job characteristics and provisions increase job satisfaction.
Although fringe benefits stand as an important piece of worker compensation
packages they have not been given much attention in the job satisfaction literature. Fringe
benefits have merely acted as controls in most studies and not as the primary subject of
scrutiny. Indeed, more than one or two measures of fringe benefits are rarely found as
independent variables in job satisfaction studies.
Rather, pensions often act as the predominant proxy for fringe benefit provision
within the job satisfaction literature and consequently the estimated impact of fringe
benefits on job satisfaction[3]. Some studies such as Artz (2008) and Donohue and
Heywood (2004) find that pensions do not significantly impact job satisfaction in
cross-section estimates. Yet Bender et al. (2005), Heywood and Wei (2006) and Bender
and Heywood (2006) find that pensions positively impact job satisfaction in cross-section
estimates. Still others find the opposite result. Heywood et al. (2002) use the 1991-1994
waves of the British Household Panel Study and find that pensions negatively impact job
satisfaction while Luchak and Gellatly (2002), in a particularly interesting study,
consider the impact of pension accrual on job satisfaction using a dedicated sample of
429 employees in a large, unionized public utility company in Canada. They posit that as
employees’ pensions increase in value over their job tenure, workers may feel more
vulnerable to job loss since firms may opportunistically layoff employees to reduce
pension liabilities. The authors use this hypothesis to explain their result that pension
accrual decreases job satisfaction[4].
Although pensions are the predominant proxy for fringe benefits, some authors
do include multiple fringe benefit measures as independent variables in their
respective models. Donohue and Heywood (2004) report positively significant Fringe benefits
estimates for such variables as paid vacation and sick pay but no significance for any
of the remaining benefits: child care, pension, profit sharing, employer provided
training/education and health insurance. Uppal (2005) uses a measure comprised of the
number of fringe benefits employees receive and finds that this is positively related to
job satisfaction. However, Benz (2005) includes most of the fringe benefits found in
NLSY waves 1994-2000 in his study of employees of non-profit organizations and finds 629
only two out of nine fringe benefits are positive and significantly related to job
satisfaction and that one is negative and significant.
The ambiguous results of past estimates arise primarily from the conflicting
theoretical effects that fringe benefits can have on job satisfaction, but theory may not be
the only explanation for the differences. Some of these mixed results may stem from the
use of alternative sources of data or from the differences in institutions or labor markets
between countries. Particularly in the USA, benefits can have a different impact on job
satisfaction than elsewhere because many benefits such as pensions and health
insurance are employer provided and do not necessarily move with an employee upon a
change in employer. This characteristic of the US labor market explains how fringe
benefits may impact job satisfaction in a different way than in other countries where
publicly provided fringe benefits are more commonplace.

3. Data and methodology

The data used are five waves of the NLSY with each wave representing every other year
from 1996 to 2004. All five waves of this USA data contain a measure of overall job
satisfaction and dozens of control variables including occupation and industry codes as
well as demographic and job characteristics. In total, there are 24,090 observations after
pooling the five waves of data. The sample includes all workers except those claiming to
be self-employed. This paper assumes fringe benefits are likely to have a different
impact on the job satisfaction of self-employed workers and thus adversely affect
the desired estimations. The mean, standard deviations and definitions of all utilized
variables are presented in Table I. The NLSY is unique in that it offers many different
kinds of fringe benefit variables for analysis as well as longitudinal data that can be
used to control for fixed effects. This paper specifically includes eight different
employer-provided or employer-offered fringe benefits as well as generated variables
representing the quantity of these benefits individuals claim to have. The latter
variables present the general importance of fringe benefits in determining job
satisfaction rather than each individual fringe benefit’s significance.
Overall, job satisfaction is measured on a scale of one to four, four representing the
highest level of job satisfaction. The typical cross-section estimate of job satisfaction fits
this Likert scale to the cumulative normal distribution through the ordered-probit
estimation. An ordered-probit estimation is commonly used in order to capture all of the
variation between subjective measures of job satisfaction that are not necessarily linear
in nature. While cross-sectional data of individual years have been used in past studies,
a problem may arise from the inability of the individual waves of the NLSY to exhibit
enough variation among fringe benefits and job satisfaction to properly estimate the
impact of fringe benefits on job satisfaction[5]. Therefore, pooled cross-section
estimation is used to expand the sample size. Since pooling longitudinal data generally
causes correlation within the individual worker observations across the waves of data,
IJM
Variables Definitions Mean SD
31,6
Job satisfaction ¼ 4 if very satisfied, 3 if satisfied, 2 if unsatisfied
and 1 if very unsatisfied 3.369 0.692
Flexible work hours ¼ 1 if employer allows flexible work hours or a
flexible schedule 0.537 0.499
630 Pension or retirement ¼ 1 if employer provides a pension or retirement
plan 0.608 0.488
Dental insurance ¼ 1 if employer offers dental insurance 0.700 0.458
Parental leave ¼ 1 if employer allows maternity or paternity leave 0.702 0.457
Child care ¼ 1 if employer provides or subsidizes child care 0.089 0.285
Vacation days ¼ 1 if worker has paid vacation days 0.829 0.376
Profit sharing ¼ 1 if employer offers profit sharing 0.270 0.444
Health insurance ¼ 1 if employer offers off the job health insurance 0.807 0.394
Female ¼ 1 if respondent is female 0.507 0.500
Black ¼ 1 if respondent in Black 0.294 0.456
Hispanic ¼ 1 if respondent is Hispanic 0.192 0.394
High school education ¼ 1 if highest grade completed is 12 0.447 0.497
College education ¼ 1 if highest grade completed is more than 12 but
less than 17 0.368 0.482
Post college education ¼ 1 if highest grade completed is more than 16 0.089 0.285
Age ¼ age of respondent 38.789 3.575
Health impairs job ¼ 1 if respondent has any health conditions that
limit job performance 0.070 0.255
Married ¼ 1 if respondent is currently married 0.626 0.484
Urban residence ¼ 1 if respondent lives in an urban area 0.744 0.437
Hourly wage ¼ wage calculation based on respondent’s usual
wage, time unit of pay and usual hours worked
per time unit 16.343 15.080
Weekly hours worked ¼ usual hours worked per week 41.416 9.621
Tenure ¼ tenure in weeks with employer as of interview
date 332.448 308.96
Recent promotion ¼ 1 if recent position change was a promotion 0.154 0.361
Union member ¼ 1 if respondent is a member of a labor union 0.175 0.380
Public employer ¼ 1 if respondent works for the government 0.183 0.386
Medium firm ¼ 1 if respondent has between 50 and 249
coworkers 0.264 0.441
Table I. Big firm ¼ 1 if respondent has greater than or equal to
Variable definitions 250 coworkers 0.303 0.459
and sample means Observations ¼ number of pooled sample observations 24,090

clustering standard errors by individual respondent is necessary to maintain consistent

estimators. After doing so, the dataset contains a larger sample size and offers more
convincing results. Additionally, sampling weights are used to adjust the survey
observations in order to fit the population better as well as correct for such survey
variations as sample under-coverage.

4. Results
The pooled cross-section estimation results are presented in Table II. The first column
shows that a continuous variable consisting of fringe benefit quantities workers claim
 Pooled cross section waves 1996-2004

Fringe benefits (0-8) 0.061 * * * (0.006) [0.024] – –

One fringe benefit – 0.091 * (0.053) [0.036] –
Two fringe benefits – 0.154 * * * (0.058) [0.062] –
Three fringe benefits – 0.142 * * * (0.054) [0.057] –
Four fringe benefits – 0.175 * * * (0.052) [0.070] –
Five fringe benefits – 0.249 * * * (0.050) [0.099] –
Six fringe benefits – 0.299 * * * (0.052) [0.119] –
Seven fringe benefits – 0.472 * * * (0.056) [0.186] –
Eight fringe benefits – 0.623 * * * (0.080) [0.239] –
Flexible work hours – – 0.224 * * * (0.022) [0.089]
Pension or retirement plan – – 0.068 * * * (0.025) [0.027]
Dental insurance – – 0.052 (0.033) [0.021]
Parental leave – – 0.117 * * * (0.028) [0.046]
Employer provided child care – – 0.105 * * (0.042) [0.042]
Employer offered vacation days – – 20.035 (0.034) [2 0.014]
Profit sharing – – 0.061 * * (0.026) [0.024]
Employer provided health insurance – – 20.078 * (0.041) [2 0.031]
Female 0.076 * * * (0.029) 0.076 * * * (0.028) 0.065 * * (0.029)
Black 20.068 * * (0.029) 2 0.076 * * * (0.029) 20.073 * * (0.029)
Hispanic 0.110 * * * (0.032) 0.108 * * * (0.032) 0.111 * * * (0.032)
High school education 20.084 * (0.043) 2 0.079 * (0.043) 20.084 * * (0.043)
College education 20.136 * * * (0.047) 2 0.131 * * * (0.046) 20.142 * * * (0.047)
Post college education 20.015 (0.065) 2 0.011 (0.065) 20.025 (0.065)
Age 0.011 (0.045) 0.009 (0.045) 0.005 (0.045)
Age squared 20.014 (0.058) 2 0.011 (0.058) 20.007 (0.058)
Married 0.066 * * * (0.024) 0.068 * * * (0.024) 0.069 * * * (0.024)
Health impairs job 20.163 * * * (0.041) 2 0.166 * * * (0.041) 20.169 * * * (0.042)
Urban residence 20.070 * * * (0.025) 2 0.071 * * * (0.025) 20.070 * * * (0.025)
Northeast region 20.131 * * * (0.043) 2 0.128 * * * (0.043) 20.123 * * * (0.043)
Midwest region 20.061 (0.039) 2 0.062 (0.039) 20.063 (0.039)
South region 20.005 (0.037) 2 0.002 (0.037) 0.007 (0.037)
Log hourly wages 0.137 * * * (0.027) 0.140 * * * (0.027) 0.149 * * * (0.027)
(continued)
Fringe benefits

results (1996-2004
Pooled cross-section

NLSY waves)
631

Table II.
 IJM
31,6

632

Table II.
Pooled cross section waves 1996-2004

Log weekly hours worked 20.047 (0.040) 2 0.037 (0.040) 0.015 (0.041)
Tenure in weeks divided by 1,000 20.642 * * * (0.114) 2 0.622 * * * (0.114) 20.575 * * * (0.115)
(Tenure in weeks divided by 1,000) squared 0.428 * * * (0.111) 0.416 * * * (0.111) 0.383 * * * (0.111)
Recent promotion 0.226 * * * (0.026) 0.224 * * * (0.026) 0.226 * * * (0.026)
Union member 20.03 (0.034) 2 0.019 (0.035) 20.006 (0.035)
Public employer 0.242 * * * (0.042) 0.256 * * * (0.043) 0.270 * * * (0.043)
Medium firm 20.156 * * * (0.029) 2 0.149 * * * (0.029) 20.135 * * * (0.029)
Big firm 20.252 * * * (0.030) 2 0.256 * * * (0.030) 20.247 * * * (0.030)
Occupation controls Yes Yes Yes
Industry controls Yes Yes Yes
Cut point 1 21.891 * * (0.890) 2 1.897 * * (0.890) 21.760 * * (0.891)
Cut point 2 21.237 (0.891) 2 1.242 (0.891) 21.101 (0.892)
Cut point 3 0.301 (0.891) 0.297 (0.891) 0.444 (0.892)
No. of observations 24,090 24,090 24,090
Notes: Represent statistical significance at the *10, * *5, * * *1 per cent levels respectively; standard errors are clustered by individual identification and
are in parentheses; marginal effects are in brackets; tenure is divided by 1,000 in order to display its coefficient in a more complete fashion
to have is significantly and positively related to job satisfaction. Also, fringe benefit Fringe benefits
quantity dummies located in the second column of Table II tell a convincing story.
The estimated impact of fringe benefits on job satisfaction generally grows both in size
and significance as the number of fringe benefits increases. Although these two
columns do not indicate which particular fringe benefits are important to worker job
satisfaction, they do show that fringe benefits are important as a whole and need to be
included in job satisfaction estimations and not overlooked as an insignificant 633
determining factor of job satisfaction.
The third column considers individual fringe benefits and shows that five out of eight
have a significantly positive impact on job satisfaction with only one, health insurance,
showing a significantly negative coefficient. As mentioned, health insurance may
decrease job satisfaction if the worker’s spouse already has it, effectively lowering the
worker’s wages for a duplicate and unnecessary fringe benefit. Also, health insurance may
create job lock if an employee cannot leave an unsatisfactory job and health insurance is
not available at other, more appealing jobs.
While direction and significance of the estimation coefficients are important, their
magnitudes are equally crucial in determining the complete impact of fringe benefit
provision on job satisfaction. Marginal effects are reported in brackets within Table II for
each fringe benefit variable, representing how much an additional benefit might change
the proportion of workers that report the highest level of job satisfaction. For instance,
if workers change from having no fringe benefits to six fringe benefits, the percentage of
workers reporting the highest level of job satisfaction can be expected to increase by
nearly 12 per cent. Similarly if an employer begins offering a pension or retirement
plan, workers reporting very high job satisfaction may increase by nearly 3 per cent.
The magnitudes reinforce the notion that fringe benefits are indeed important in
determining employee job satisfaction.
Each benefit, such as health insurance, illustrates a distinct explanation for its
individual impact on job satisfaction and offers broad conclusions regarding the type of
worker that may value a particular fringe benefit. In order to examine the prevalence of the
role of fringe benefits, the pooled dataset is split four ways and cross-sections are
estimated for each demographic sub-sample of the dataset. The results are shown in
Table III. First, the sample is split by gender and pooled cross-sections of each sub-sample
are estimated. Both males and females seem to value similar fringe benefits including
flexible work hours, parental leave and employer provided child care. However, only
females significantly value pensions while only males value profit sharing[6].
Next, the impact of fringe benefits on job satisfaction for union members is compared
to non-union members. As a result of collective bargaining, union members generally
have more fringe benefits than non-union members. Specifically, though, non-union
workers seem to value fringe benefits more than union workers. Both groups value
flexible work hours and profit sharing, yet only non-union workers additionally value
pensions, parental leave and employer provided childcare. It is evident that fringe
benefits are not as important in determining the job satisfaction of union workers as they
are for non-union workers. Union workers may take fringe benefit provision for granted
due in part to their bargaining power, effectively reducing the relationship between
fringe benefits and job satisfaction[7].
The sample is then split by marital status indicated at the time of the respondent’s
interview. Flexible work hours, pensions and parental leave are positively
 IJM
31,6

634

Table III.

sub-samples
results of selected
Pooled cross-section
Non-union No children
Females Males Union members members Married Single Children at home at home

Flexible work hours 0.208 * * * (0.031) 0.232 * * * (0.032) 0.126 * * (0.057) 0.248 * * * (0.024) 0.257 * * * (0.028) 0.158 * * * (0.034) 0.235 * * * (0.027) 0.206 * * * (0.037)
Pension or (0.045)
retirement plan 0.113 * * * (0.035) 0.026 (0.037) 0.071 (0.059) 0.067 * * (0.028) 0.074 * * (0.032) 0.068 * (0.040) 0.073 * * (0.031) 0.059
Dental insurance 0.047 (0.049) 0.056 (0.046) 0.088 (0.083) 0.036 (0.036) 0.071 * (0.042) 0.027 (0.053) 0.068 * (0.040) 0.031 (0.057)
Parental leave 0.133 * * * (0.047) 0.114 * * * (0.036) 0.070 (0.071) 0.127 * * * (0.030) 0.124 * * * (0.034) 0.100 * * (0.048) 0.136 * * * (0.034) 0.081 * (0.048)
Employer provided (0.070)
child care 0.090 * (0.052) 0.111 * (0.067) 2 0.057 (0.129) 0.139 * * * (0.042) 0.070 (0.053) 0.176 * * * (0.062) 0.147 * * * (0.051) 0.004
Employer offered (0.057)
vacation days 2 0.035 (0.046) 2 0.015 (0.051) 2 0.169 * (0.091) 0.006 (0.036) 2 0.070 (0.044) 0.020 (0.052) 2 0.066 (0.041) 0.021
Profit sharing 0.033 (0.037) 0.082 * * (0.038) 0.171 * * (0.070) 0.047 * (0.028) 0.068 * * (0.033) 0.041 (0.042) 0.065 * * (0.032) 0.058 (0.046)
Employer provided (0.067)
health insurance 2 0.087 (0.061) 2 0.055 (0.057) 2 0.052 (0.135) 2 0.099 * * (0.043) 2 0.109 * * (0.053) 2 0.029 (0.062) 2 0.078 (0.051) 2 0.095
Normal controls Yes Yes Yes Yes Yes Yes Yes Yes
Occupation controls Yes Yes Yes Yes Yes Yes Yes Yes
Industry controls Yes Yes Yes Yes Yes Yes Yes Yes
Cut point 1 2 1.674 (1.283) 2 1.409 (1.240) 2 3.168 (2.186) 2 1.459 (0.981) 2 1.911 * (1.093) 2 1.784 (1.559) 2 1.398 (1.080) 2 2.910 * (1.649)
Cut point 2 2 1.052 (1.282) 2 0.704 (1.243) 2 2.572 (2.187) 2 0.783 (0.981) 2 1.197 (1.094) 2 1.194 (1.558) 2 0.737 (1.080) 2 2.248 (1.652)
Cut point 3 0.404 (1.282) 0.942 (1.244) 2 0.956 (2.187) 0.755 (0.981) 0.388 (1.094) 0.292 (1.559) 0.823 (1.080) 2 0.717 (1.653)
No. of observations 12,214 11,876 4,212 19,878 15,084 9,006 16,717 7,373
Notes: Represent statistical significance at the *10, * *5, * * *1 per cent levels respectively; standard errors are clustered by individual identification and are in parentheses
and significantly related to job satisfaction for both groups, yet interesting differences Fringe benefits
remain. First, employer provided child care is only important for single workers.
After all, married workers may not need employer provided child care if their spouse
has time to look after the children. Second, employer provided health insurance
is significantly and negatively related to job satisfaction only for married workers.
This may be a result of the wasteful duplication of fringe benefit provision among
spouses and the reduction of wages that results. 635
Finally, the sample is split between workers who have children at home and those
that do not. Fringe benefits are often suited for workers with families so those with
dependents at home are more likely to value fringe benefits than those that do not.
As anticipated, six of the eight fringe benefits positively impact job satisfaction of workers
with children at home while only two significantly impact the job satisfaction of those with
no dependents at home.

5. Robustness checks
The pooled cross-section results may be biased due to unobservable individual
characteristics that are correlated with both job satisfaction and fringe benefits,
generating a potential problem of workers sorting themselves into jobs with particular
characteristics. This possible selection bias has often been controlled for by researchers
in their cross-section estimates. Bender and Heywood (2006), McCausland et al. (2005)
and Bryson et al. (2005) use instruments to control for worker selection into a particular
occupation, sector or performance payment system and find that selection is
significant. Green and Heywood (2008), Cornelissen et al. (2008) and Pouliakas and
Theodossiou (2009) use panel data and a fixed effects estimator to control for worker
sorting. These studies confirm that researchers agree non-random worker sorting into
various workplace characteristics is evident.
Without accounting for worker sorting, the pooled cross-section results of Table II
may be unreliable. Unobservable individual preferences decide, at least in part, the
worker’s job satisfaction but also what fringe benefits workers receive. In order to
discover the true impact of fringe benefits on job satisfaction, we must first hold the
effects of unmeasured individual preferences on job satisfaction fixed and only allow
observable worker and job characteristics including the provision of fringe benefits to
vary. The possibility of worker sorting necessitates the use of a fixed effects estimator
and is only possible by using panel data. As workers move from job to job, their
unobservable characteristics are assumed to remain constant but their fringe benefits
are allowed to vary. Therefore, if worker job satisfaction changes, it is due to changes
only in fringe benefits and other measurable characteristics.
A straightforward translation to a fixed effects estimator is provided by a
dichotomous logit procedure, the conditional logit[8]. This method estimates whether
or not a worker is very satisfied using the NLSY waves as panel data and is based on
the cumulative logistic distribution rather than the cumulative normal distribution.
The logistic and normal distributions are quite similar, thus the logit and probit
estimation results are comparable[9].
In the conditional logit procedure, individual workers are first categorized into
groups by their NLSY identification number across all five waves so that each group,
or individual worker, consists of all the observations attributable to that worker.
The conditional logit procedure then removes from the estimation all of the groups
IJM that do not exhibit a change in their job satisfaction at all across the five waves. These
31,6 individuals do not contribute to the likelihood function and, therefore, have no effect on
the conditional probability of very high job satisfaction. Finally, the conditional logit
removes any independent variables from the regression that do not vary at all over the
five waves of data. These variables exhibit no within-group variation across the waves
and include not only race and gender but also characteristics among workers that are
636 unobservable and assumed to remain constant. The result is a conditional probability
of very high job satisfaction based entirely on observed characteristics of workers and
jobs and not on fixed effects. The conditional logit model is more thoroughly presented
in the Appendix.
The results of the fixed effects estimation are presented in Table IV. The fringe benefit
variable portrayed in the first column remains a positive and significant determinant of
job satisfaction[10]. Also, five individual fringe benefits remain significant and positive
while health insurance remains negatively related to job satisfaction, though it lost its
significance. In addition, most of the fringe benefit dummies reflecting the quantity of
benefits retain their positive and significant relationship with job satisfaction. Thus, the
unobservable individual characteristics of workers do not greatly alter the pooled
cross-section results. These results may also be biased if worker job satisfaction and fringe
benefit provision are simultaneously determined or endogenous. In other words,
unmeasured determinants of job satisfaction might also determine fringe benefits for
employees.
A formal test of endogeneity between fringe benefits and job satisfaction has not
been undertaken. Although not with job satisfaction, fringe benefits such as pensions,
health insurance and paid vacations have been found to be endogenous in wage
regressions and thus result in simultaneity bias in ordinary least squares estimates
( Jensen and Morrisey, 2001). Since wages and job satisfaction are highly related, it is
possible that endogeneity between fringe benefits and wages could raise a similar
simultaneity bias between fringe benefits and job satisfaction. Therefore, a test for
endogeneity should be employed to be certain that a two-stage least squares estimation
is not required to control for the correlation in the error terms that jointly determine
job satisfaction and fringe benefits.
Normally, an instrument is used in a two-stage least squares procedure to correct for
endogeneity between the dependent variable and independent variable of interest.
However, in this case, job satisfaction and fringe benefit variables are categorical rather
than continuous and so it is problematic to employ the same two-stage least squares
approach. Rather, a recursive bivariate probit approach can detect the presence of
endogeneity by testing if the cross-equation correlation between the identification
equation and the job satisfaction equation is significantly different from zero. Consider
the following structural equations explaining job satisfaction and fringe benefits:

JS k ¼ b1 X k þ b2 F k þ 1 k
Fk ¼ a1 X k þ a2 I k þ mk

where k indexes an individual worker, F is the worker’s fringe benefit provision, X is a

vector of independent variables influencing both worker job satisfaction and fringe
benefit provision, I is an instrument that impacts fringe benefit provision but is not
significantly related to job satisfaction and 1 and m are random error terms. If fringe
 Fixed effects estimation

Fringe benefits (0-8) 0.125 * * * (0.016) – –

Flexible work hours – 0.291 * * * (0.050) –
Pension or retirement plan – 0.284 * * * (0.059) –
Dental insurance – 0.158 * (0.082) –
Parental leave – 0.142 * * (0.066) –
Employer provided child care – 0.264 * * * (0.086) –
Employer offered vacation days – 0.048 (0.082) –
Profit sharing – 0.005 (0.062) –
Employer provided health insurance – 20.146 (0.097) –
One fringe benefit – – 0.280 * * (0.123)
Two fringe benefits – – 0.206 (0.133)
Three fringe benefits – – 0.165 (0.131)
Four fringe benefits – – 0.393 * * * (0.125)
Five fringe benefits – – 0.535 * * * (0.123)
Six fringe benefits – – 0.654 * * * (0.125)
Seven fringe benefits – – 0.928 * * * (0.136)
Eight fringe benefits – – 1.192 * * * (0.175)
Normal controls Yes Yes Yes
Occupation controls Yes Yes Yes
Industry controls Yes Yes Yes
No. of observations 13,202 13,202 13,202
Notes: Represent statistical significance at the *10, * *5, * * *1 per cent levels respectively; standard errors are in parentheses
Fringe benefits

Fixed effects results

637

Table IV.
IJM benefits are indeed endogenous to job satisfaction, then the error terms in the structural
31,6 equations should be significantly correlated.
In order to proceed with the recursive bivariate probit approach, job satisfaction is first
transformed into a binary job satisfaction variable that equals one when respondents’ job
satisfaction is very high and zero otherwise. A dummy variable is also generated for each
particular fringe benefit as well as the minimum quantity of fringe benefits each worker
638 claims to have, from one benefit to eight. For instance, the dummy variable representing
four fringe benefits equals one for all individuals claiming to have at least four fringe
benefits and zero for all those with less than four. In all, sixteen generated binary forms of
fringe benefit provision are used in sixteen different recursive bivariate probit estimations.
The chosen instrument (I) used in the recursive bivariate probit procedure is
whether or not the respondent or the respondent’s spouse has additional sources of income
besides their jobs. This may include interest income, dividends, annuity payments or any
other regular source of income. The instrument is positively related to fringe benefits but
not significant in determining job satisfaction. Fringe benefits may be considered a normal
good. That is, a worker with an additional source of income may be inclined to purchase
more fringe benefits from the primary employer. At the same time, the additional source
of income may not have an impact on the worker’s job satisfaction. The strength of
the instrument is tested by considering the F-statistic in the first stage regression of fringe
benefit provision. According to Staiger and Stock (1997), an instrument is weak if the
F-statistic in the first stage regression is ,10 when there is only one endogenous
regressor. Regardless of which measure of fringe benefit provision is used, the F-statistic is
sufficiently large (.10) to ensure that the instrument is not weak.
Estimation of the recursive bivariate probit model explained above and endogeneity
testing follows from Monfardini and Radice (2008). The authors explain that, unless there
is a very large sample, the likelihood ratio test of the cross-equation correlation coefficient
is the best test for endogeneity of the fringe benefit dummy variable. If the correlation
coefficient between the error terms 1 and m is significantly different from zero, then the
fringe benefit dummy is indeed endogenous. This paper’s results show the opposite and
are presented in Table V.
The first column shows the pooled cross-section result when job satisfaction is a
dichotomous dummy and a probit estimation is used only for each individual measure of
benefits as the independent variable of interest. The second column shows the estimated
coefficient of each fringe benefit measure on job satisfaction within the recursive
bivariate probit framework. In this way job satisfaction and each fringe benefit dummy
variable are simultaneously determined within the model, effectively controlling for
potential endogeneity. The third column shows the results of the likelihood ratio test for
the cross-equation correlation equal to zero. If the likelihood ratio test reflects that this
correlation is not significantly different than zero, we cannot accept the hypothesis
of endogeneity. Indeed, all but two of the correlations are far from a significance level of
5 per cent. Only profit sharing reports a cross-equation correlation that is significantly
different from zero at the 5 per cent significance level while the fringe benefit dummy
reflecting workers with at least seven fringe benefits has a cross-equation correlation
significantly different from zero at the 10 per cent level. Although these two show evidence
of endogeneity, their coefficients are still positive and significant, confirming that the
relationship still remains even after controlling for endogeneity.
Worker has at least Probit results Bivariate probit results Likelihood ratio test of r ¼ 0

One benefit 0.169 * * * (0.049) 0.274 * * (0.127) 0.31

Two benefits 0.150 * * * (0.038) 0.221 * * (0.090) 0.185
Three benefits 0.119 * * * (0.035) 0.139 (0.163) 0.681
Four benefits 0.152 * * * (0.032) 0.121 (0.098) 0.935
Five benefits 0.168 * * * (0.027) 0.125 (0.158) 0.928
Six benefits 0.165 * * * (0.026) 20.072 (2 0.34) 0.299
Seven benefits 0.260 * * * (0.034) 0.516 * * * (0.149) 0.064
Eight benefits 0.355 * * * (0.066) 0.358 (0.311) 0.859
Flexible work hours 0.193 * * * (0.017) 20.082 (20.448) 0.55
Pension or retirement plan 0.101 * * * (0.020) 0.174 (0.144) 0.609
Dental insurance 0.076 * * * (0.021) 0.065 (0.130) 0.931
Parental leave 0.126 * * * (0.021) 0.320 * * (0.136) 0.155
Employer provided child care 0.182 * * * (0.030) 20.01 (20.221) 0.382
Employer offered vacation days 0.014 (0.024) 0.131 (0.111) 0.281
Profit sharing 0.111 * * * (0.020) 0.504 * * * (0.163) 0.021
Employer provided health insurance 0.050 * * (0.024) 0.109 (0.089) 0.491
Observations 24,090 24,090 24,090
Notes: Represent statistical significance at the *10, * *5, * * *1 per cent levels respectively; likelihood ratio test of r ¼ 0: Probability r . x 2 Standard
errors are in parentheses
Fringe benefits

Bivariate probit results

and comparison
Table V.
639
IJM An additional concern is the potential for wage endogeneity in the job satisfaction
31,6 equation. The results thus far follow much of the job satisfaction literature and assume
wages are exogenous in job satisfaction estimations, though recent research suggests
that wages should not be treated as such (Pouliakas and Theodossiou, 2009;
McCausland et al., 2005). As a test to see if wages are indeed endogenous, the log wage
variable is removed from the bivariate probit models. The resulting likelihood ratio tests
640 confirm that now six out of the 16 cross-equation correlations suggest the possibility
of endogeneity within this dataset. Yet still the majority of the tests discount the
significance of endogeneity between job satisfaction and fringe benefits[11]. Thus, the
endogenous relationship between fringe benefits and job satisfaction is not a problematic
issue in this NLSY data. The estimates obtained from the pooled cross-section data
should be used to study the impact of fringe benefits on job satisfaction.

6. Conclusion
Fringe benefits make up a significant portion of employer compensation packages in
the USA, but their impact on worker job satisfaction has yet to be given much
attention. Fringe benefits can affect job satisfaction in opposing ways. First of all, since
fringe benefits are generally less taxed than wages, they can be purchased at less cost
through an employer than if bought on the market. Second, fringe benefits are often
desirable pieces of compensation packages and so increase job satisfaction. However,
if spousal compensation packages already provide particular fringe benefits, those
offered by an employer might be considered wasteful. Since past research has shown
that wages often suffer in exchange for fringe benefits, those that are wasteful may
decrease job satisfaction. Finally, workers can feel locked into a particular job or
employer if the fringe benefits they need or desire cannot be gained at a better, more
satisfying job elsewhere. This job-lock can decrease job satisfaction.
In order to estimate the relationship between fringe benefits and job satisfaction,
five waves of the NLSY are used as a pooled dataset and then are split into particular
demographic sub-samples to analyze the extent of the impact of fringe benefits on job
satisfaction. Specific fringe benefits and also the quantity of benefits are revealed as
important determinants of job satisfaction in the entire sample and also in expected
demographic sub-samples. As a robustness check, a fixed effects estimator is used to
control for potentially biasing unobservable individual characteristics captured and not
controlled for in the pooled cross-section estimation. The results generally carry over
from the pooled cross-section estimation to the fixed effects estimation, suggesting
that unobservable worker characteristics do not seem to compromise the significant
relationship between fringe benefits and job satisfaction.
Moreover, fringe benefits may be simultaneously determined with job satisfaction.
If fringe benefits are indeed endogenous, then their estimated impact on job satisfaction
will be biased. Using a recursive bivariate probit approach, this endogenous relationship
is tested using the likelihood ratio test of the cross-equation correlation of the structural
equations for job satisfaction and fringe benefit provision. Most of the correlations
are indeed not significantly different from zero, confirming that endogeneity is not a
problem in this dataset. The pooled cross-section estimates remain as accurate
depictions of the positive impact that fringe benefits have on job satisfaction.
Consequentially, a complete measure of fringe benefits should be included as a
control variable in future job satisfaction estimations. Additionally, studies focusing
on particular fringe benefits may more fully explain the idiosyncratic affects that fringe Fringe benefits
benefits might have on job satisfaction.

Notes
1. The definition of civilians includes all workers in the private non-farm economy, excluding
households and the public sector, excluding federal government employees.
2. While most studies fail to control for fixed effects, some do not. These include Vieira et al.
641
(2005), Donohue and Heywood (2004), Heywood et al. (2002), Benz (2005) and Heywood and
Wei (2006), although some do not report coefficients on fringe benefits after controlling for
fixed effects.
3. Another prominent subject in the literature is the role of profit sharing in determining job
satisfaction. Profit sharing is included in this paper as a fringe benefit since it is not directly
related to an individual’s effort, but as such it is commonly considered a method of
performance pay. Green and Heywood (2008) and Cornelissen et al. (2008) have found that
performance pay, including profit sharing, often increases job satisfaction.
4. This result is even more impressive since all of the survey respondents in the study are union
workers. Union workers are generally more difficult to layoff due to their relative labor
market power. So union workers feeling more risk of layoff from higher pension accruals is a
striking result indeed.
5. Individual NLSY waves display some variation in the strength and significance of fringe
benefits’ relationship with job satisfaction and these results are available from the author.
6. Artz (2008) includes profit sharing as a form of performance pay and shows that
performance pay indeed has an impact on males’ job satisfaction but not females. However,
the author’s primary focus was individual performance pay rather than broad forms such as
profit sharing.
7. In order to see if individual fixed effects are at play here, fixed effects estimation is used for
both sub-samples of workers. For union workers, only pensions are significantly and
positively related to job satisfaction whereas five out of the eight fringe benefits have a
significant and positive impact on job satisfaction for non-union workers. Thus, it is possible
that union workers have individual unobservable characteristics that determine largely
whether or not they value fringe benefits.
8. An alternative to the conditional logit method is a fixed-effects ordered logit technique.
Though less computational, the conditional logit estimation method carries the potential of
losing large amounts of information since the dependent variable is condensed from multiple
categories into only two. This problem was highlighted by Ferrer-i-Carbonell and Frijters
(2004), but Jones and Schurer (2009) use an alternative approach of the conditional logit to
mitigate some of the information loss. Their binary dependent variable uses the mean of
every group’s job satisfaction rather than the highest level as the arbitrary threshold. In this
way, the conditional logit captures more variation in job satisfaction and does not lose as
much information. The results of the conditional logit using the highest level of job
satisfaction as the threshold are very similar to the conditional logit that uses the mean of job
satisfaction as the threshold.
9. The ordered-probit model’s results generally carry over to the ordered-logit model and also
to the dichotomous logit model. These are available from the author to interested readers.
10. The size of the fringe variable coefficient in the first column of Table IV (0.125) is noticeably
larger than the same variable’s coefficient (0.061) in the ordered probit cross-section estimate.
The difference can at least partly be explained by the change in estimation methodology
between the pooled ordered probit estimation and the fixed effects estimation that instead
IJM uses a dichotomous dependent variable and assumes a logistic distribution rather than a
cumulative normal. The difference in coefficient sizes is reduced when a pooled binary logit
31,6 cross-section estimate is used rather than the ordered probit. In the binary logit model, the
coefficient on the same fringe benefit variable is (0.092). The remaining difference, though
smaller, could still be attributed to unobservable effects or also differences in estimation
sample sizes between the cross-section and fixed effects estimations.
642 11. As an additional measure to test the impact of wage endogeneity on the likelihood ratio tests,
the fitted values of log wage are estimated including all the normal controls except for age,
which serves as a convenient instrument that is not significantly related to job satisfaction
or fringe benefits in this particular NLSY dataset but significantly related to wages.
The likelihood ratio tests of the cross-equation correlations of the estimations with fitted
values of log wage are almost identical to those when wages are simply removed from the
bivariate probit models. This implies that although wages may be considered endogenous to
job satisfaction in this dataset, the simultaneity bias is not strong enough to significantly
change the primary result that the pooled cross-section estimation can be used to identify the
significant relationship between fringe benefits and job satisfaction. Many thanks to an
anonymous reviewer who suggested methods of testing for wage endogeneity.

References
Adams, S. (2004), “Employer-provided health insurance and job change”, Contemporary
Economic Policy, Vol. 22 No. 3, pp. 357-69.
Akerlof, G.A., Rose, A.K. and Yellen, J.L. (1988), “Job switching and job satisfaction in the US
labor market”, Brookings Papers on Economic Activity, Vol. 2, pp. 495-582.
Alpert, W.T. (1987), “An analysis of fringe benefits using time-series data”, Applied Economics.,
Vol. 19, pp. 1-16.
Artz, B. (2008), “The role of firm size and performance pay in determining employee job
satisfaction”, Labour: Review of Labour Economics and Industrial Relations, Vol. 22 No. 2,
pp. 315-43.
Baughman, R., DiNardi, D. and Holtz-Eakin, D. (2003), “Productivity and wage effects of
‘family-friendly’ fringe benefits”, International Journal of Manpower, Vol. 24 No. 3,
pp. 247-59.
Bender, K.A. and Heywood, J.S. (2006), “Job satisfaction of the highly educated: the role of gender,
academic tenure and earnings”, Scottish Journal of Political Economy, Vol. 52 No. 2,
pp. 253-79.
Bender, K.A., Donohue, S.M. and Heywood, J.S. (2005), “Job satisfaction and gender segregation”,
Oxford Economic Papers, Vol. 57, pp. 479-96.
Benz, M. (2005), “Not for the profit, but for the satisfaction? Evidence on worker well-being in
non-profit firms”, KYKLOS, Vol. 58 No. 2, pp. 155-76.
Bryson, A., Cappellari, L. and Lucifora, C. (2005), “Why so unhappy? The effects of unionization
on job satisfaction”, Oxford Bulletin of Economics and Statistics, Vol. 72 No. 3, pp. 357-80.
Clegg, C.W. (1983), “Psychology of employee lateness, absence and turnover: a methodological
critique and an empirical study”, Journal of Applied Psychology, Vol. 68, pp. 88-101.
Cornelissen, T., Heywood, J. and Jirjahn, U. (2008), “Performance pay, risk attitudes and job
satisfaction”, German Socio-Economic Panel Study Papers, No. 136, DIW, Berlin.
Dale-Olsen, H. (2006), “Wages, fringe benefits and worker turnover”, Labour Economics, Vol. 13
No. 1, pp. 87-105.
Donohue, S.M. and Heywood, J.S. (2004), “Job satisfaction and gender: an expanded specification Fringe benefits
from the NLSY”, International Journal of Manpower, Vol. 25 No. 2, pp. 211-34.
Ferrer-i-Carbonell, A. and Frijters, P. (2004), “How important is methodology for the estimates of
the determinants of happiness?”, The Economic Journal, Vol. 114, pp. 641-59.
Freeman, R.B. (1978), “Job satisfaction as an economic variable”, The American Economic
Review, Vol. 68 No. 2, pp. 135-41.
Green, C. and Heywood, J. (2008), “Does performance pay increase job satisfaction?”,
643
Economica, Vol. 75 No. 300, pp. 710-28.
Heywood, J.S. and Wei, X. (2006), “Performance pay and job satisfaction”, Journal of Industrial
Relations, Vol. 48 No. 4, pp. 523-40.
Heywood, J.S., Siebert, W.S. and Wei, X. (2002), “Worker sorting and job satisfaction: the case of
union and government jobs”, Industrial & Labor Relations Review, Vol. 55 No. 4,
pp. 595-609.
Jensen, G.A. and Morrisey, M.A. (2001), “Endogenous fringe benefits, compensating wage
differentials and older workers”, International Journal of Health Care and Economics,
Vol. 1, pp. 203-26.
Jones, A. and Schurer, S. (2009), “How does heterogeneity shape the socioeconomic gradient in
health satisfaction”, Journal of Applied Econometrics (December 14, 2009).
Luchak, A.A. and Gellatly, I.R. (2002), “How pension accrual affects job satisfaction”, Journal of
Labor Research, Vol. 23 No. 1, pp. 145-62.
McCausland, W.D., Pouliakas, K. and Theodossiou, I. (2005), “Some are punished and some are
rewarded: a study of the impact of performance pay on job satisfaction”, International
Journal of Manpower, Vol. 26 Nos 7/8, pp. 636-59.
McEvoy, G.M. and Cascio, W.F. (1985), “Strategies for reducing employee turnover: a meta
analysis”, Journal of Applied Psychology, Vol. 70, pp. 342-53.
Maddala, G.S. (1987), “Limited dependent variable models using panel data”, The Journal of
Human Resources, Vol. 22 No. 3, pp. 307-38.
Mangione, T.W. and Quinn, R.P. (1975), “Job satisfaction counter-productive behavior and drug
use at work”, Journal of Applied Psychology, Vol. 60, pp. 114-16.
Monfardini, C. and Radice, R. (2008), “Testing exogeneity in the bivariate probit model: a monte
carlo study”, Oxford Bulletin of Economics and Statistics, Vol. 70 No. 2, pp. 271-82.
Pouliakas, K. and Theodossiou, I. (2009), “Confronting objections to performance pay: a study of
the impact of individual and gain-sharing incentives on the job satisfaction of British
employees”, Scottish Journal of Political Economy, Vol. 56 No. 5, pp. 662-84.
Staiger, D. and Stock, J.H. (1997), “Instrumental variables regression with weak instruments”,
Econometrica, Vol. 65 No. 3, pp. 557-86.
Uppal, S. (2005), “Workplace characteristics and job satisfaction”, International Journal of
Manpower, Vol. 26 No. 4, pp. 336-49.
Vieira, J.C., Menezes, A. and Gabriel, P. (2005), “Low pay, higher pay and job quality: empirical
evidence for Portugal”, Applied Economics Letters, Vol. 12, pp. 505-11.
Weiss, A. (1984), “Determinants of quit behavior”, Journal of Labor Economics, Vol. 2 No. 3,
pp. 371-87.
Woodbury, S. (1983), “Substitution between wage and nonwage benefits”, American Economic
Review, Vol. 73 No. 1, pp. 166-82.
IJM Appendix
This appendix briefly presents the conditional logit model assuming two time periods and
31,6 discusses its extension to more than two periods.
Consider the logit model as:
expðb0 xit þ ai Þ
Probð yit ¼ 1Þ ¼
1 þ expðb0 xit þ ai Þ
644 where ai are the unobservable characteristics that help determine the state ( yit) that
each worker is in.
The conditional logit eliminates the fixed effects thereby isolating how the changes of
the observed characteristics explain changes in the dichotomous job satisfaction variable.
The elimination of fixed effects in a two-period conditional logit model is constructed below.
First, if Syit ¼ 0 or Syit ¼ 2, then the worker never changes states and so contributes
nothing to the likelihood function. These observations are then discarded so only those workers
changing states remain in the model. We are then left with two possible outcomes:
1 expðb0 xi2 þ ai Þ
Probð0; 1Þ ¼ ·
1 þ expðb xi1 þ ai Þ 1 þ expðb0 xi2 þ ai Þ
0

(probability of state yi2 being true)

expðb0 xi1 þ ai Þ 1
Probð1; 0Þ ¼ ·
1 þ expðb0 xi1 þ ai Þ 1 þ expðb0 xi2 þ ai Þ
(probability of state yi1 being true)
Since both states are mutually exclusive:
Probð1; 0Þ exp½b0 ðxi1 2 xi2 Þ

Prob½ð1; 0Þjð1; 0Þorð0; 1Þ
¼ ¼
Probð1; 0Þ þ Probð0; 1Þ 1 þ exp½b0 ðxi1 2 xi2 Þ

and:
Probð0; 1Þ 1
Prob½ð0; 1Þjð1; 0Þorð0; 1Þ
¼ ¼
Probð1; 0Þ þ Probð0; 1Þ 1 þ exp½b0 ðxi1 2 xi2 Þ

Thus, the fixed effects (ai’s) have been cancelled out and so only a standard logit model remains
to be estimated. This can of course be extended to include five periods (as is represented in this
paper) rather than two. Then we must not only consider the case when Syit ¼ 1 but also when
Syit ¼ 1, 2. . .4. Still, since all of the states are mutually exclusive, we can cancel out the ai’s and
leave only a standard logit model remaining. For details see Maddala (1987).

About the author

Benjamin Artz is an Assistant Professor of Economics at the American University of Sharjah,
Sharjah, UAE. Benjamin Artz can be contacted at: bartz@aus.edu

To purchase reprints of this article please e-mail: reprints@emeraldinsight.com

Or visit our web site for further details: www.emeraldinsight.com/reprints