Does Low Job Satisfaction
Lead to Job Mobility?
Nicolai Kristensen
CCP, Aarhus School of Business
Niels Westergård-Nielsen
CCP, Aarhus School of Business
and IZA Bonn
Discussion Paper No. 1026
February 2004
IZA
P.O. Box 7240
53072 Bonn
Germany
Phone: +49-228-3894-0
Fax: +49-228-3894-180
Email: iza@iza.org
This paper can be downloaded without charge at:
http://ssrn.com/abstract=511722
An index to IZA Discussion Papers is located at:
http://www.
An index to IZA Discussion Papers is located at:
http://www.iza.org/publications/dps/
Any opinions expressed here are those of the author(s) and not those of the institute. Research
disseminated by IZA may include views on policy, but the institute itself takes no institutional policy
positions.
The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center
and a place of communication between science, politics and business. IZA is an independent nonprofit
company supported by Deutsche Post World Net. The center is associated with the University of Bonn
and offers a stimulating research environment through its research networks, research support, and
visitors and doctoral programs. IZA engages in (i) original and internationally competitive research in
all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research
results and concepts to the interested public.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be
available on the IZA website (www.iza.org) or directly from the author.

IZA Discussion Paper No. 1026
February 2004
ABSTRACT
Does Low Job Satisfaction Lead to Job Mobility?
This paper seeks to analyse the role of job satisfaction and actual job change behaviour. The
analysis is based on the European Community Household Panel (ECHP) data for
Danish families 1994-2000. The results show that inclusion of job satisfaction, which is a
subjective measure, does improve the ability to predict actual quit behaviour: Low overall job
satisfaction significantly increases the probability of quit. Various job satisfaction domains are
ranked according to their ability to predict quits. Satisfaction with Type of Work is found to be
the most important job characteristic while satisfaction with Job Security is found to be
insignificant. These results hold across age, gender and education sub-groups and are
opposed to results for UK, where job security is found to be the most important job domain.
This discrepancy between UK and Denmark might be due to differences in unemployment
insurance benefits and indicates that there are “invisible” benefits inherited in the welfare
state insurance system because employees in Denmark don’t worry about job security.
JEL Classification: J28, J30, J60, M50
Keywords: job satisfaction, quits, personnel economics
Corresponding author:
Niels Westergård-Nielsen
Department of Economics
Aarhus School of Business
Silkeborgvej 2
8000 Aarhus C
Denmark
Email: nwn@asb.dk
1. Introduction
Given the demographic evolution in Denmark and many other OECD countries firms
are likely to experience an increased competition in attracting qualified personnel in the
coming years. Furthermore, firms often invest in training and education for their
employees and the rate of return to such investments crucially depends on retaining the
trained employee after the training period. It is therefore important for firms to gain a
good understanding of what determines quits in order to enable the firms to retain their
employees longer.
A large literature in economics has considered determinants of labour market mobility.
However, very few studies have analysed the importance of job satisfaction for quits.
Freeman, in his seminal paper from 1978, is the first to analyse the connection between
quit and job satisfaction. His analysis is based on panel data from two different US
sources. More recent studies which use British or German panel data and link realised
quits to previous levels of job satisfaction include Clark (2001); Clark et al. (1998); and
Lévy-Garboua et al. (2001). Ward and Sloane (2000) apply British cross section data
together with work histories of the respondents. The general finding is that workers
reporting dissatisfaction with their jobs are statistically more likely to quit than those
with higher levels of job satisfaction.
Shields and Price (2002) and Appelbaum et al. (2003) look at intentions to quit and how
these relate to job satisfaction. Shields and Price carry out a case study of nurses in
Britain while Appelbaum et al. analyse low-wage, low-skilled workers in US hospitals.
Both studies find that intentions to quit strongly correlate with job dissatisfaction.
This paper is the first to analyse the effect of job satisfaction on quits in Denmark.
Numerous studies have shown that employees in Denmark have the highest job
satisfaction in Europe (Kristensen and Westergård-Nielsen (2003) and the references
therein). This is somewhat opposed by the fact that average tenure in Denmark is one of
the lowest in the developed world (Bingley and Westergaard-Nielsen, 2003) and makes
it less clear whether job satisfaction necessarily predicts quits in Denmark. The paper
2
also incorporates an alternative measure of job satisfaction, namely whether or not the
employee is searching for a new job. Since this type of information often is a part of
private companies internal “climate surveys” among their employees it is of general
interest to companies to explore whether this measure provides an equally good (or
perhaps even better) prediction of quit behaviour.
Lastly, we explore an approach first suggested by Clark (2001), and seek to establish a
ranking of what matters in a job. This is done by replacing the overall job satisfaction
score by a series of domain job satisfaction variables. The ranking is based on the
ability to predict quits and hence it links what people say with what people do, cf.
section 2 for a discussion of subjective measures such as job satisfaction.
The paper is organised as follows. Endogeneity problems are inherited in this type of
analysis, and we discuss this issue in the next section. Section 3 and 4 describe the data
source and includes some initial descriptive analysis of the data. In section 5 we briefly
present the econometric method applied while results are presented in section 6.
Conclusion and policy recommendations are given in section 7.
2. Identification and endogeneity
What is measured by job satisfaction? Job satisfaction is a subjective measure, and
many economists’ reactions to subjective data, is that such data might be untrustworthy
because they measure “what people say rather than what people do” (Freeman, 1978),
and hence don’t reveal actual behaviour.
If job satisfaction only depends on personal sentiments and standard observed variables,
such as age and education, then no new information would be contained in job
satisfaction and it would only appear as an endogenous variable with no meaningful
impact. But this is not likely to be the case. Supposedly, job satisfaction measures a
multitude of factors some of which intrinsically are subjective and psychological. Other
factors measured will be objective but unobserved. Examples are organisation of work,
physical work conditions and the like, which give the job satisfaction variable
3
systematic exogenous variation. Hence, job satisfaction shall be seen as a proxy for
unobserved objective factors, such as the employee’s evaluation of “the quality of the
match”, and this makes job satisfaction a relevant variable for predicting quits.
If relevant aspects of the work place are left out of the estimation of quit behaviour, and
if these aspects are correlated with the observable factors included in the estimation,
then the parameters of the observed factors will be biased. Consistency will be gained
by including a proxy for the unobservable factors – and job satisfaction could be such a
proxy1.
On the other hand, it is also conceivable that job satisfaction will depend on possible
alternatives, for instance we would expect individuals with good outside job
opportunities to be less satisfied than individuals with no outside job opportunities,
ceteris paribus.
Figure 1 Unemployment and average job satisfaction
0
2
4
6
8
10
12
14
16
18
20
4.4 4.6 4.8 5 5.2 5.4
Average JS
Macro unemployment rate (%)
Source, unemployment: Statistics Denmark, Statistical ten-year review, 2002.
In Figure 1 we have plotted the aggregate unemployment rate for various age groups
and gender across years and calculated the average job satisfaction for each of these
sub-groups. Figure 1 shows that the average job satisfaction increases with the
1 If the assumption about strict independence between the regressors and unobserved individual random
effects holds then the parameter estimates will not be biased. A fixed effect panel logit model would also
ensure unbiased parameter estimates of observed variables. However, efficiency (predictive power) is
likely to be higher if a good proxy for the unobserved components is included and adds to the analysis
because the job satisfaction variable can be given a clear economic interpretation.
4
unemployment rate. This indicates that job satisfaction is endogenous. Hence,
introducing job satisfaction directly in a model for quit behaviour will induce biased
estimates because quit behaviour is negatively correlated with unemployment
(Frederiksen and Westergaard-Nielsen, 2003).
Earnings is also likely to be an endogenous variable in a model with quit as dependent
variable, and direct inclusion of income will in that case also induce bias - not just in
one but in all parameters in the model.
In this paper we follow a pragmatic procedure where we don’t explicitly model the
possible endogeneity. Instead, we analyse how sensitive the main conclusions are to the
specification of the model. This is also the approach taken by other authors in the field
and direct comparison with results from other studies is therefore feasible. Given the
inherit lack of useful instruments an alternative approach would be to estimate a series
of equations jointly.
5
3. Data
We use the data set “Welfare of Danish Families”, which is the Danish equivalent to the
European Community Household Panel (ECHP) data set. For Denmark, data have been
collected from 1994 to 2000 and include personal interviews of about 5,600 individuals
in 19942. These constitute a random sample of the population aged 16 years or more.
Hence, pensioners, students, unemployed and others out of the labour force are also
included in the sample. Despite attempts to supplement the sample with new individuals
from 1995 and onwards the sample suffers increasingly from sample attrition. The
sample size in 2000 includes only about 3,200 individuals.
The data set is very rich and includes detailed information about the individuals
socioeconomic background and family background, employment and unemployment
history, education and training as well as detailed information on the individuals own
perception of satisfaction with main activity, i.e. job satisfaction if employed.
Furthermore, all respondents, including current full-time employees, were asked
whether they search for a new job or not.
Separations are identified as individuals who report a start date for their current job
which falls in-between the last interview and the current interview. For 1994 separations
are identified as new jobs which started within the last year. Quits are identified as
separations reported as voluntary and which are not promotions within the same firm.
Clearly, there is a risk that a respondent may report a separation to be voluntary even
though he/she actually was laid off – hence we should expect some bias due to this
potential reporting error.
With this definition of quit it is clear that the causality runs from job satisfaction, or
changes in job satisfaction, to the decision of whether to quit or not.
2 After cleaning the data.
6
4. Descriptive analysis
Job satisfaction is generally found to be high in Denmark. Eurobarometer (1997) finds
Danes to be the most satisfied workers among all the 16 European nations included in
their survey. The ECHP data confirm that job satisfaction indeed is very high.
Figure 2 Overall Job Satisfaction among employees, by year and gender
0 .5 0 .5
0 2 4 6 0 2 4 6
1994, males 1994, females
2000, males 2000, females
Density
JobSatisfaction
Graphs by year and female
Note: The questions on overall job satisfaction reads: “How satisfied are you with your work or other
main activity?” 1 “Not satisfied at all”…..6 “Fully satisfied”.
Figure 2 show that on a scale from 1 to 6 where 6 is “fully satisfied” very few
employees rank their job satisfaction below 4. Furthermore, there does not appear to be
any major gender difference in job satisfaction while there seems to be a decline in the
level of satisfaction from 1994 to 2000. On the other hand less than 2 percent of the
respondents in 2000 rank their job satisfaction below 3 while in 1994 about 4 percent
rank their job satisfaction below 3.
Looking at how many search for a new job, how many actually quit, and the extent to
which these two interact we observe that about 67 percent of those who quit did not
7
answer that they searched for a new job in the period before they actually quit, cf. Table
1.
Table 1 Actual quits and search
no yes total
search
no 13,693 432 14,125
89.6% 67.2% 88.7%
yes 1,585 211 1,796
10.4% 32.8% 11.3%
total 15,278 643 15,921
100.0% 100.0% 100.0%
Quit
It is somewhat surprising that such a large share of those who quit did not reply that they
searched. This indicates that information on whether the employee search for a new job
or not is a poor predictor of quit behaviour. Since job satisfaction is expected to be
highly correlated with search this could also indicate that job satisfaction conveys little
information about quit behaviour. In order to see whether this is the case or not we first
seek to characterise individuals who quit and/or individuals who indicate they search for
another job compared to individuals who don’t search or quit, cf. Table 2.
Standard deviations (not shown) are large due to the low number of observations in
columns 3-5 so not too much emphasis should be put on the mean values in Table 2.
However, three main observations can be made: One, there seems to be a strong linkage
between search and job satisfaction as well as search and changes in job satisfaction.
Changes in job satisfaction are very small for respondents who do not search (columns
1-3) while average job satisfaction has decreased quite a lot for those who search, in
particular for respondents who later quit (column 5). It is therefore of interest to see
whether job satisfaction and search convey the same information with respect to quits,
i.e. whether they are equally good as predictors of quit behaviour.
8
Table 2 Mean values of selected variables, by search and quit behaviour
1 2 3 4 5
Variable All
no search,
no quit
no search,
quit
search, no
quit
search,
quit
Subjective satisfaction
overall job satisfaction 5.0 5.0 4.9 4.3 4.1
earnings satisfaction 4.4 4.4 4.3 3.9 3.9
security satisfaction 4.8 4.9 4.8 4.1 4.5
typework satisfaction 4.9 5.0 4.8 4.3 4.1
work hours satisfaction 4.9 4.9 4.6 4.6 4.7
work time satisfaction 5.0 5.1 4.8 4.7 4.5
work environment satisfaction 4.8 4.9 4.8 4.3 4.3
distance to work satisfaction 5.0 5.0 4.8 4.7 4.6
Changes in satisfaction
overall job satisfaction -0.05 -0.01 -0.14 -0.40 -0.56
earnings satisfaction -0.01 0.01 -0.03 -0.14 -0.48
security satisfaction 0.03 0.07 -0.03 -0.31 -0.20
typework satisfaction -0.04 0.01 -0.04 -0.38 -0.44
work hours satisfaction -0.02 0.00 -0.23 -0.11 0.03
work time satisfaction -0.03 -0.01 -0.13 -0.15 -0.14
work environment satisfaction -0.04 -0.01 0.03 -0.30 -0.40
distance to work satisfaction -0.02 -0.02 0.10 -0.12 0.06
Other characteristica
public sector 0.41 0.42 0.33 0.39 0.36
wage 12,650 12,748 12,582 11,954 11,665
tenure 7.9 8.4 3.1 4.9 3.8
health (ranked 1-5) 1.6 1.6 1.5 1.6 1.4
hours per week 37.4 37.4 38.0 36.9 37.7
age 40.6 41.4 33.2 36.2 33.2
# obs 15,921 13,693 432 1,585 211
Two, respondents who quit (columns 3 and 5) seem to have some characteristics
common: on average they are young, healthy, have a low level of tenure and are more
prone to work in the private sector. At the same time they seem to differ with respect to
job satisfaction and changes therein, and this raises the question as to whether job
satisfaction or changes in job satisfaction are good predictors of quits in Denmark.
Three, among the domain characteristics, Type of Work seems to be the variable that
mirror overall job satisfaction the most. Earnings Satisfaction is generally at a lower
level than all the other satisfaction domains.
As mentioned above, the standard errors are quite large and the descriptive statistics are
therefore not significant. We therefore proceed to multivariate analysis.
9
5. Econometric analysis
In this section we briefly describe the model, which is used to estimate determinants for
whether an individual choose to quit or not.
The data set applied includes employees only, i.e. self-employed are not included. Data
are right censored since we don’t necessarily observe the end of the employment spell
during the 7 years of observations. Furthermore, we allow individuals to have more than
one spell of employment. This rests on the assumption that the individual random
effects included in the model capture the higher propensity to quit, which these
individuals seem to have.
At each point in time a person is faced with a binary choice of whether to quit or not.
Applying a logit model to estimate this binary choice the probability of quit can be
written as
P(quit=1 | Xit, ci) = Λ(Xitβ+ ci), t = 1,…,T
Where i is an index across individuals; ci is an unobserved individual specific random
effect; Xit is a vector including observable variables for person i in period t; and β is a
vector of parameters. Λ signifies the logit model, i.e.
it i
it i
it i
exp(X ß+ c )
(X ß+ c )
1 exp(X ß+ c )
Λ ≡
+
Assuming the individual random effects follow a normal distribution, i.e.
(0, 2 ) i c c 􀀀N σ , allow the unobserved effects to be integrated out. The likelihood
contribution of an individual i that stays in the same job for T periods can be written as3
2/ 2 2 1
1
1
Pr( | ,..., ) Pr( 0) Pr( )
2
i c i
i i i
c T
iT i iT it T i
c t
quit X X e quit quit dc
σ
πσ
∞ − −
−∞ =
 
=  = × 
 
∫ Π
where
Pr( 0) 1
it 1 exp( )
it i
quit
X β c
= =
+ +
and
3 See Wooldridge, 2002, ch. 15 for details.
10
exp( )
Pr( )
i 1 exp( )
it i
T
it i
X c
quit
X c
β
β
+
=
+ +
if the individual quits in period T
Pr( ) 1
Ti 1 exp( )
it i
quit
X β c
=
+ +
if the observation is right censored
The functional form of the likelihood of the random effects logit model implies that the
random effects are assumed to be independent of the observed explanatory variables
(i.e. the likelihood is assumed to be separable).
6. Estimation results
Parameter estimates from random effects logit models are given in Table 3, overleaf.
The variable of main interest is job satisfaction, which is seen to be very significantly
negative, i.e. the higher job satisfaction the lower the probability of quit. Various
different specifications of this variable have been tested, including the simple job
satisfaction score (assuming cardinality), dummies for various satisfaction levels and
transformation of the variable into a z-score.4 Here we use the z-score since it conveys
the information in just one parameter– similar results are obtained from other
specifications.
The other parameter estimates reveal that age and tenure, as expected, are very strong
predictors of quit behavior. The youngest age group has a much higher probability of quit
than the reference group of 30-49 year olds while the older age group has a significantly
lower probability of quit. The longer the tenure the lower the probability of quit. Firms
with 100 employees or more are likely to have an internal labour market within the firm
and this is probably the reason why the probability of quit is lower for large firms. People
who own their home as opposed to being tenants are expected to be less mobile and this
results in a significantly lower probability for owners to quit their job.
4 The z-score transformation amounts to a rescaling to a unit normal distribution (or some other
symmetric probability distribution). This is done by subtracting the mean job satisfaction from any given
response and dividing by the standard deviation. This procedure yields a continuous variable (Freeman,
1978).
11
Table 3 Quit and overall job satisfaction, probability model
Quit Coef P-value Coef P-value Coef P-value Coef P-value
Job satisfaction -0.625 *** 0.000 -0.505 *** 0.000 -0.514 *** 0.000 -0.525 *** 0.000
Age below 30 0.504 *** 0.000 0.499 *** 0.000 0.566 *** 0.002
Age above 49 -1.051 *** 0.000 -1.031 *** 0.000 -0.992 *** 0.003
Cohabitant/married 0.118 0.304 0.112 0.453 0.150 0.423
Female -0.211 ** 0.041
Number of Children -0.053 0.446 -0.177 * 0.071 0.086 0.386
Children under 12 (dummy) -0.237 0.112 -0.458 ** 0.025 0.025 0.908
Ln(wage) 0.159 0.438 0.220 0.375 0.037 0.922
Ln(hours per week) 0.013 0.962 0.048 0.903 0.086 0.844
Primary/lower secondary education -0.122 0.359 -0.015 0.928 -0.411 0.104
More than secondary education 0.165 0.154 0.058 0.724 0.252 0.133
Public sector (dummy) -0.061 0.580 -0.225 0.173 0.113 0.477
Manager/Professional 0.120 0.366 0.173 0.339 0.101 0.615
Blue collar 0.096 0.402 0.043 0.786 0.149 0.398
Firm size below 20 employees -0.024 0.831 -0.122 0.412 0.156 0.385
Firm size 100 employees or more -0.225 * 0.053 -0.128 0.384 -0.357 * 0.078
Tenure 0-1 year 1.108 *** 0.000 1.288 *** 0.000 0.822 *** 0.000
Tenure 1-4 years 0.491 *** 0.000 0.649 *** 0.000 0.259 0.160
Tenure 11-15 years -0.830 *** 0.000 -0.870 *** 0.006 -0.783 ** 0.012
Tenure 16 years or more -1.007 *** 0.000 -1.276 *** 0.002 -0.883 *** 0.007
Owner of residence -0.194 * 0.057 -0.259 ** 0.048 -0.121 0.463
Permanent contract -0.091 0.528 -0.248 0.171 0.117 0.631
Year dummies No Yes Yes Yes
Constant Yes Yes Yes Yes
Number of individuals 3754 3754 1934 1818
Number of observation 12478 12478 6476 6002
Log likelihood -2236.02 -2046.32 -1140.65 -892.22
Note: * significance at the 10% level; ** significance at the 5% level; *** significance at the 1% level.
All All Men Women
12
Finally, females are seen to be significantly less prone to quit. These other results are in
accordance with results in Frederiksen and Westergaard-Nielsen (2003).
Contrary to results from studies on British and German data we find no significant effect
from wages on the probability of quit. This is most likely because the Danish net wage
distribution is very equally distributed. This does not necessarily mean that wages don’t
affect quit behavior in Denmark. The relevant parameter might be the relative wage
compared to the wage level in the same firm rather than the level of the wage as such.
Bingley and Westergaard-Nielsen (2004) show that the relative wage difference between
the current wage and a potential wage is important for the mobility decision.
Unfortunately, the data at hand do not allow for a good estimation of the relevant
comparison wage.
The number of children and the dummy variable for small children (under 12 years old)
are insignificant for the overall regression including all individuals. Looking at the
gender specific estimates reveals that the probability of quit is significantly lower for
men from a household with children than with no children. The same stabilizing effect
from children is not seen among women. Another gender difference is that the firm size
effect only adheres to women while it is insignificant for men. Apart from these two
differences the estimates for men and women are generally very similar.
Following Clark (2001) the overall job satisfaction variable was substituted with
various satisfaction variables for various job domains. By comparing the log-likelihood
value we are able to rank which satisfaction parameter yields the highest likelihood and
hence reveals most about the quit probability. In this manner we obtain a ranking of
what matters in a job.
Overall job satisfaction should encompass all aspects of a job and one should therefore
expect this variable to be a better predictor of quits than the various domain satisfaction
variables. This is also the case when all individuals are included in the estimation as
well as when the model is estimated across sub-groups of individuals - with a few
exceptions (Appendix, Table 7).
13
Overall, the most important job domain is found to be satisfaction with Type of Work.
This is also the most important job domain for all sub-groups (gender, age and
education) except the lowest educated, cf. Table 4. Satisfaction with Earnings is
generally found to be the second most important job domain despite the insignificant
wage parameter. Since wages and hours of work are among the job domains and at the
same time included as explanatory variables we re-run the models without wage and
hours worked (Appendix, Table 6). This only has a very minor impact on the
satisfaction parameter estimates and the log-likelihood values but it does mean that the
earnings domain for the regression including all individuals becomes marginally
smaller than the domain Working Time. For the lowest skilled the Work Environment
is ranked highest (even above overall job satisfaction).5
Table 4 Ranking of job satisfaction domains, by gender, age and education
Most important
Second most
important
UK, most
important
All Type of work Earnings Security
Men Type of work Working time Security
Women Type of work Earnings Initiative
Age below 30 Type of work Working hours Initiative
Age 30 and above Type of work Earnings Pay
Work itself
Job security
More than secondary education Type of work Working hours na
Upper secondary education Type of work Earnings na
Lower secondary eduation or primary Environment Earnings na
Note: “Most important” after overall job satisfaction with the exception of More than secondary
education and Lower secondary education or primary where “most important” is more important than
overall job satisfaction. For UK, people aged 30 are included in the youngest age category, i.e. age 30
and below. Security and Distance are generally found to be insignificant (for Distance with the exception
of Upper secondary education).
The ranking of the most important job domain based on UK data (Clark, 2001) is
presented in the right-most column of Table 4. In UK, security satisfaction is generally
5 These results generally confirm results found by the Danish Ministry of Finance (2000).
14
found to be the most important satisfaction domain. Why is security an issue in the UK
and not in Denmark, where security satisfaction is insignificant for all sub-groups? This
difference might be due to a lower level of unemployment insurance benefits in UK
compared to DK. If this is the case it means that there is an “invisible” benefit inherited
in the welfare state insurance system because employees in Denmark don’t worry about
job security.
Next we turn to comparison of search, job satisfaction and changes in job satisfaction as
predictors for quit. A priori it is not clear which variable should be the best predictor.
The results presented in Table 5 reveal that Search yields the highest log-likelihood value
and likelihood ratio tests reveals that it is a significantly better predictor of quit behavior
than job satisfaction. Changes in job satisfaction perform poorest. One implication of this
is that a question about search should be included in companies “climate surveys” since it
is closely linked to actual behavior.
Table 5 Comparison of alternative measures of satisfaction
Job
Satisfaction
Change in Job
Satisfaction Search
Parameter -0.547 -0.261 1.022
P-Value 0.000 0.000 0.000
Log-likelihood -1580.7 -1592.9 -1565.9
Note: Number of observations: 9,666; Number of individuals: 3,251.
7. Discussion and Conclusions
This paper is the first to study the link between job mobility and job satisfaction on
Danish data. The main results found are threefold:
One, the results have confirmed that subjective data on job satisfaction are highly useful
and convey a lot of information about unobserved job characteristics which matter in
relation to the decision of whether to quit or not. Hence, inclusion of job satisfaction as a
proxy for the unobserved factors significantly improves the prediction of quit behavior.
This is of importance for the HRM policy of a firm that wants to retain workers.
15
Two, ranking of more explicit job domains reveals that Type of Work is the most
important job feature in Denmark and that Earnings are only second in importance. This
does not preclude that earnings play an important role when an outside job offer is
considered. This is opposed to the UK where Security is found to be the most important
job satisfaction domain – while it is insignificant in Denmark, which indicates that the
relatively high unemployment benefit in Denmark perhaps has the effect that employees
worry less about their job security. Instead they have “the luxury” of mainly focusing on
the quality of the type of job. This also means that companies who seek to diminish the
number of quits simply should focus more on describing the job content and the career
opportunities so that the type of job is as well known as possible before the employment
contract is signed.
Finally, we find that search is a better signal of an upcoming quit than job satisfaction or
changes in job satisfaction. That search is a strong predictor of quits is hardly surprising.
Nevertheless, it indicates that questions about intentions to quit or whether an employee
search for another job might add valuable information to the traditional “climate surveys”
– which for some companies already include these types of questions.
There are several relevant extensions to the analysis in this paper. It is technically
feasible to link the survey data applied in this paper with data from Statistics Denmark’s
registers. This would greatly improve the opportunities for calculating comparison wage
and would increase the amount of information particularly prior to 1994. Another
relevant extension would be do model quit and job satisfaction (or income)
simultaneously in order to circumvent some of the potential endogeneity problems.
16
8. Appendix
Table 6 Comparison of models with and without wage and hours of work per week, parameter estimates and log-likelihood
values
Overall Earnings Security
Type of
work
Working
hours
Working
time
Environment
Distance
All, controlling for wages and hours
Satisfaction coefficient -0.505 -0.134 0.009 -0.230 -0.096 -0.124 -0.112 -0.042
P-value 0.000 0.000 0.785 0.000 0.005 0.000 0.002 0.176
Log-likelihood -2046.3 -2061.3 -2069.0 -2051.1 -2065.3 -2062.1 -2064.3 -2068.2
All, Not controlling for wages and hours
Satisfaction coefficient -0.501 -0.126 0.009 -0.227 -0.095 -0.123 -0.111 -0.043
P-value 0.000 0.000 0.789 0.000 0.006 0.000 0.002 0.166
Log-likelihood -2046.7 -2062.5 -2069.2 -2051.6 -2065.5 -2062.3 -2064.5 -2068.2
Type of job satisfaction
17
Table 7 Quits and job satisfaction domain, by gender, age and education
Overall Earnings Security
Type of
work
Working
hours
Working
time
Environment
Distance
All
Satisfaction coefficient -0.505 -0.134 0.009 -0.230 -0.096 -0.124 -0.112 -0.042
P-value 0.000 0.000 0.785 0.000 0.005 0.000 0.002 0.176
Log-likelihood -2046.3 -2061.3 -2069.0 -2051.1 -2065.3 -2062.1 -2064.3 -2068.2
Number of observations = 12478 Number of individuals = 3754
Men
Satisfaction coefficient -0.280 -0.097 -0.023 -0.234 -0.119 -0.159 -0.062 -0.026
P-value 0.000 0.043 0.581 0.0 0.0 0.0 0.2 0.5
Log-likelihood -1140.7 -1151.6 -1153.5 -1143.3 -1150.5 -1147.1 -1152.8 -1153.4
Number of observations = 6476 Number of individuals = 1936
Women
Satisfaction coefficient -0.286 -0.183 0.051 -0.236 -0.071 -0.077 -0.175 -0.069
P-value 0.000 0.000 0.343 0.000 0.182 0.130 0.001 0.163
Log-likelihood -892.2 -896.8 -902.4 -894.7 -902.0 -901.7 -897.8 -901.9
Number of observations = 6002 Number of individuals = 1818
Age below 30
Satisfaction coefficient -0.256 -0.052 -0.024 -0.211 -0.090 -0.065 -0.131 -0.064
P-value 0.000 0.368 0.650 0.000 0.102 0.211 0.021 0.199
Log-likelihood -677.3 -684.9 -685.2 679.1 -684.0 -684.6 -682.8 -684.5
Number of observations = 2189 Number of individuals = 1055
Age 30 or above
Satisfaction coefficient -0.296 -0.182 0.023 -0.245 -0.099 -0.163 -0.096 -0.028
P-value 0.000 0.000 0.586 0.000 0.029 0.000 0.038 0.482
Log-likelihood -1359.0 -1365.9 -1374.0 -1362.4 -1371.9 -1367.1 -1372.1 -1373.9
Number of observations = 10289 Number of individuals = 2962
More than secondary education
Satisfaction coefficient -0.522 -0.061 0.032 -0.291 -0.120 -0.089 -0.106 0.060
P-value 0.000 0.294 0.589 0.000 0.046 0.121 0.085 0.249
Log-likelihood -779.0 -786.2 -786.6 -777.2 -784.8 -785.5 -785.3 -786.0
Number of observations = 4627 Number of individuals = 1490
Upper secondary education
Satisfaction coefficient -0.319 -0.176 0.024 -0.227 -0.140 -0.130 -0.079 -0.128
P-value 0.000 0.001 0.624 0.000 0.004 0.005 0.138 0.003
Log-likelihood -936.5 -945.7 -951.2 -943.0 -947.4 -947.6 -950.3 -947.2
Number of observations = 5599 Number of individuals = 2091
Lower secondary or primary education
Satisfaction coefficient -0.239 -0.232 -0.077 -0.208 0.093 -0.130 -0.243 0.018
P-value 0.027 0.019 0.346 0.036 0.354 0.125 0.008 0.843
Log-likelihood -306.8 -306.4 -308.8 -307.1 -308.8 -308.1 -305.7 -309.2
Number of observations = 2252 Number of individuals = 898
Type of job satisfaction
18
9. References
Appelbaum, E., Berg, P, Frost, A., and Preuss, G. (2003): “The effects of restructuring
on Low-Wage, Low-Skilled Workers in U.S. Hospitals”, Ch. 3 (pp. 77-117) in “Low-
Wage America: How Employers Are Reshaping Opportunity in the Workplace”, (Eds):
Eileen Appelbaum, Annette Bernhardt, Richard J. Murnane. The Russel Sage
Foundation, 2003.
Bingley, P. and Westergaard-Nielsen, N. (2003): “Returns to tenure, firm-specific
human capital and worker heterogeneity”, International Journal of Manpower, Vol. 24,
No. 7, 2003.
Bingley, P. and Westergaard-Nielsen, N. (2004): “Tenure, career making and pay over
the life cycle – an options value model”, Working Paper (forthcoming), Center for
Corporate Performance, Aarhus.
Clark, A. E. (2001): “What really matters in a job? Hedonic measurement using quit
data”, Labour Economics 8 (2001), pp. 223-242.
Clark, A.E., Georgeellis, Y. ans Sanfey, P. (1998): “Job Satisfaction, wage changes and
quits: evidence from Germany”, Research in Labor Economics, Vol. 17, pp. 95-121.
Eurobarometer, (1997): “Employment, unemployment and the quality of life – The employment
on Europe survey 1996”, Report prepared for the European Commission. Available at
http://europa.eu.int/comm/public_opinion/archives/eb/ebs_098_en.pdf.
Frederiksen, A. and Westergaard-Nielsen, N. (2003): “Where did they go?” Discussion
Paper, IZA, Bonn.
Freeman, R.B. (1978): “Job Satisfaction as an Economic Variable”, American
Economic Review, Vol. 68, No. 2, pp. 135-141.
Kristensen, N. and Westergård-Nielsen, N. (2003): “A survey of job satisfaction and
subjective well-being studies for Denmark”, Manuscript, Center for Corporate
Performance, Aarhus.
Lévy-Garboua, L., Montmarquette, C. and Simonnet, V. (2001): “Job Satisfaction and
Quits: Theory and Evidence from the German Socioeconomic panel”, TEAM Working
Paper, 27:2001, Available at http://team.univ-paris1.fr/
Minstry of Finance (2000) “Motivation in the Danish Central Government Sector –
Impetus for growth and innovation”, Report by Personalestyrelsen, MoF, Denmark.
Available at http://www.perst.dk/db/filarkiv/4327/MotivationUK.pdf
Shields, M. A. and Price, S.W. (2002): “Racial Harassment, Job Satisfaction and
Intentions to Quit: Evidence from the British Nursing Profession”, Economica (2002),
Vol. 69, pp. 295-326.
19