job satisfaction, mobility decisions and wage gains by gender
DESCRIPTION
HR fieldTRANSCRIPT
Job Satisfaction, Mobility Decisions and Wage Gains by Gender
Theodora Xenogiani
LSE, CEP
October 2003
Abstract
Workers tend to change jobs several times during their working life, generally towards better
career prospects. Most individuals face at some point of their working lives, the choice whether
to remain in the same job or change to a di¤erent job, in a di¤erent …rm etc. Their decision is
determined by the expected bene…ts of a job change (both monetary and non pecuniary payo¤s)
and the cost of moving.
If there are di¤erences in tastes as well as cost functions between men and women and women
are more constrained by household and family responsibilities, this should have an impact on their
job mobility behaviour. Furthermore, if women are less motivated by pecuniary aspects of the job
than men, then this could partly explain their lower returns to job mobility found in the literature.
Fist we investigate the determinants of job quitting for men and women distinguishing between
pecuniary and non pecuniary aspects of the job. Second we estimate the returns to job mobility by
gender. Although the link between the two has been extensively investigated in theory, there is very
limited empirical evidence, especially when it comes to gender issues.
We use the British Household Panel Data (BHPS) to undertake this investigation. The special
feature of the BHPS is that it permits a distinction between voluntary and involuntary job separa-
tions. It also contains a large number of variables for job attributes and personal characteristics.
1
JEL: J60, J63, J28, J16
Keywords: Job mobility, wage growth, gender, job satisfaction
2
1 Introduction
Workers tend to change jobs voluntarily several times during their working life, generally
towards better career prospects. Most individuals face at some point of their working lives the
choice whether to remain in the same job or change to a di¤erent job, in a di¤erent …rm etc.
This decision is determined by the expected bene…ts of a job change (monetary payo¤s and non
pecuniary payo¤s) and the cost of moving. It depends crucially on the type and quantity of
their human capital and the degree of satisfaction in their current job. Each individual chooses
an optimal career path depending on her own preferences as expressed by the utility she derives
from every possible outcome and her characteristics, especially education and human capital
investment1 . If the expected bene…ts in the current job are greater than those in potential
alternative jobs, the worker will decide not to move. This expected value is partly formed
according to workers’ past experience which determines their valuation and expectations. In
this paper we argue that women and men make di¤erent job mobility choices and thus their
career paths di¤er. These di¤erences may steam both from heterogeneity in preferences (tastes)
and the cost functions.
In this project we use British Household Panel Data (BHPS) to investigate …rst the deter-
minants of job mobility, separately for men and women. Moreover we look at job satisfaction
in order to gain a better understanding of the di¤erences in job preferences between men and
women. Second we are interested in the consequences of job mobility in terms of wage gains. The
special feature of the BHPS is that it permits a distinction between voluntary and involuntary
job separations. It also contains information on job attributes. In particular the BHPS dis-
criminates between di¤erent types of training i.e. general further education, job related training
and training undertaken to facilitate the transition to a new job. Moreover it provides informa-
tion on other job attributes such as distance to work, usual hours of work, over time and rich
information on overall job satisfaction and detailed aspects of job satisfaction.
In the …rst part we look at the factors determining job satisfaction. We want to investigate
whether these parameters di¤er between men and women. One would expect them to be the same
if tastes and preferences were the same for the two genders. The same holds for job constraints
1 Preferences may have both a direct and an indirect e¤ect on career decisions. The indirect e¤ect operates viaeducational choices and previous working experience (work choice decisions in the past).
3
and the cost functions. Next we look at the determinants of voluntary job mobility, namely
quits. Although we also look at the determinants of promotions, this is not the main focus of
the paper. We distinguish between quits originating for family reasons and those for non family
reasons, but this is not part of the estimation because of the very small number of observations.
The decision to leave a job is determined by a broad set of characteristics which include both
personal characteristics of the worker and features of the job. We are also interested in the e¤ect
of job satisfaction on the decision to leave a job2 . We conduct the analysis separately for men
and women. This is because women are likely to make di¤erent career choices from men because
their time opportunity cost may be di¤erent in terms of family obligations. Opportunity cost
may a¤ect job mobility directly or indirectly through job search intensity. This could imply that
women search for new and better jobs less intensively than men, and thus they move less, or in
a less strategic way than men. In addition the constraints can di¤er between men and women.
In the …nal part we are interested in the consequences of job change in terms of earnings and
thus we attempt to estimate the returns to job mobility. In particular we want to investigate the
existence of di¤erent wage gains from job mobility for men and women. If this is the case, we want
to investigatewhether there is a part of thewage gap which can be attributed to di¤erent mobility
patterns. It has been shown in the literature that women’s decisions are motivated less by money
(pecuniary, monetary payo¤s) than men. In contrast, it is non pecuniary characteristics of the
job that play a more important role for them, such as hours ‡exibility, distance from home etc.
2 Related Literature
Empirical evidence suggest that men and women are di¤erent in terms of job search con-
straints as well as tastes, which partly induce di¤erent mobility patterns. Women are more
constrained by household and family responsibilities and pay is found to play a more important
role in job switching for men than for women. As a consequence, there are also gender di¤erences
in the returns to job mobility. These di¤erences imply di¤erent preferences for the two groups
as well as di¤erences in the degree of labour market attachment.
2 Subjective measures of satisfaction have been widely used in sociology but only recently have economistsstarted using them in economic research (Clark 1997, 1998, 2001, Ward and Sloanne 2000, Oswald, Levy- Garbouaet al 1999 etc).
4
There is evidence (Manning 2003 for the UK, Sicherman 1996, Keith and McWilliams 1999
for the US3) suggesting that although overall job mobility rates are similar for the two genders,
the reasons for this mobility may di¤er (see Booth et al 1999). Women are more likely to leave
their jobs for non market reasons. Job to job changes are likely to be less responsive to wages for
women than for men and thus wage gains from mobility may be higher for men than for women.
Manning (2003) estimates separation elasticities with respect to the wage and he …nds that the
elasticity of separations to non employment is higher for women than men but the elasticity of
separations to other jobs is only weakly higher for men.
In this paper we are particularly interested in the impact of job mobility on wage growth.
The relationship between the two has been investigated in theory but there is very limited
empirical evidence. There is evidence that voluntary job mobility generates wage gains, relative
to non mobility or involuntary job change. (Bartel and Borjas 1981, Mincer 1986). Murphy and
Welch (1990) …nd strong evidence of rapid wage growth at the early stages of workers’ careers.
Topel and Ward (1992) …nd that a substantial part of the wage growth can be attributed to job
change.
Human capital theory4 suggests that job shifts are not necessary to increase earnings, which
can only take place with the accumulation of human capital. Newly recruited workers are more
likely to undertake investment in …rm speci…c human capital, which implies rather ‡at earnings
tenure pro…les after a certain period in the same job. This type of model predicts steeper
earnings pro…les following a job change, although there may be a downward shift of the pro…le
just after the job change. High tenure workers are likely to move to a new …rm which will o¤er
them the option of investment in speci…c human capital and thus experience higher wages and
wage growth than if they remain in their current job.
Other models which again explain the positive cross-section relationship between earnings
and job tenure are those of search and matching. A worker’s productivity is assumed to be
constant while employed in the same job. According to the matching models (Jovanovic 1979,
Burdett 1978), in the beginning the match is characterised by uncertainty regarding the pro-
3 They …nd that although the returns to di¤erent types of job mobility are similar for men and women, theincidence of these moves are di¤erent. Women are found to be more likely to quit for family related reasons andless likely to have a job to job voluntary change. In their 1999 paper they …nd that the returns to job search aresimilar for men and women but women undertake less on the job search.
4 See Neal (1995), Topel (1986, 1991) and Farber (1994).
5
ductivity of the worker. After a certain period of time the true productivity is revealed and the
match can be either extended or ended. Workers will decide to leave their job if they believe
this is not a good match and they hope to switch to one which will be of better quality. If a
good match is found workers will be less likely to leave it. Search models predict an increase in
earnings following a job change. Workers engage in job search and will decide to move to a new
job if this o¤ers a higher wage than his current job. In addition search models suggest that the
probability of a job change falls with labour market experience. First because search intensity
is likely to fall and second because more experienced workers are more likely to be already in
a high paying job, given that they have been engaged in job search for a longer period. This
implies lower job mobility for more experienced workers since mobility costs are higher.
We will analyse the role of job mobility on wage growth and investigate whether this is
di¤erent for men than for women. It may be di¤erent in the same way that their career paths
di¤er signi…cantly. Job preferences di¤erences between men and women may imply di¤erent
trade o¤s between wages and non pecuniary job aspects across genders. For example women
may prefer a job with more ‡exible hours so that they can also take care of their families and
spend time in home work. Some women (especially in the old cohorts) are likely to choose jobs
with these desirable non pecuniary aspects although they may involve lower wages or lower wage
growth. Women are particularly interested in working hours and occupational characteristics
of their job. Discrimination may be one more reason for their di¤erent career paths in the
sense that certain high paying jobs may not be open to women. If women are more likely to
leave the labour market (in short periods) then employers would be less willing to hire them in
positions that require training and are associated with higher earnings. Both the di¤erences in
work tastes and gender discrimination in the labour market may result in lower wage growth for
women which further widens the gender wage gap.
Women do not simply follow di¤erent career paths than men. Even for those with stronger
attachment to the labour market, there are persistent di¤erences in the returns to job mobility.
According to the human capital model, there are gender di¤erences in productivity which lead to
higher wages for men than for women. One possible reason for these di¤erences is suggested to
be the weaker labour force attachment of women, due to household and family responsibilities.
To test the predictions of the human capital approach one can distinguish between actual and
6
potential work experience. Di¤erences in the returns to potential experience can be explained
in terms of gender di¤erences in experience levels. In order to test for common returns to
experience we need data on actual labour market experience (see Light and Ulreta, 1995). The
returns to tenure are supposed to capture the returns to investment in job (…rm) speci…c human
capital. If women are more likely to leave the …rm, then the incentives to investment in job
speci…c human capital are lower than those for men. Thus we would expect women to invest
less in job speci…c human capital. This in turn would imply lower returns to tenure for women.
For the UK there is no strong evidence of that5.
The model of monopsony (see Manning 2003) suggests that labour market transition rates
and the reservation wage of men and women may be important in explaining the gender wage
gap. It has been found that women are more concerned with non pecuniary aspects of their job,
that is work hours (hours ‡exibility: Altonji and Paxson, 1988, 1992), job location etc. In that
sense they might accept or decide to move to a new job which is more ‡exible in terms of hours
worked but o¤ers lower pay.
Keith and McWilliams (1997) …nd that women have fewer involuntary job moves than men
and that the returns to job mobility are the same across gender, if one takes into account the type
of job mobility. Other studies show that the mobility wage gains for women are smaller than for
men. Loprest (1992) argues that job characteristics play an important role in mobility decisions
across women and men which partially explains di¤erences in the returns to job mobility. She
focuses on di¤erences between men’s and women’s patterns of job mobility and wage growth in
their …rst four years of working full time in the labour market. In her paper she tries to explore
the extent to which di¤erences in job mobility, returns to job mobility and the characteristics
of the jobs men and women hold can account for part of the gender wage gap. She …nds that
job mobility and wage growth rates in years of no job change are similar for men and women.
However wage growth for those women who change jobs is half of that of male job changers.
Khan and Griesinger (1989) show that women’s gains may be lower because women usually care
relatively more about job characteristics other than earnings. Light and Ulreta (1995) …nd that
although there are gender di¤erences in mobility patterns the returns to job mobility are very
similar for both genders. They …nd that women have longer and more frequent non working
5 M. Myck and G. Paull, 2001.
7
spells than men in their early careers and thus women tend to require relatively more time to
accumulate a given amount of work experience. Royalton (1998) …nds that less educated women
have lower numbers of job to job (and higher numbers of job to unemployment) transitions
compared to their male counterparts and more educated women have more frequent job to job
(and job to unemployment) transitions.
Keith and McWilliams (1999) estimate the returns to job search, job mobility and the inter-
action between the two for a sample of young men and women, using the National Longitudinal
Survey of Youth. In their paper they suggest that men and women may have di¤erent returns
to job search activities. This can be because they may exert di¤erent search intensity, their
reservation wages may be di¤erent or their wage and o¤er functions may be di¤erent. Given
their household activities, the opportunity cost of search may be higher for women, which will
a¤ect both their reservation wage (lower reservation wage) and their search intensity (via lower
returns to search). They argue that young workers are more likely to search for a new job. This
is because workers in low wage jobs can gain more from search than highly paid workers. If
wages are positively correlated with experience, then younger workers should be more likely to
search.
The estimation of the returns to job mobility is not an easy task. In the existing empirical
literature, researchers have used three di¤erent methods to estimate the returns to job mobility.
The …rst involves the estimation of separate wage growth equations, one for job stayers and
one for job movers (Holmlund 1984). Then the two di¤erent group results are evaluated at
the mean observed characteristics of job movers and the di¤erence between these two mean
predictions is taken as the wage premium associated with a job change. However it is likely that
these estimates of job mobility wage gain underestimate the true value. This is because of the
heterogeneity between the two groups of workers. Wages may be correlated with unobserved
ability and thus if stayers have unobserved characteristics which lead to higher wages, using the
mean characteristics of movers would overstate the earning of movers had they not changed jobs.
It turns out that when this simulated change is compared with the actual change, the returns to
job mobility are underestimated. One way to solve this issue is to control for selection using the
standard Heckman procedure. Stayers are found to get higher wages than those which movers
would have got had they not changed jobs. The hypothetical wage growth derived with this
8
method is then compare to the actual wage growth of movers in order to get an estimate of
mobility wage gains.
The second approach which has been employed in the literature involves the inclusion of
a dummy variable in the wage growth equation (Abbott and Beach 1994, sample of Canadian
women with data from the 1986-1987 Labour Market activity survey). The underlined assump-
tion necessary for this model is that the coe¢cients in the stayers and movers equations are
the same. In the same way as in the previous method, on-the-job wage growth of job stayers
is likely to be higher than that of job movers had they not moved (remember that workers
decide to change jobs if the expected payo¤ in a new job is higher than that in the current job).
Some of the papers in this strand of the literature attempt to …nd a third group which is more
comparable to current period job movers than job stayers and whose wage growth can better
approximate that of job movers if they had not moved. This is suggested to be current period
stayers who change jobs next period6.
As already discussed, estimating log wage change equations introduces two di¤erent sources
of bias. The …rst is the one due to unobserved individual characteristics which may be correlated
with mobility. The second source of bias is due to the endogeneity of job mobility in an earnings
equation arising because shocks to earnings may in‡uence mobility. A worker will decide to
change jobs if the expected value of his current job is lower than that o¤ered in a di¤erent job.
High wages in current job will make job change less likely and thus the error term in a wage
change equation will be correlated with job mobility. To correct for the …rst type of bias we use
individual …xed e¤ects (note that the problem will only be solved if the unobserved individual
component is constant over time). To reduce the possibility of the second type of bias we use
IV.
Finding an instrument for quits in the earnings equation is not an easy task. A possible
instrument is job satisfaction. As we will show in section 5.3, voluntary job change is decreasing
6 Campell (2001), distinguishes between short and long run wage gains from job mobility. In addition, in hispaper he stresses the need to include initial wages among the control variables (This set of explanatory variablesincludes: change in marital status, educational attainment and hours per week worked change). He argues thatthe initial wage in the wage change equation helps to identify any change in the coe¢cient of the explanatoryvariables between the two dates. However the inclusion of this variable in the regression is likely to introduce abias if the initial wage is correlated with the error term. To solve this correlation problem, Campell uses predictedwages instead of actual wages. He estimates a standard wage equation with controls for tenure, age, region andregional unemployment and uses the predicted wage as an instrument for initial wage in the wage growth equation.
9
in the level of job satisfaction. In the mobility equations of section 5.3 we use satisfaction with
job security, with the hours of work and the work itself. In all cases the e¤ect of these variables
is negative and in most cases signi…cantly di¤erent from zero. In addition we do not expect
initial satisfaction to a¤ect wage change conditional on job change. This suggests that measures
of job satisfaction in the initial job could be used as instruments for job mobility in the wage
change equation.
3 The BHPS
3.1 Dataset Description
In this work we use the …rst eleven waves7 of the British Household Panel Survey8 (BHPS) in
combination with the job life history and occupation life history …les9. These have been merged
by the depositor to create a unique …le containing all the information about each respondent’s
work life. The second and last version of this …le was released in 2000 and it also contains
identi…ers which allow links with the household and individual record information provided in
each of the eleven waves. The sample will consist of an unbalanced panel of individuals with
labour market data for at least two years.
The BHPS is being carried out by the ESRC UK Longitudinal Studies Centre in the Institute
for Social and Economic Research, (ISER) at the University of Essex. The main objective of
the survey is to ”improve our understanding of social and economic change at the individual
and household level in Britain, to identify, model and forecast such changes, their causes and
consequences in relation to a range of socio-economic variables”.
The BHPS was designed as an annual survey of each adult (16+) member of a nationally
representative sample of more than 5,000 households, making a total of approximately 10,000
individual interviews. The same individuals are re-interviewed in successive waves and, if they
split-o¤ from original households, all adult members of their new households are also interviewed.
7 Information used from these …les inlude education, training (general and job speci…c), marital status, age,demographics, job satisfaction etc.
8 Reference: ”BHPS, User Manual, Volume A, Introduction, technical summary and appendices”, edited byMarcia Freed Taylor with John Brice, Nick Buck and Elaine Prentice-Lane.
9 From the life and occupation history …les we get the following variables: industry, occupation, sector, employersize, pay, hours of work, managerial duties, boss etc. in each job held in the respondent’s life. In addition reasonwhy left each job, whether this involved employer change. All these provide links to the respective wave.
10
Children are interviewed once they reach the age of 16; there is also a special survey of 11-15
year old household members from Wave Four onwards. Thus the sample should remain broadly
representative of the population of Britain as it changes through the 1990s. Additional sub-
samples were added to the BHPS in 1997 and 199910.
3.2 The Sample
One decision to be taken was about the starting point for each individual. One could argue
that given that the …rst labour market experience is important in one’s life, this would be a
good starting point. It would be indeed interesting to look at the moment of entry in the labour
market. However it is rather di¢cult to identify the …rst move and thus we decided to start with
the annual information provided in the eleven waves. Moreover if we only looked at individuals
whose date of entry in the labour market takes place within the eleven years of the panel, then
we would be left with a very small number of observations. In other words job mobility is
examined between two interviews, held around September of each year. This method does not
account for multiple job changes within a calendar year and for this reason the sample has been
restricted to respondents who changed at most one time during the year.
An important issue is the distinction between voluntary and involuntary job changes. In
section 4, we report job changes for di¤erent reasons. We have conducted the econometric
analysis looking at quits versus non change and promotions (i.e. all stayers) as well as quits
versus non change (or separately quits, promotions and no changes, in the multinomial logit
section) .
10 All individuals enumerated in respondent households became part of the longitudinal sample. The samplefor the subsequent waves consists of all adults in all households containing at least one member who was residentin a household interviewed at Wave One, regardless of whether that individual had been interviewed in WaveOne. The following rules applied in subsequent waves di¤ered from the sampling rules in Wave One in only onerespect. In both sets of rules, eligibility depended on domestic residence in England, Wales, or Scotland south ofthe Caledonian Canal. In waves after Wave One, however, OSMs were followed into institutions (unless in prisonor in circumstances where the respondent was not available for interview e.g. too frail, mentally impaired etc.)or into Scotland north of the Caledonian Canal.
11
3.3 The Dependent variable
In the wage equations, wages are gross weekly (and hourly) payments11 for full time, non
self employed workers.
The BHPS contains information on workers’ satisfaction. They are asked questions related
to their satisfaction levels from work. The respective question is ”how satis…ed are you with a
speci…c aspect of your job?”, there being seven di¤erent aspects12. The answers take the values
1-7 with 1 being ”not satis…ed at all” and 7 being ”completely satis…ed”. Using that information
we create a dummy variable which takes the value one if the worker reports that he is satis…ed
with his job and zero otherwise. We use the overall satisfaction index which does not focus on
speci…c aspects of the work, either as a dummy variable or as an ordered outcome variable.
Job change will be measured in many di¤erent ways to serve the purpose of each speci…c part
of the analysis. First we use a dummy variable to capture quits as opposed to no change and
within …rm promotions and second we de…ne job change as quits versus no change, i.e. omitting
promotions. We have also estimated multinomial logit models for the three di¤erent outcomes,
namely quits, promotions and no job change.
3.4 The Explanatory Variables
The …rst group of regressors contains the usual family and social background controls. We
use a dummy for gender (equal to one if female), for marital status (married), two dummies for
race (white and black) and region dummies. In addition controls for the number of children of
four age groups are used (age groups: 0-4, 5-11, 12-15 and above 16). Moreover we use controls
for formal education (…ve dummies for the di¤erent quali…cation levels13).
The special feature of the BHPS is that it allows us to merge the job history …les with the
11 This is constructed using the answers to the questions about the last payment and the period that it covers incombination with the question about whether that payment was the usual one. The questions are the following:”The last time you were paid, what was your gross pay, that us including any overtime, bonuses, commission,tips or refund but before any deductions for tax, national insurance or pension contributions, union dues and soon?” and ”how long a period did that cover?”
12 The detailed satisfaction variables in the BHPS are the following:JBSAT1: promotion prospects, JBSAT2: Job satisfaction: total pay, JBSAT3: Job satisfaction: relations with
boss, JBSAT4: Job satisfaction: security, JBSAT5: Job satisfaction: use of initiative, JBSAT6: Job satisfaction:work itself, JBSAT7: Job satisfaction: hours worked, and the overall measure of job satisfaction, JBSAT.
13 These quali…cation dummies are the following: qual1: higher degree, qual2: nursing, qual3: A-level, qual4:O-level, commercial, apprentiship, qual5: no quali…cations.
12
eleven waves of the panel. In that way one can track back the working life of the individual
in order to construct labour market experience variables as well as experience in the speci…c
occupation and/or industry. However actual work experience and actual job tenure are endoge-
nous to the mobility decision (the time until leaving a particular job is a choice variable in the
model). For that reason we use predicted work experience and job tenure by assigning to each
individual the average value of work and unemployment experience for each individual’s region.
We have also done this by education-region cell and the choice between the two does not make
any signi…cant di¤erence14.
Other controls include a dummy for work in the public15 sector, three dummies for employer
size (the …rst for less than 50 employees, the second for between 50 and 200 and the third
for above 200 employees) and sectoral dummies. Dummy variables for other measures of job
satisfaction, related to speci…c aspects of the job16 are also used. One dummy for job satisfaction
with respect to security, one for job satisfaction related to the work itself and one for the hours
worked17. Other work related controls include times of day usually work (four dummies: work
during the day (reference category), work morning only or afternoon only or lunch time and
afternoon only, work evening and night, and shifts or varying times), travel time to work, dummy
for managerial duties, dummy for promotion opportunities. The BHPS asks the respondents to
evaluate their current …nancial situation relative to the past year. Using this information we
construct three dummies. The …rst takes the value one if the …nancial condition is better now,
the second is one if the …nancial situation is worse now and the third (the reference group) if it
is the same. The same questions were asked about the …nancial expectations of the respondents
for the next year. We construct the same dummies as for the …nancial situation questions.
14 So we decide to use the …rst version in this paper.15 i.e. civil servants, central government, local government, NHS, higher education, nationalised industries, non
pro…t organisations16 One may argue that the variables re‡ecting non pecuniary aspects of job satisfaction may be correlated with
the overall measure of job satisfaction. However this overall level of job satisfaction is not a composite measureconstructed from the di¤erent sources of job satisfaction.
Dummy =1 if satis…ed, =0 otherwise.17 Ideally we would like to test other apsects of job satisfaction as well but the questions on total ”relationship
with boss”, promotion prospects and use of initiative were not asked in the last four years of the panel (1998-2001).
13
Table 1: Percentage of Job Change by Reason for ChangeWOMEN
no change Promotion Non Family Family Other InvoluntaryReasons Reasons Change
all 77.23 8.03 6.69 0.68 4.2 3.19
Marital Status 81.04 7.22 4.64 0.71 3.84 2.58married 73.12 8.91 8.91 0.65 4.59 3.84
not married
Marital status- Kidsmarried, kids 77.88 8.51 5.5 0.84 4.51 2.79
married, no kids 82.87 6.47 4.14 0.62 3.46 2.46not married, kids 71.58 7.55 9.72 1.24 5.08 4.86
not married, no kids 73.53 9.27 8.7 0.5 4.47 3.57Source: Waves 1-11 of the BHPS.Notes: Question Asked: "Why did you leave your previous job?" Each row should add up to 100, although thismay be slightly higher because of rounding.
4 Data Patterns
In Tables 1 and 2 we tabulate the reasons for voluntary job changes reported by employed
women and men, working full time. This is done for di¤erent sub groups de…ned by marital
status and the existence of children. We have grouped the answers into promotions, quits for non
family reasons, quits for family reasons and quits classi…ed in the ”other” category18 . Moreover
we report the percentage of women or men in each group who do not change jobs between the
two interviews and those who experience involuntary job loss19.
The main di¤erence between men and women is in the percentage of those who quit for
family related reasons. 0.68% of women per annum working full time versus 0.09% per annum
for men working full time.
In table 3 we tabulate the aspects of the job which make it attractive, for the sample of full
time workers who report to have their previous job for a better one.
Focusing on full time workers, the most important factor explaining the attraction of the
current job is pay. This is true for both men and women, although the percentage for women
(28%) is lower than that for men (37%). The second most important aspect (leaving out the
18 These groups are de…ned in the following way. The answers given which were considered as separations forfamily reasons: ”left to have baby”, ”children, home care”, ”care of other person” and ”moved away for familyreasons”. Those recorded as non family reasons separations are: ”left for better job”, ”started college, university”.Then we de…ne another category which covers the ”other reason” separations. Involuntary separations consist ofthe following answers: ”made redundant”, ”dismissed or sacked”, ”retirement”, and ”stopped for health reasons”.
19 Each row should add up to 100, although this may be slightly higher because of rounding.
14
Table 2: Percentage of Job Change by Reason for ChangeMEN
no change Promotion Non Family Family Other InvoluntaryReasons Reasons Change
all 78.18 7.39 6.58 0.09 3.78 4.01
Marital Statusmarried 80.91 6.88 5.05 0.03 3.62 3.54
not married 73.67 8.23 9.11 0.18 4.06 4.79
Marital status- KidsMarried, kids 79.23 7.57 5.55 0.03 4.26 3.38
Married, no kids 82.98 6.04 4.42 0.02 2.83 3.74Not married, kids 69.72 7.74 10.76 0.37 5.31 6.12
not married, no kids 74.55 8.34 8.74 0.14 3.78 4.49Source: Waves 1-11 of the BHPS.Notes: Question Asked: "Why did you leave your previous job?" Each row should add up to 100, although thismay be slightly higher because of rounding.
”other reasons” response) is promotion prospects. 13% of men believe this is the main attraction
of their new job, whereas only 8% of women report the same. For women, the speci…c type of
their new job as well as its interest content seems to be very important, and actually more
important than for men. Moreover ”less commuting” does matter more for women than for
men,which can be well linked to family responsibilities and home production for women. The
evidence in this table can be seen as a …rst expression of di¤erences in tastes and preferences
between men and women. In the same table we report tabulations for married versus non
married men and women. Looking …rst at pay as the main attraction of the job, the di¤erence
between the percentage of married and unmarried who report it is much larger than that for
men. For both genders though it seems that pay is more important for the unmarried ones.
The same holds for promotion prospects and and the type of work (however this is no very clear
whether it relates to content of work). On the contrary, it appears that marriage makes security
a more important factor in the determination of an attractive job. For example 9% of full time
married women who changed job for a better one report job security to be the reason, whereas
this percentage falls to 6% for non married women. The respective numbers for men are 9 and
5.5%. We note that there is an important di¤erence between married and non married workers
suggesting ‡exible hours as the positive aspect of their new ”better” job. In particular 5.5%
(3.4%) of married women (men) report this as the main reason, versus 2% for unmarried (same
for men). Finally there is a di¤erence in the percentage of married versus non married men and
15
Table 3: Attraction of current job relative to previous job (percentage), 1991- 2001
Full Time WorkersWOMEN MEN
married non married married non married
more/better money 29.77 27.7 30.93 36.67 35.16 37.97promotion prospects 8.31 6.34 9.28 13.22 11.3 14.83more responsibility 2.45 1.9 2.75 1.53 1.86 1.24
job security 7.2 9.09 6.19 7.12 9.19 5.39more interesting job 7.8 9.51 6.87 5.37 5.71 4.98speci…c type work 8.91 6.77 10.08 4.58 3.48 5.5
be own boss 0.37 0.21 0.46 0.79 0.75 0.83greater initiative 2.15 1.9 2.29 1.58 1.86 1.35less commuting 4.31 5.5 3.67 3.22 4.22 2.39
less hours 1.11 1.27 1.03 1.3 1.61 1.04more ‡exible hrs 3.12 5.5 1.83 2.6 3.35 1.97health reasons 0.22 0.63 0.45 0.62 0.31to use skills 4.01 3.81 4.12 3.84 3.11 4.46
less demanding wrk 1.04 1.9 0.57 1.02 1.24 0.83prefer this job 6.01 5.07 6.53 5.08 4.84 5.29new job better 3.79 2.96 4.24 3.9 3.73 4.05
other 9.43 9.94 9.16 7.74 7.95 7.57Total 100 100 100 100 100 100
Number of Observations 1347 473 873 1770 805 964
Source: Waves 1-11 of the BHPS.Notes: Question Asked: "What is the attraction of your current job relative to your previous one?" Question onlyasked to workers who answered to have left their job for a better one. Only allowed to give one answer.
women who …nd less commuting as the good aspect of their new job.
Another way of looking at the di¤erences in male and female preferences is to estimate
satisfaction equations using as the dependent variable a measure of overall job satisfaction. On
the right hand side we include only the di¤erent components of job satisfaction, described in
section 3.3. Table 4 reports the estimated coe¢cients separately for full time men and women.
One can see that women are less concerned with pay as well as promotion prospects in their
current job. In contrast they place more weight than men on the relations with their boss issue as
well as the work itself. Moreover the component of job satisfaction attributed to working hours is
more important for women than for men. The di¤erences in the coe¢cients are signi…cant with
the exception of promotion prospects, use of initiative and hours worked. The results are very
similar when full time and part time workers are pooled, with a more pronounced di¤erential
on the impact of hours of work.
In the appendix, section 8.2, we present graphs (1- 9) of the proportion of men and women
16
Table 4: Aspects of Job Satisfaction, 1991-2001 (full time workers)
WOMEN MEN ALL interactionsfemale* satisf.
(1) (2) (3) (4)promotion prospects 0.103 0.114 0.115 -0.013
(0.007)** (0.006)** (0.006)** (0.010)total pay 0.088 0.119 0.120 -0.033
(0.008)** (0.007)** (0.007)** (0.010)**relations with boss 0.172 0.143 0.143 0.031
(0.008)** (0.007)** (0.007)** (0.010)**security 0.090 0.133 0.134 -0.045
(0.008)** (0.006)** (0.006)** (0.010)**use of initiative 0.106 0.115 0.115 -0.009
(0.010)** (0.008)** (0.008)** (0.013)work itself 0.369 0.318 0.317 0.054
(0.010)** (0.008)** (0.008)** (0.013)**hours worked 0.134 0.141 0.139 -0.002
(0.009)** (0.007)** (0.007)** (0.011)Constant -0.118 -0.323 -0.310
(0.075) (0.060)** (0.046)**Observations 9419 13657 23076R-squared 0.53 0.51 0.52
Chow Test F(7, 16121)= 8.40(0.000)
Notes: Fixed e¤ects job satisfaction regression. Overall measure of jobsatisfaction as the dependent variable. No additional controls. Standard errors in parentheses
17
(of certain groups) who remain in the same job, get promoted and those who quit their jobs, in
the 10 years covered by the BHPS.
5 Model and Empirical Procedure
5.1 Wage Equation
In the …rst stage we estimate a standard wage equation in order to derive the residuals that
will be used in the next section. The estimated equation has the following form:
wit = Xit³ + "it (1)
where wit is the log wage and vector X contains the usual individual characteristics (age, age
square, marital status, general health dummies, whether there are kids of speci…c age groups
(or number of children of this age) along with job related controls. This last set of variables
consists of occupation and industry dummies as well as times of the day the individual usually
worked and distance to work in minutes20 . "it = ®i + ®t + uit, a standard error term in panel
data models. ®i is the individual speci…c component of the error which remains constant across
time. uit is the time varying component. ®t is a set of time dummies. The …rst term can be
thought of as unobserved ability whereas the second can be pure luck. Dispersion between the
wage paid to a particular worker and that paid to individuals with similar characteristics can
be explained by di¤erences in unobserved ability or luck. The …rst is represented by ®i in the
error and the second by uit.
The results from the wage regression are presented in table 5, separately for men and women.
Interpreting the results from the wage equation is not our objective in this paper and thus we
proceed to the next section.
20 The BHPS contains information on the times of the day that the respondent ussually works. From this weconstruct the following four dummies: wktime1=1 if working during the day, wktime2=1 if working morning only,afternoon only, or lunch time and evenings, or other times during the day, wktime3=1 if working during the nightand evening, wktime4=1 if shifts and varying times.
18
Table 5: Wage Regression, Full time Workers, 1991-2001
Women Men
age 0.066 0.099(0.009)** (0.008)**
agesq -0.001 -0.001(0.000)** (0.000)**
higher degree 0.052 0.045(0.030) (0.024)
nursing 0.146 -0.078(0.039)** (0.116)
A-level + 0.005 0.044(0.033) (0.026)
O-level and -0.060 -0.006(0.032) (0.026)
married -0.048 0.007(0.009)** (0.009)
professional 0.130 0.057(0.022)** (0.014)**
managerial 0.142 0.081(0.013)** (0.010)**
non skil led 0.057 -0.016manual (0.013)** (0.011)skilled 0.034 0.023manual (0.013)* (0.008)**
manufacturing 0.028 -0.034(0.031) (0.017)*
services -0.035 -0.056(0.030) (0.017)**
other -0.056 -0.083sectors (0.030) (0.017)**public 0.010 -0.001
(0.012) (0.012)size [50.200] 0.039 0.054
(0.008)** (0.006)**size 200+ 0.071 0.074
(0.009)** (0.007)**good/fair -0.014 -0.003
health (0.006)* (0.005)bad, very bad -0.153 -0.080
health (0.029)** (0.027)**Constant 4.695 3.988
(0.359)** (0.245)**Observations 16665 23187R-squared 0.2107 0.2122
Note: These are the results from a wage regression with …xed e¤ects. Time dummies are also included in all regressions.Standard errors in parentheses.
19
5.2 Satisfaction Equation
The relationship between job changes and job satisfaction has been investigated in the
literature in the past. Mainly it has been analysed in terms of predicting job quits for dissatis…ed
workers. Freeman (1978) provides evidence that reported job satisfaction is a good predictor
for job mobility over and above the e¤ect of past wages. Akerlof, Rose and Yellen (1988) and
Clark, Georgellis and Sanfey (1997) con…rm Freeman’s …ndings using US and German data.
However job satisfaction can be very interesting in its own right. By examining the de-
terminants of job satisfaction one can shed some light on di¤erences in preferences and work
tastes between men and women when choosing a speci…c job and thus make inference about job
mobility decisions.
The relationship between wages and job satisfaction has also been extensively investigated
in the literature. Job satisfaction can be thought of as a function of both the actual wage (wage
in current period) and an indicator of the relative wage. Wages are likely to determine job
satisfaction, not only in absolute terms but also in relative terms. Individuals derive satisfaction
from their wage level and also from the way this compares to the wages of other workers21 .
In this section we estimate job satisfaction as a function of the worker’s characteristics
(vector X), the wage gap (measured by the residuals from the wage equation, of section 5.1),
c"it = (wit ¡ cwit) and/or actual wages22. In addition we attempt to control for non pecuniary
aspects of job satisfaction (vector Z). These we capture either by including non pecuniary aspects
of satisfaction directly or by including controls for work related aspects such as ”times of day
usually work”, hours of work, whether the worker would like to work more or fewer hours,
distance to work, public sector dummy, employer size, etc. This information is available in the
21 This in turn can be either a measure of wage relative to other individuals or one which compares the actualwage of the individual to that received by the same worker in the last period. The …rst measure has been usedby Philippe Moguerou (2002). He constructs a measure of relative earning and then using its residuals in the jobsatisfaction regression. An alternative solution suggested again by Moguerou is the creation of a dummy variablewhich takes the value one for the well paid individuals (that is those with positive residuals) and zero otherwise.
Job satisfaction and wages are likely to be simultaneously determined. As suggested by Chevalier and Lydon(2002) it would be more appropriate to estimate the two as a system of equations using identifying restrictions.Their results show they cannot reject the hypothesis that wages are endogenous in a job satisfaction regression.It is very di¢cult to …nd a valid instrument for wages in the job satisfaction equation (They use partner’s wagesas an instrument for satisfaction in the wage equation.). In their paper they add controls for occupation inthe satisfaction regression instead of accounting for job conditions and thus make an attempt to account forcompensating di¤erentials.
22 In some versions we also use predicted wages instead of actual wages. For more details see table 6.
20
BHPS and included in vector Z. So the latent job satisfaction has the form,
S¤it = ¯1 + Xit¯2 + Zit¯3 +¯4c"it + 'it (2)
The observed job satisfaction is Sit =
8<:
1 if S¤it > 0
0 if S¤it · 0
9=;.
The purpose of looking at job satisfaction is to get a better understanding of the di¤erent
aspects shaping worker’ preferences with respect to job choice. That is we examine the di¤erent
aspect of work which may a¤ect the way people perceive their occupation. This will be useful
later on, in the attempt to explain mobility decisions and career choices. In particular we want
to investigate whether there are di¤erences in the factors determining job satisfaction for men
and women. If there are di¤erences in preferences and tastes, then these should be captured in
sign and/ or magnitude of the coe¢cients.
We want to analyse the factors determining job satisfaction, with particular interest in
possible di¤erences between men and women. If the common belief that women have a weaker
attachment to the labour market is true, we would expect to …nd pecuniary aspects of the job to
be more important for males than females. On the contrary hours of work, especially ‡exibility
of working hours, should matter more for women.
We investigate the determinants of job satisfaction. Here the dependent variable is a simple
binary variable which takes the value one if the individual reports to be satis…ed with his job
and zero otherwise23. We estimate a standard probit model as described in equation 2. Table
6 presents the results in the form of marginal e¤ects separately for men and women. These are
de…ned as the partial change in the predicted probability of job satisfaction for a change in one
of the explanatory variables24 .
De…ne
Pr(Sit = 1) = F(X
Wikt¯k) (3)
where W is a vector containing X;Z; c"it: Then the marginal e¤ect with respect to say Wk is:
23 See sections 3.3 and 3.4 for a detailed description of the dependent variable and the controls used.24 One should be very cautious when interpreting marginal e¤ects in the job satisfaction equation. This is
because there no measurement units for job satisfaction and thus we can only comment on the direction of theimpact but not literally interpret it.
21
@F(X
Wikt¯k)
@Wk= f(
XWikt¯k)¯k; (4)
which for the probit model, can be written as:
@F(X
Wikt¯k)
@Wk= Á(
XWikt¯k)¯k; (5)
where Á is the standard normal density function.
In Table 6, in the …rst speci…cation we use the logarithm of current wage instead of the
residual whereas in the second speci…cation we use the residuals, c"it. With this last term
we attempt to capture the relative wage e¤ect, documented in the literature. In the third
speci…cation we use the residuals, c"it and the set of variables regarding current …nancial situation
or expectations about future …nancial situation.
Promotion prospects in current job have a positive impact on job satisfaction but this e¤ect
is smaller for women than for men. More importantly female workers become unsatis…ed the
bigger the distance to work and there is a large di¤erence between that coe¢cient for males and
females. As expected the number of working hours has a negative e¤ect on job satisfaction for
women. Moreover this coe¢cient becomes larger when interacted with the presence of children
aged 5-11 for women. We believe this as preliminary evidence of di¤erent preferences for men and
women as well as distinct constraints (family related obligations) for the two groups. Consistent
with the existing evidence we …nd that satisfaction is U-shaped in age and married workers
seem to be more satis…ed. The size of the workplace seems to a¤ect negatively the level of job
satisfaction, and we con…rm the standard e¤ect of the general state of health.
Considering the e¤ect of wages on job satisfaction, we …nd the well documented positive
e¤ect for men but not for women. The higher the wage in current job, the more satis…ed the
worker. For the relative wage measure, the e¤ect is larger than that of the actual wage for women
and it is also statistically signi…cant. It has the expected positive sign, meaning that workers
who are paid more than other individuals with similar characteristics, tend to be more satis…ed
with their job, on average. For men, the e¤ect of relative wages is also positive but smaller
than that of actual wages. However it remains larger for men than for women (5.5 versus 3.6),
implying that wages (either absolute or relative) are more important for men than for women.
22
Table 6: Determinants of Job Satisfaction
(Equation 3, Marg. e¤ects)
WOMEN MEN
(1) (2) (3) (4) (5) (6)age -0.007 -0.006 -0.006 -0.018 -0.013 -0.009
(0.003)** (0.003)* (0.003)* (0.003)** (0.002)** (0.002)**age squared 0.000 0.000 0.000 0.000 0.000 0.000
(0.000)** (0.000)** (0.000)** (0.000)** (0.000)** (0.000)**married 0.048 0.049 0.046 0.026 0.032 0.031
(0.009)** (0.009)** (0.009)** (0.010)* (0.010)** (0.010)**public 0.012 0.013 0.013 0.000 -0.000 0.006
(0.010) (0.010) (0.010) (0.013) (0.013) (0.013)size [50.200] -0.032 -0.030 -0.029 -0.037 -0.029 -0.028
(0.011)** (0.011)** (0.010)** (0.010)** (0.010)** (0.010)**size 200+ -0.043 -0.039 -0.039 -0.043 -0.031 -0.029
(0.010)** (0.010)** (0.010)** (0.010)** (0.010)** (0.010)**good/fair -0.070 -0.071 -0.066 -0.080 -0.084 -0.075
health (0.009)** (0.009)** (0.009)** (0.009)** (0.009)** (0.009)**bad, very bad -0.137 -0.137 -0.112 -0.194 -0.218 -0.186
health (0.049)** (0.049)** (0.046)* (0.049)** (0.049)** (0.049)**promotion opportunities 0.081 0.080 0.076 0.121 0.126 0.116
(0.008)** (0.008)** (0.008)** (0.008)** (0.008)** (0.008)**morning afternoon 0.010 0.013 0.009 -0.044 -0.039 -0.040
(0.024) (0.024) (0.024) (0.032) (0.032) (0.032)evening, night 0.014 0.014 0.008 -0.079 -0.081 -0.084
(0.027) (0.027) (0.027) (0.025)** (0.026)** (0.025)**shifts varying -0.004 -0.002 -0.002 -0.037 -0.032 -0.032
(0.012) (0.012) (0.012) (0.011)** (0.011)** (0.011)**travel to work time -0.027 -0.025 -0.025 -0.029 -0.021 -0.020
(0.012)* (0.012)* (0.012)* (0.009)** (0.009)* (0.009)*whether 0.033 0.034 0.048 0.007 0.007 0.025kids 0-2 (0.015)* (0.015)* (0.014)** (0.012) (0.012) (0.011)*whether 0.023 0.023 0.022 0.021 0.024 0.025kids 3-4 (0.017) (0.017) (0.017) (0.011) (0.011)* (0.011)*whether 0.021 0.020 0.017 0.015 0.016 0.014kids 5-11 (0.011)* (0.011) (0.011) (0.010) (0.010) (0.010)whether 0.033 0.030 0.031 0.023 0.021 0.023
kids 12-15 (0.010)** (0.010)** (0.010)** (0.010)* (0.010)* (0.010)*Number of -0.001 -0.001 -0.001 0.001 0.001 0.001hours worked (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
wants -0.101 -0.102 -0.100 -0.132 -0.130 -0.126fewer hours (0.008)** (0.008)** (0.008)** (0.008)** (0.008)** (0.008)**
wants -0.014 -0.016 -0.008 -0.079 -0.083 -0.067more hours (0.018) (0.018) (0.017) (0.015)** (0.015)** (0.015)**
wage 0.036 0.026 0.055 0.031residuals (0.014)** (0.014) (0.014)** (0.014)*log wage 0.009 0.068
(0.010) (0.009)**…nancial situation 0.016 0.063
better now (0.008)* (0.008)**…nancial situation -0.063 -0.080
worse now (0.010)** (0.010)**exp …n -0.010 0.013
better than now (0.007) (0.007)exp …n -0.060 -0.073
worse than now (0.013)** (0.013)**Observations 14511 14358 14350 19754 19518 19507
Notes. These are probit regressions. Robust standard errors reported, corrected for within group correlation(clustering). Many of the non signi…cant variables are not reported in the table. Additional controls for occupation,industry, region and time dummies included in all regressions.
23
In columns (3) and (6) we estimate the satisfaction with additional controls. In particular
we want to look at the responsiveness of women to di¤erent aspects of their current …nancial
situation (relative to that of last year) and their expectations for the following year. These as
further evidence of the pecuniary e¤ect on job satisfaction on top of the actual and/or relative
pay e¤ect. This table veri…es what we found earlier, that women are less motivated by pay and
their …nancial situation. Current …nancial conditions better than those of the previous year do
increase the satisfaction level of both men and women, but the e¤ect is much smaller for the
females. The same holds for the expectation variables.
Overall, in this section, we …nd that the factors shaping job satisfaction may di¤er between
men and women. The later are less interested in pay and other pecuniary aspects of the job. In
contrast, they are more concerned with non monetary aspects of their job.
5.3 Job Change Equation
An individual will change jobs if the expected value of his alternative job, EV mit ; exceeds
that of the expected value of his current job, EV sit; plus the cost of job change, Cit. Thus he
will move if
M¤it = EV mit ¡EV sit ¡Cit > 0 (6)
EV sit (EV mit ) is the expected value of the current job (new job/ outside opportunities) forecasted
in the future conditional on experience at time t. Note that the expected value of the current job
and outside option is not formed only on the basis of pecuniary payo¤s. Non pecuniary aspects
of the job, such as the content of the job, working hours and ‡exible arrangements, travel time
to workplace etc, are also expected to be important.
We have …ve di¤erent sets of variables: individual characteristics, pecuniary aspects of job,
non pecuniary aspects of job, cost parameters and family responsibilities (i.e. constraints).
We can write the latent model as:
M¤it = Xit°1+Kit°2 +Cit°3 +°4j
JX
j=1
satjit +°5
2X
j=1
HRjit + ³it (7)
where ³it = ®0i + ®0t + u0it, a standard error term in panel data models. ®0
i is the individual
24
…xed e¤ect and u0it a standard error term. ®0i is again a vector of time dummies. The estimated
job change equation includes controls for personal characteristics Xit: Vector Kit (vector K is a
subset of vector Z, of equation 2) contains controls for job attributes which can be seen as proxies
for the di¤erence in the expected value as presented in (6): In addition it contains variables which
re‡ect the expectations of the individuals about the future as well as their judgement about their
current …nancial situation relative to last year. In particular, we include two dummy variables,
the …rst equals one if the respondent believes he is better o¤ at present and the second takes
the value one if he is worse o¤ (the default is that the worker believes his …nancial situation
remained the same or he is unsure about it). These two variables are also constructed for the
expectations of the worker for the next year regarding his …nancial situation. However these
two variables are likely to be endogenous in the mobility decision, in the sense that they take
account of planned job change. Moreover we include controls for hours worked (vector HR).
Vector Cit is the cost of moving and will be proxied by family constraints i.e. marital status,
presence of children of di¤erent age groups. The age groups are the following: 0-4, 5-11, 12-15
and 16-18. Furthermore this vector includes regional dummies to control for local labour market
characteristics. We expect gender di¤erences to exist both in preferences and constraints.
The observed binary variable for job change is given by:
Mit =½
1 if M¤it > 0
0 if M¤it · 0
¾; (8)
or using equation 7, the decision to change jobs between t and t+1 will be taken according to:
Mit =
8<:
1 if Xit°1 +Kit°2 + Cit°3 +°4jPJj=1 satjit°5 +
P2j=1HRjit + ³ it > 0
0 if Xit°1 + Kit°2 +Cit°3 + °4jPJj=1 satjit°5 +
P2j=1 HRjit + ³it · 0
9=; (9)
Job change will also be estimated in a multinomial logit model to investigate the determinants
of job mobility distinguishing between quits and promotions relative to workers who remain in
the same job. In addition it would be interesting to distinguish between those who stay with
the same employer and those who change employers along with jobs. This will be left for future
research.
25
5.3.1 Binary Choice Model
This section presents the results from job mobility equations with the dependent variable
being quits versus non job change, without promotion. The results are presented in tables 7
and 8 separately for men and women. In tables 16 and 17 in the appendix we report additional
results where the dependent variable is a dummy which takes the value one if the worker quits
his job and zero if he remains in the same job or is promoted (thinking of promoted workers as
successful stayers). Booth and Francesconi (1999) …nd that the quit rates of men and women
are quite similar and di¤erences can only be found in the layo¤ probabilities. Moreover Booth,
Francesconi and Frank (2001) …nd that women are as likely as men to be promoted. However
there are underlined di¤erences in the factors determining quit decisions between men and
women. These are the ones we will investigate in this paper
Highly educated workers are more likely to change jobs voluntarily. This can be explained
in terms of larger variety of alternatives that better quali…ed individuals may have. This is
equally true for men and women. In addition we …nd a negative e¤ect of both job tenure and
general labour market experience on the probability of voluntary job change, con…rming previous
…ndings in the literature. Given that this is not the primary focus of the paper we do not present
these results in tables 7, 8, 17 and 16 but they are available upon request.
To start with, comparing the results from tables 7, 8, 16 and 17, we can see that it does
not make a great di¤erence whether we look at quits versus non change only or quits versus
promotions and non change25. Married women are less likely to quit (by 2%), the same holds
for men, but the e¤ect is half the size of that for women. Quits are less frequent in the public
sector and TU membership reduces the probability of quitting.
The …rst interesting e¤ect, related to the di¤erences in family constraints and cost functions
story, is that of the travel to work time variable. The marginal e¤ect of this variable appears
positive and statistically signi…cant in almost all speci…cations. It implies that the larger the
distance from the current workplace, the more likely to quit within the next period. This
marginal e¤ects is much larger for women, 2.5-3.5% than for men, 0.6-1.3%. This can be seen
as a …rst piece of evidence that household responsibilities may be more important for women
which makes their opportunity cost of non home production higher. Moreover the di¤erence in
25 More on this in the multinomial logit analysis, in the following section.
26
the coe¢cient is statistically di¤erent from zero.
If the hypothesis that women are less concerned with pay and career prospects is true, then
this should be re‡ected on the coe¢cients of the pay variables in the job change equation. Sur-
prisingly we …nd exactly the opposite. The marginal e¤ect of log wage is statistically signi…cant
in all regressions, for both men and women. It has the expected negative sign and it is larger
in magnitude for women. The di¤erence in the coe¢cient is not signi…cant, but it still appears
that women do make decisions regarding quitting, taking into account pecuniary aspects of the
job, unlikely what our initial suspicion was and what our job satisfaction results of the previous
section suggested.
The other wage variable, the residuals from the wage equation, is used as a relative wage
measure. The higher the residuals, the larger the wage gap between the actual wage paid
to worker i and all other workers with similar characteristics (note that this may be due to
unobserved ability or luck). The higher the wage residual, the less likely to leave one’s job. This
relative wage measure is negative and large in magnitude especially for men, the marginal e¤ect
being -6.2%. For women this is much smaller (less than half that for men, i.e. -3.1%) and the
di¤erence between the two is signi…cant.
One more variable which may re‡ect strategic career planning and might reveal something
about di¤erences in preferences between men and women is the promotion prospects variable.
This takes the value one if there are promotion prospects in current job, and zero otherwise. It
comes up with a negative sign, as expected. Individuals in jobs with promotion prospects are
less likely to quit their jobs. For women, this e¤ect is not statistically signi…cant in any of the
speci…cations. For men, it is much larger (of size -1%) and comes up signi…cant in almost all
speci…cations. The negative e¤ect holds for both genders but it is larger for men (however the
di¤erence is not signi…cant).
The job satisfaction measures seem to determine quitting. Job satisfaction with pay, hours
of work and the work itself have negative marginal e¤ects. The e¤ects of the …rst two appear to
be the same for men and women. Workers who are satis…ed with di¤erent aspects of their job
are less likely to quit. It is worth noting that satisfaction with the work itself is more important
for women, and actually its e¤ect is larger than that of satisfaction with pay for this group of
workers.
27
A last set of variables which are likely to shed some light on workers’ quitting behaviour are
those related to family commitments. Speci…cally we use two variables. The …rst is a dummy
variable which takes the value one if the worker had to work fewer hours because of family
commitments and zero otherwise. The second dummy is one if family constraints prevented job
search. We …nd that the …rst is positively correlated with quitting for women, the marginal e¤ect
(of the order of 6%) being signi…cant at the 10% level. However job search because of family
constraints does not seem to impede changing jobs. For men both variables have no statistically
signi…cant e¤ects on quits. This is direct evidence -although weak- that family responsibilities
do impact on the mobility decisions for women, whereas this does not seem to be the case for
men.
5.3.2 Multiple Choice Model: Promotions, Quits and no Job change
In this section we estimate a multinomial logit model by treating promotions, quits and no
job change as three distinct outcomes. We can then test whether two of these groups can be
pooled together. We de…ne:
M0it =
8>>><>>>:
0 if no change
1 if promotion
2 if quit
9>>>=>>>;
(10)
Pr(M0= m) =
e°0jN
P2j=0 e°0jN
; where j = 0;1;2
The marginal e¤ects in the multinomial logit are given by:
µj =@ Pr(M 0 = m j N )
@Nk= Pr(M
0= m j N)
"°km ¡
2X
k=0
°kj Pr(M0= j j N )
#
The results are given in tables 9 and 10. For purposes of presentation, we only report the
marginal e¤ects for the variables of interest. All regressions include controls for quali…cations,
tenure and experience, number of children in speci…c age groups, occupation, dummy for public
sector, establishment size and time dummies. We have performed Wald tests in order to see
whether two categories can be pooled together. In all speci…cations and for both men and
28
Table 7: Quits versus no change (Marg. e¤ects)
Binary Choice Model, Full Time Women
(1) (2) (3) (4) (5) (6) (7)Promotion -0.007 -0.004 -0.004 -0.004 0.001 0.000 0.002
Opportunities (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) (0.008)married -0.022 -0.021 -0.022 -0.018 -0.016 -0.014 -0.010
(0.006)** (0.007)** (0.007)** (0.006)** (0.006)* (0.007)* (0.009)number of -0.001 -0.000 -0.000 -0.001 -0.000 -0.000 0.000overtime (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
number of 0.000 0.001 0.001 0.000 0.000 0.001 0.002working hours (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)* (0.001)
TU -0.017 -0.012 -0.013 -0.015 -0.020 -0.016 -0.015(0.007)* (0.008) (0.008) (0.007)* (0.007)** (0.008)* (0.011)
Travel to 0.027 0.035 0.034 0.029 0.025 0.029 0.035work time (0.009)** (0.010)** (0.010)** (0.009)** (0.009)** (0.010)** (0.014)*log wage -0.036 -0.025 -0.034 -0.034
(0.008)** (0.009)** (0.008)** (0.010)**wage -0.031
residuals (0.016)*satisfaction -0.012 -0.008with money (0.002)** (0.002)**satisfaction -0.004 -0.005 -0.008with hours (0.002)* (0.002)* (0.003)**satisfaction -0.014 -0.015 -0.016
with initiative (0.002)** (0.002)** (0.003)**satisfaction 0.001 -0.003 0.001
with work itself (0.002) (0.002) (0.003)wants -0.012 -0.003
fewer hours (0.007) (0.009)wants 0.016 0.020
more hours (0.016) (0.021)…nancial situation 0.009
better now (0.007)…nancial situation 0.014
worse now (0.009)…nancial exp 0.029
better than now (0.007)**…nancial exp 0.031
worse than now (0.013)*fam. Commit. 0.059
work fewer hours (0.038)fam. Commit. 0.012
prevent job search (0.026)Observations 8659 7870 7869 8655 8621 7809 4053
Notes. Robust standard errors reported, corrected for within group correlation (clustering).Controls for occupation, industry, region and time dummies included in all regressions. Standard errors reportedbelow the marginal e¤ects.
29
Table 8: Quits versus no change (Marg. e¤ects)
Binary Choice Model, Full Time Men
(1) (2) (3) (4) (5) (6) (7)Promotion -0.010 -0.011 -0.010 -0.005 -0.002 -0.006 0.002
Opportunities (0.004)* (0.005)* (0.005)* (0.004) (0.004) (0.005) (0.006)married -0.012 -0.009 -0.011 -0.011 -0.010 -0.003 0.004
(0.005)** (0.005) (0.005)* (0.005)* (0.005)* (0.005) (0.006)number of 0.000 0.001 0.001 0.000 0.000 0.000 0.001overtime (0.000) (0.000)* (0.000) (0.000) (0.000) (0.000) (0.000)
number of 0.000 0.001 0.001 0.000 0.000 0.001 0.000working hours (0.000) (0.000)* (0.000)* (0.000) (0.000) (0.000) (0.000)
TU -0.018 -0.017 -0.016 -0.018 -0.019 -0.018 -0.027(0.005)** (0.005)** (0.005)** (0.005)** (0.005)** (0.005)** (0.007)**
Travel to 0.007 0.013 0.011 0.007 0.006 0.012 0.005work time (0.005) (0.005)* (0.005)* (0.005) (0.005) (0.005)* (0.006)log wage -0.032 -0.016 -0.026 -0.024
(0.006)** (0.006)** (0.005)** (0.007)**wage -0.062
residuals (0.012)**satisfaction -0.012 -0.008with money (0.001)** (0.001)**satisfaction -0.002 -0.004 -0.004with hours (0.001) (0.001)** (0.002)*satisfaction -0.008 -0.010 -0.007
with initiative (0.002)** (0.002)** (0.002)**satisfaction -0.002 -0.005 -0.005
with work itself (0.002) (0.002)** (0.002)**wants -0.008 -0.006
fewer hours (0.005) (0.006)wants 0.016 0.013
more hours (0.010) (0.012)…nancial situation 0.006
better now (0.005)…nancial situation 0.014
worse now (0.006)*…nancial exp 0.026
better than now (0.005)**…nancial exp -0.004
worse than now (0.008)fam. Commit. -0.023
work fewer hours (0.017)fam. Commit. -0.002
prevent job search (0.020)Observations 12025 10914 10914 12016 11987 10810 5720
Notes. Robust standard errors reported, corrected for within group correlation (clustering).Controls for occupation, industry, region and time dummies included in all regressions. Standard errors reportedbelow the marginal e¤ects.
30
women, the null hypothesis (that two categories can be collapsed) is clearly rejected. This
con…rms our decision to investigate separately quits, promotions and no job change.
In table 9 we report the results for women. As expected, women who are in jobs with
high promotion opportunities are more likely (5%) to be promoted than those in jobs without
promotion prospects, versus no job change. On the contrary, high promotion opportunities
reduce the probability of quitting. Women who work longer hours are more likely to be promoted
(this without any signi…cant e¤ect of overtime).
The travel to work time variable con…rms the results previously found in the binary model.
That is, the longer the travel to work time, the higher the probability that a female worker
leaves her job. The marginal e¤ect is of the order of 3%.
Actual wage has a positive impact on the probability of promotion of the order of 1.4-3.2%
and a negative and signi…cant impact on the probability of quitting of -3.5%. The residuals
from the wage equation enter the promotion equation with a negative sign, suggesting that in
this case it may be the luck parameter which is captured by the residuals. That is it may be
that workers who are paid more than those with similar characteristics, are less likely to be
promoted. This can be because the higher pay suggests that they have been luckier than the
others and thus their probability of being promoted is lower. In other words this could be seen
against the existence of unobserved ability (captured by the residuals).
The satisfaction variables are mostly negative both when quits and promotions are concerned.
It is worth noting that work satisfaction with initiative seems to be the strongest aspect of job
satisfaction determining mobility decisions.
It is interesting to look at the variables re‡ecting the workers’ expectations about their
…nancial situation, one year later26. When individuals believe their …nancial situation will be
worse than at the date of the interview, they are more likely (by 4.5%) to quit their job. This
is as if they are trying to avoid the down turn of their …nancial situation by changing jobs. In
addition those who believe they’ll be better o¤ are more likely to be promoted (marginal e¤ect
of 2.5%) but also more likely to leave their job. In the …rst case, it seems that they actually
foresee their promotion in the following year. In the second case, they expect an improvement
in their …nancial situation possibly because they anticipate a quit and a better job in the near
26 The ones which compare their current …nancial situation to that of one year ago are not statisticaly di¤erentfrom zero, for women.
31
future. It could also be that they have already started searching for a new job, although still
employed in their current job. However these variables are likely to be endogenous in the sense
that they may take account of future quits, as already discussed in the previous section.
Finally the two variables re‡ecting further family commitment e¤ects suggest the following.
Individuals who report that family commitments oblige them to work fewer hours are signif-
icantly less likely to be promoted, the marginal e¤ect being -7%. What is surprising is the
positive and signi…cant e¤ect of the ”family commitments prevent job search” variable on the
probability of quitting.
Turning to men now, the results are reported in table 10. The e¤ect of promotion opportu-
nities is slightly smaller for men than for women, but it remains sizeable and signi…cant. Being
married has a negative impact on quitting for men too, but its size is again smaller than for
women. The travel to work time variable has a much smaller e¤ect for men. This is in line
with our initial expectations given that it may be seen as expressing the fact that women may
be more constrained by household responsibilities and thus distance to work can be by itself an
important job attribute for them and thus a primary determinant of quitting behaviour.
However we …nd that actual wages for men have a negative e¤ect on quits but this is smaller in
magnitude than that for women. But it is still sizeable of -2%. Wage residuals are only signi…cant
when concerning quits. For men we …nd a marginal e¤ect of -5% suggesting that those individuals
who perceive being better paid than individuals of the same observable characteristics, are
less likely to quit their job. Note that this ”over compensation” can be explained by either
unobserved ability di¤erences, or simply luck.
The satisfaction variables are even more signi…cant for men than for women. In particular,
satisfaction with initiative at work is the …rst aspect, with a marginal e¤ect of the order of
-6% to -8%. Then satisfaction with the work itself and hours worked seem to be of the same
importance, among the di¤erent aspects of job satisfaction.
Workers who expect to be in better …nancial situation next year, are again 3-3.6% more
likely to be promoted, and 1-2% more likely to quit their jobs. In addition men who believe
they are in a worse …nancial situation in period t relative to that in t-1, are 1% more likely to
leave their job.
It is very interesting that the family commitment variables are not signi…cant for men in
32
Table 9: Multinomial Logit: Promotions, Quits versus no change (Marg. e¤ects)
Ful l time women
(1) (2) (3) (4) (5) (6)promotion quit promotion quit promotion quit
Promotion 0.054 -0.014 0.055 -0.008 0.053 -0.011Opportunities (0.006)** (0.006)* (0.006)** (0.006) (0.008)** (0.008)
married -0.004 -0.015 0.001 -0.012 0.001 -0.006Number of kids: (0.005) (0.007)* (0.005) (0.006) (0.008) (0.009)
0-4 -0.006 0.026 -0.002 0.026 0.014 0.011(0.009) (0.009)** (0.009) (0.008)** (0.014) (0.014)
5-11 0.011 0.003 0.012 0.004 0.015 0.004(0.005)* (0.005) (0.005)** (0.005) (0.007)* (0.007)
12-15 0.000 -0.005 0.001 -0.004 0.011 0.000(0.006) (0.007) (0.006) (0.007) (0.008) (0.009)
16-18 -0.003 -0.014 -0.003 -0.011 -0.003 -0.017(0.012) (0.015) (0.012) (0.015) (0.016) (0.022)
number of 0.001 -0.000 0.001 -0.000 0.001 -0.000overtime (0.000) (0.001) (0.000)* (0.001) (0.001) (0.001)
number of 0.001 0.001 0.002 0.001 0.002 0.001working hours (0.000)** (0.001) (0.001)** (0.001) (0.001)* (0.001)
TU 0.002 -0.013 -0.000 -0.017 0.001 -0.017(0.007) (0.007) (0.007) (0.007)* (0.010) (0.010)
Travel to -0.014 0.033 -0.013 0.029 -0.009 0.033work time (0.009) (0.009)** (0.009) (0.009)** (0.012) (0.012)**log wage 0.026 -0.031 0.013 -0.037 0.032 -0.035
(0.008)** (0.009)** (0.007) (0.007)** (0.010)** (0.010)**wage -0.048 -0.015
residuals (0.013)** (0.015)satisfaction -0.003 -0.004 -0.001 -0.006with hours (0.002) (0.002) (0.003) (0.003)*satisfaction -0.004 -0.013 -0.006 -0.013
with initiative (0.002) (0.002)** (0.003)* (0.003)**satisfaction 0.001 -0.004 0.000 0.001
with work itself (0.002) (0.002) (0.003) (0.003)wants -0.014 -0.009 -0.015 -0.002
fewer hours (0.006)* (0.006) (0.009) (0.009)wants 0.022 0.005 0.027 0.007
more hours (0.012) (0.013) (0.016) (0.016)fam. Commit. -0.070 0.021
work fewer hours (0.033)* (0.022)fam. Commit. 0.009 0.045
prevent job search (0.027) (0.021)*Constant -0.324 0.084 -0.258 0.220 -0.412 0.173
(0.048)** (0.055) (0.052)** (0.054)** (0.084)** (0.077)*Observations 8761 8761 8700 8700 4545 4545
Notes. Robust standard errors reported, corrected for within group correlation (clustering).Controls for occupation, industry and time dummies included in all regressions. Standard errors reported belowthe marginal e¤ects.
33
contrast to what was true for women. As expected they are weaker determinants of both
promotions and quitting behaviour for men.
Finally, it is worth looking at the existence of children in the family and in particular the
number of children in speci…c age groups. These marginal e¤ects are also reported in tables 9 and
10. The existence of young children (aged 0 to 4) increases the probability of quitting, especially
for women but also for men. For women one extra kid aged 0-4 increases the probability of
quitting by 2.6%. The respective marginal e¤ect for men being 1%.
Overall as expected, workers in jobs with promotion opportunities are less likely to leave their
jobs and more likely to be promoted. Married people are also less likely to quit, the e¤ect being
stronger for women. Both working hours and overtime increase the probability of promotion.
Here we …nd again the positive impact of distance to work, on the probability of quit. It is
positive and signi…cant and it is a stronger predictor of quitting behaviour for women than for
men.
34
Table 10: Multinomial Logit: Promotions, Quits versus no change (Marg. e¤ects)
Ful l time men
(1) (2) (3) (4) (5) (6)promotion quit promotion quit promotion quit
Promotion 0.046 -0.012 0.045 -0.007 0.037 -0.002Opportunities (0.005)** (0.004)** (0.005)** (0.004) (0.007)** (0.005)
married -0.008 -0.008 -0.006 -0.003 -0.002 0.004Number of kids: (0.005) (0.004)* (0.005) (0.004) (0.007) (0.005)
0-4 0.003 0.010 0.003 0.010 -0.002 0.006(0.006) (0.005)* (0.006) (0.005)* (0.009) (0.006)
5-11 -0.001 -0.001 -0.000 0.000 0.002 0.003(0.004) (0.003) (0.004) (0.003) (0.006) (0.004)
12-15 -0.005 -0.009 -0.005 -0.008 -0.010 -0.013(0.005) (0.004)* (0.005) (0.004) (0.008) (0.006)*
16-18 -0.020 0.002 -0.020 -0.004 -0.028 0.000(0.013) (0.009) (0.013) (0.009) (0.019) (0.011)
number of 0.001 0.000 0.001 0.000 0.001 0.000overtime (0.000)* (0.000) (0.000)* (0.000) (0.000)* (0.000)
number of -0.000 0.001 0.000 0.001 0.000 0.000working hours (0.000) (0.000)* (0.000) (0.000) (0.001) (0.000)
TU 0.003 -0.014 0.005 -0.016 0.014 -0.023(0.005) (0.005)** (0.005) (0.004)** (0.008) (0.006)**
Travel to 0.007 0.007 0.006 0.007 0.011 0.002work time (0.006) (0.004) (0.006) (0.004) (0.007) (0.004)log wage 0.002 -0.013 -0.001 -0.024 -0.011 -0.021
(0.006) (0.005)** (0.006) (0.005)** (0.008) (0.006)**wage -0.016 -0.052
residuals (0.012) (0.010)**satisfaction 0.002 -0.003 0.001 -0.003with hours (0.002) (0.001)* (0.002) (0.001)*satisfaction -0.002 -0.008 -0.002 -0.006
with initiative (0.002) (0.001)** (0.002) (0.002)**satisfaction -0.001 -0.004 0.000 -0.005
with work itself (0.002) (0.001)** (0.003) (0.002)**wants -0.005 -0.004 -0.006 -0.002
fewer hours (0.005) (0.004) (0.008) (0.005)wants 0.007 0.013 0.006 0.008
more hours (0.009) (0.006)* (0.012) (0.008)fam. Commit. 0.015 -0.005
work fewer hours (0.022) (0.017)fam. Commit. 0.033 -0.026
prevent job search (0.024) (0.026)Constant -0.241 -0.002 -0.245 0.085 -0.163 0.061
(0.040)** (0.030) (0.047)** (0.036)* (0.065)* (0.049)Observations 11947 11947 11843 11843 6293 6293
Notes. Robust standard errors reported, corrected for within group correlation (clustering).Controls for occupation, industry and time dummies included in all regressions. Standard errors reported belowthe marginal e¤ects.
35
6 Wage growth
Having examined the determinants of job mobility separately for men and women, we now
turn to the estimation of the returns to job mobility. We want to see …rst if job movers experience
wage growth after the move and whether this is greater than the one they would have experienced
if they had not moved. Second we want to investigate if there are di¤erences in the returns to
job change between men and women. As already mentioned in the literature review in section
2, there are di¤erent ways to estimate the returns to job mobility. The …rst method consists
of estimating a wage growth equation for stayers and use this to get predicted wage growth for
movers if they had not moved. Then this predicted wage growth is compared to actual wage
growth of movers to derive the wage gains from job mobility.
We make use of the selection model proposed by Lee (1978, 1982) and applied to job mobility
in di¤erent papers, and more recently by Manning (2003). Assuming the wage growth equations
are di¤erent for movers and stayers, and dropping the time subscript, we write the one for
movers as:
¢wmi = µm1 + Xiµm2 + ²mi (11)
and for stayers:
¢wsi = µs1 + Xiµs2 + ²si (12)
The decision to move is taken according to the following rule:
M¤i = ±1 + Ki±2 + ±3(¢wmi ¡¢wsi) + vi (13)
where individual i decides to move to a new job if M¤i > 0: That is we observe ¢wmi only
if M¤i > 0 and we observe ¢wsi otherwise. Vector X contains the standard human capital
and personal characteristics of the worker and vector K contains other than wage gap factors
determining the probability of job change, described in section 5.3. This a standard model
with selection bias. OLS on equations 11 and 12 would give inconsistent estimates because
E(²mi j M¤i > 0) 6= 0 and E(²si j M¤
i · 0) 6= 0:
36
Table 11: Average Wage and Wage Growth
Average Wage Average Wage Growth(in period t) (in period t+1) (in period t+1)
Women Men Women Men Women Men
no Change 5.568 5.885 5.598 5.912 0.13 0.134promotion 5.652 5.934 5.756 6.027 0.109 0.113
quit. non family reasons 5.469 5.705 5.546 5.857 0.144 0.141quit. Family reasons 5.359 5.685 5.415 5.894 0.036 0.434
no Change 5.568 5.885 5.598 5.912 0.13 0.134promotion 5.652 5.934 5.756 6.027 0.109 0.113
quit 5.464 5.705 5.541 5.857 0.139 0.145
In table 11, we report average wages for di¤erent groups of people, by gender. In the
…rst row we report average wages for stayers, then for promoted workers and the for quitters
distinguishing between those who quit for family and non family reasons, as de…ned in section
4. We report wages both in period t (before the move, if any, is observed) and in period t+1.
In addition, in the last two columns of the table we report average wage growth for the di¤erent
groups. We notice that there is no di¤erence in average growth for stayers between men and
women. Yet wage growth for promoted individuals is higher for men than for women, although
the di¤erence is not very large. More importantly we …nd a small di¤erence in average wage
growth for those workers who quit for non family reasons, and a very large one for those who
quit for family reasons.
If we estimate log wage change equations, we are possibly introducing two types of bias. The
…rst comes from unobserved individual characteristics which may be correlated with mobility
and are included in the error terms. We use individual …xed e¤ects to correct for this …rst source
of bias. Yet this will only be solved if the unobserved individual component is constant over
time. The second bias will arise if job mobility is endogenous in an earnings equation. In other
words, expected wage growth is likely to determine job mobility. As already seen in section
5.3.1 a worker will decide to change jobs if the expected value of his current job is lower than
that o¤ered in a di¤erent job. High wages (as well as better wage growth prospects) in current
job will make job change less likely and thus the error term in a wage change equation will
be correlated with job mobility. To reduce the possibility of this source of bias we use IV and
37
Table 12: Wage gains from job mobility
(Quits versus non Change)
Women Menno Heckman Heckman selection no Heckman Heckman selection
actual ¢ws 0.124 0.133 0.128 20.132(d¢wmjM = 0) 0.056 0.134 0.021 0.135actual ¢wm 0.22 0.221 0.246 0.246average gain 0.17 0.087 0.226 0.11
Wage gains from job mobility(Quits versus Promotions and non Change)
Women Menno Heckman Heckman selection no Heckman Heckman selection
actual ¢ws 0.132 0.13 0.13 0.133(d¢wmjM = 0) 0.068 0.134 0.034 0.135actual ¢wm 0.22 0.22 0.243 0.246average gain 0.152 0.086 0.209 0.111
instrument quitting in the wage growth equation.
There are unobservable characteristics which make the choice of staying more attractive
for stayers. The same holds for unobservable characteristics which make moving worth more
for movers. Workers who decide to move are those for whom wage growth in the current job is
smaller than wage growth in the new job. This implies that using stayers to predict wage growth
for movers will underestimate the true returns to job mobility. In order to see the direction of
bias we write wage growth conditional on moving as: E(¢wsi j i moves) = µs1+Xicµs2+E(²si j i
moves), where E(²si j imoves) < 0: This implies that µs1 + Xicµs2 > E(¢wsi j i moves) and
thus we overestimate wage growth for movers if we use wage growth for stayers to estimate their
wage growth if they had stayed. This further implies that we underestimate the returns to job
change.
One can use the standard Heckman two step procedure to solve the problem of selection bias
that arises in this case. We can rewrite the wage growth equation for movers, conditional on
moving as:
¢wmi = µm1 +Xiµm2 + ½¾mÁ(g0mb±)©(g0mb±)
+ ´m (14)
38
and that of stayers conditional on staying:
¢wsi = µs1 + Xiµs2 ¡ ½¾mÁ(g0mb±)
1 ¡ ©(g0mb±)+ ´s (15)
where ¾m is the standard deviation of ²m and ½ is the correlation between ²m and the error and
g0
m contains all the variables of the …rst stage probit determining the job change decision.
The second method we can use to estimate the wage gains from job mobility is an instru-
mental variable technique. By imposing the restriction that the parameters are the same for
movers and stayers, then we can view the sample selection model as one with an endogenous
dummy variable. In that case:
¢wi = µ +Xiµ2 +¹Mi + "i (16)
with Mi = 1 if M¤i = ±1 + Ki±2 + v0i > 0 and Mi = 0 otherwise.
In this case, we need to get estimates for ±2 from a standard probit model (say eq. 13) and
compute the residuals from this model. In the second step this residual should be added in the
¢wi equation which can then be estimated for movers and stayers.
For both of these methods we require exclusion restrictions. As already mentioned we will
use job satisfaction to predict job change. More speci…cally we use job satisfaction with other
than pay factors, and particularly satisfaction with the work itself and the hours of work. We
chose these two aspects of job satisfaction which are likely to predict quits but are not correlated
with wage growth.
In table 12 we present a summary of the results from the Heckman selection model separately
for men and women. In the …rst part of the paper we report the results looking at job quitters
versus stayers (that is those who do not change jobs) whereas in the second part we include
promoted workers among the stayers. In the …rst row of each block we report the average actual
wage growth for stayers. This is very similar for men and women. In the second row we …nd
the average wage growth of movers, conditional on staying on the same job, that is using the
stayers equation with the movers characteristics to predict their wage growth had they staid
in the same job. In the second raw we report actual wage growth for movers. Wage growth
following a quit is much higher for men than for women. Men who quit their job, get on average
39
Table 13: Wage Growth: The Impact of Job Change
Full Time Workers (Quit versus non Change)Fixed E¤ects Estimates
All Women Men(1) (2) (3)
job Change OLS 0.077 0.053 0.092(0.011)** (0.016)** (0.014)**
IV 0.331 0.069 0.451(0.102)** (0.174) (0.124)**
(Note: in IV we use job satisfaction as the instrument)
the double wage, whereas the increase is smaller for women. In the …nal row of the table we
report the di¤erence between the two conditional means, which is expected to represent returns
to mobility. We …nd that the returns to quitting for men are on average twice as large than
those of women. Comparing the two blocks of table 12 we see that the average wage gain after
job quitting is slightly larger when promoted individuals are not considered among stayers.
In table 13 we report the results from the second method, using the IV technique. We only
report the …xed e¤ects estimates. The returns to quits were slightly higher in the random e¤ects
models. OLS reveals a 5-9% wage gain (in terms of log wage change) due to quits. These returns
appear to be smaller for women. In particular, job mobility increases women wage growth by 5%
versus 8% for men. When IV is used the returns to quitting go up. IV and …xed e¤ects reveal
statistically signi…cant and very large wage gains for men, of 42%. However the corresponding
number for women is only 7% and the e¤ect is not signi…cantly di¤erent from zero. We have
estimated the same IV and OLS wage growth equations for quitters versus workers who don’t
change jobs and those who get promoted. As expected the returns to job change are slightly
lower there given that we include promoted individuals in the non changers group. These results
are reported in table 15 in the appendix.
One might think that these sharp di¤erences in the returns to quits are mainly due to
di¤erent career choices of older women. For that reason we have done the same analysis for
di¤erent age groups and the results remain the same. Even for the group of young workers, who
have just entered the labour market, we …nd a positive and substantial wage gain for men but
this appears much smaller and insigni…cant for young women. We report a summary of these
results in table 14.
40
Table 14: Wage Growth: The Impact of Job Change
Full Time Workers (Quit versus non Change)IV, …xed e¤ects estimates
Age group 15-29 30-45 46+ 15-35 36+women 0.191 -0.543 1.548 0.279 -0.214
(0.207) (0.442) (1.100) (0.178) (0.557)men 0.303 0.534 1.206 0.434 0.447
(0.147)* (0.218)* (0.673) (0.128)** (0.302)
7 Conclusion
In this paper we are trying to shed more light on job preferences and the factors determining
job satisfaction separately for men and women. In addition we want to analyse the determinants
of job mobility and investigate the existence of di¤erences between male and female workers.
There is a vast literature on mobility patterns, focusing on gender di¤erences. Certain papers
in this literature focus on quits, others on promotions and others on layo¤ probabilities for men
and women. In this paper we …rst investigate the determinants of quitting versus job staying
(considering promoted individuals as stayers) and then we distinguish between promotions, quits
and stayers.
For this purpose we are using the British Household Panel Dataset which permits a distinc-
tion between voluntary and involuntary job separations. In addition it contains rich information
on job attributes and personal characteristics.
In the …rst part of the paper we look at job satisfaction. In particular we want to investigate
the existence of di¤erences between men and women in association with the job attributes which
determine their job satisfaction. In addition we can check what job satisfaction means for female
and male workers. We …nd that there are di¤erent parameters determining job satisfaction for
female and male workers. Pay is equally important for the two genders but other aspects of pay
(or monetary payo¤) are di¤erent. In addition no pecuniary aspects of the job, such as travel
to work time or working hours, appear to be more important for women.
Next we are trying to analyse the set of parameters which are likely to a¤ect men’s and
women’s mobility decisions in a distinct way. The decision to leave a job is determined by
a broad set of characteristics which include both personal characteristics of the worker and
features of the job. In particular we …rst look at family factors and household composition.
41
We expect to …nd that women are more restricted by family responsibilities and thus these
parameters should be of greater importance in determining their mobility decisions. Second we
are interested in the e¤ect of pecuniary variables (pay) on job mobility. Again, we would expect
that if women were more constrained by family commitments, they would then be less motivated
by pay. Our results suggest that non monetary aspects of the job seem to have a higher weight
in the quitting decision for women than for men. However we …nd that women are equally, if
not more, motivated by pay as men. Yet promotion prospects matter more for men, which can
be seen as evidence of more strategic career planning for them.
In the …nal part of the paper we are interested in the wage growth following a job change.
In other words we attempt to estimate the returns to job mobility. We want to investigate the
existence of di¤erent wage gains from job mobility for men and women. If women’s decisions
are motivated less by monetary payo¤s than men, then it may be that wage gains from mobility
may be lower for them. We estimate returns to job mobility …rst correcting for selection in the
two states and second using an IV technique. With both methods we …nd that wage growth
for women following a quit is much lower than that for men. This suggests that career choices
and job mobility could explain part of the gender wage gap. In the sense that women do
make di¤erent career choices from them, which is mainly due to di¤erent family and household
responsibilities and which lead to unequal gains from job moves for the two groups.
42
References
[1] Abbott M. and C. Beach (1994), ”Wage Changes and Job Changes of Canadian Women:
Evidence from the 1986-87 Labour Market Activity Survey”, The Journal of Human Re-
sources, Vol.29, Issue 2, Special Issue: Women’s Work, Wages and Well-Being, Spring 1994,
pp.429-460.
[2] Bartel, A. P., and N. Sicherman, ”Technological Change and Wages: an Interindustry
Analysis”, JPE, Vol. 107, No.2 (1999).
[3] Booth, A. and M. Francesconi (1999), ”Job Mobility in 1990s Britain: Does Gender Mat-
ter?”, unpublished paper, September 1999.
[4] Booth, A., M. Francesconi and J. Frank (2001), ”A Sticky ‡oors Model of Promotion, Pay,
and Gender”, August 2001.
[5] Booth, A., M. Francesconi and Garcia-Serrano (1999), ”Job Tenure and job mobility in
Britain”
[6] Bunchisky, M., (1999), ”Wage mobility in the US”, Review of Economics and statistics,
1999.
[7] Clark, A., A.Oswald and P.Warr, (1996) ”Is job satisfaction U-shaped in age?”, Journal of
occupational and organizational psychology”, Vol.69, pp. 57-81.
[8] Dolton, P., G.H. Makepeace and W.Van der Klaauw, (1989) ”Occupational Choice and
Earnings Determination: the Role of Sample Selection and non-pecuniary factors”,
[9] Euwals R., (1997), ”Hours Constraints within and Between Jobs”, June 1997.
[10] Euwals R., (2000), ”Female Labour Supply, Flexibility of Working Hours, and Job Mobil-
ity”, IZA Discussion Paper No. 2419, April 2000.
[11] Freed Taylor, M., J. Brice, N. Buck and E. Prentice-Lane, ”BHPS, User Manual, Volume
A, Introduction, technical summary and appendices”.
[12] Gottschalk, P., (2001), ”Wage mobility within and between job”
43
[13] C. Grund and D. Sliwka, (2001), ”The Impact of Wages Increases on Job Satisfaction-
Empirical Evidence and Theoretical Implications”, October 2001.
[14] Keith, K. and A. McWilliams (1997), ”Job Mobility and Gender-based Wage Growth Dif-
ferentials”, Economic Inquiry, Vol 35, April 1997, pp. 320-333.
[15] Keith, K. and A. McWilliams (1995), ”The Wage E¤ects of Cumulative Job Mobility”,
Industrial and Labour Review, Vol.49, Issue 1, October 1995, pp.121-137.
[16] Kidd, P., (1998), ”Job Changes, occupational mobility and human capital acquisition: an
empirical analysis”, Bulletin of Economic research, 1998**
[17] Lee, L. F., (1983) ”Generalised Econometric Models with Selectivity”, Econometrica, Vol.
51, pp. 507-512.
[18] Le Grand, C. and M. Tahlin, ”Job Mobility and Earnings Growth”, Swedish Institute for
Social Research.
[19] Levy-Garboua, L., C. Montmarquette and V. Simmonnet, (1998), ”Job Satisfaction and
Voluntary External Mobility: An Analysis based on the German Socio-Economic Panel
(1984-1994)”.
[20] Levy-Garboua, L., and C. Montmarquette (1996), ”Reported job Satisfaction: What does
it Mean?”
[21] Light, A., and K.McGarry, (1998), ”Job Change patterns and the wages of young men”,
Review of Economics and Statistics 1998.
[22] Light, A., and M. Ureta, (1995), ”Early-career work experience and gender wage di¤eren-
tials”, Journal of Labor economics, Vol. 13, Issue 1, Jan. 1995, pp. 121-154.
[23] Loprest, P., (1992), ”Gender Di¤erences in Wage Growth and Job Mobility”, The American
Economic Review, Vol. 82, Issue 2, Papers and Proceedings of the hundred and fourth
annual meeting of the American Economic Association, May 1992, pp.526-532.
[24] Lydon, R., and A.Chevalier, (2002), ”Estimates of the E¤ect of Wages on Job Satisfaction”,
CEP DP 531, May 2002.
44
[25] Maddala, G. S., (1983), ”Limited Dependent and Qualitative Variables in Econometrics”,
Cambridge University Press, 1983.
[26] Manning, A., (2003), ”Monopsony in Motion: Imperfect Competition in Labour Markets”,
Princeton University Press, 2003.
[27] McCall, B.P., (1990), ”Occupational matching: a test of sorts”, JPE, 1990.
[28] McCue, K., (1996), ”Promotions and Wage Growth”, Journal of Labor Economics, Vol.14,
Issue 2, April 1996, pp.175-209.
[29] McFadden, D., (1973), ”Conditional Logit Analysis of Qualitative Choice Behavior”, in P.
Zarembka (ed.), Frontiers in Econometrics, 105-142, Academic Press, New York, 1973.
[30] Munasinghe, L., and K. Sigman, (2000), ”Prior mobility and wages”
[31] Munasinghe, L., and T. Reif, (2001), ”The gender gap in wage returns to labor market
experience and job tenure”, June 2001.
[32] Myck, M., and G. Paull (2001), ”The Role of Employment Experience in Explaining the
Gender Wage Gap”, IFS WP 01/18, 2001.
[33] Neal, D., (1998), ”The complexity of job mobility among young men”, NBER WP 6662.
[34] Parent, D., (2000), ”Industry Speci…c Capital and the Wage Pro…le: Evidence from the
National Longitudinal Survey of Youth and the Panel Study of Income Dynamics”, JLE,
2000, vol.18, no2.
[35] Parent, D., (1999), ”Wages and Mobility: The Impact of Employer-Provided Training”,
JLE, 1999, vol.17, no2.
[36] Paxson, Christina H. and Nachum Sicherman (1996), ”The Dynamics of Dual Job Holding
and Job Mobility,” Journal of Labor Economic, 14(3):357-93.
[37] Scoones, D., and D.Bernhardt (1998), ”Promotion, Turnover and discretionary human cap-
ital acquisition”, JLE, 1998.
[38] Shaw, K., (1987), ”Occupational Change, Employer Change, and the Transferability of
Skills”, Southern Economic Journal, January 1987, Vol. 53, No 3.
45
[39] Sicherman, N., and O. Galor (1990). ”A Theory of Career Mobility,” Journal of Political
Economy, 98(1):169-92.
[40] Tackseung, J., and L. Munasinghe, (2001), ”Mobility wage gains across gender and race”
46
8 Appendix
8.1 Additional Results
Table 15: Wage Growth: The Impact of Job Change
Full Time Workers (Quits versus Promotions and non Change)Fixed E¤ects Estimates
Al l Women Men(1) (2) (3)
job Change OLS 0.066 0.042 0.081(0.010)** (0.016)** (0.014)**
IV 0.342 0.174 0.419(0.104)** (0.180) (0.126)**
(Note: in IV for each group we use job satisfaction as the instrument)
47
Table 16: Quits versus no change or promotion (Marg. e¤ects)
Binary Choice Model, Full Time Women
(1) (2) (3) (4) (5) (6) (7)Promotion -0.013 -0.010 -0.010 -0.010 -0.005 -0.005 -0.002
Opportunities (0.005)* (0.006) (0.006) (0.005) (0.005) (0.006) (0.008)married -0.020 -0.019 -0.019 -0.016 -0.015 -0.014 -0.009
(0.006)** (0.006)** (0.006)** (0.006)** (0.006)* (0.006)* (0.009)number of -0.001 -0.000 -0.000 -0.001 -0.000 -0.000 0.000overtime (0.000) (0.001) (0.001) (0.000) (0.000) (0.001) (0.001)
number of 0.000 0.001 0.001 -0.000 0.000 0.001 0.001working hours (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
TU -0.015 -0.011 -0.011 -0.014 -0.017 -0.014 -0.012(0.007)* (0.007) (0.007) (0.007)* (0.007)** (0.007)* (0.010)
Travel to 0.025 0.033 0.032 0.027 0.023 0.027 0.032work time (0.009)** (0.009)** (0.009)** (0.008)** (0.009)** (0.009)** (0.012)*log wage -0.034 -0.027 -0.033 -0.038
(0.008)** (0.009)** (0.007)** (0.010)**wage -0.024
residuals (0.015)satisfaction -0.011 -0.007with money (0.002)** (0.002)**satisfaction -0.003 -0.004 -0.007with hours (0.002) (0.002)* (0.002)**satisfaction -0.012 -0.013 -0.013
with initiative (0.002)** (0.002)** (0.003)**satisfaction 0.001 -0.003 0.000
with work itself (0.002) (0.002) (0.003)…nancial situation 0.007
better now (0.006)…nancial situation 0.012
worse now (0.008)…nancial exp 0.023
better than now (0.006)**…nancial exp 0.029
worse than now (0.012)*fam. Commit. 0.062
work fewer hours (0.037)fam. Commit. 0.017
prevent job search (0.025)Observations 9510 8659 8658 9506 9469 8593 4509
Notes. Robust standard errors reported, corrected for within group correlation (clustering).Controls for occupation, industry, region and time dummies included in all regressions. Standard errors reportedbelow the marginal e¤ects.
48
Table 17: Quits versus no change or promotion (Marg. e¤ects)
Binary Choice Model, Full Time Men
(1) (2) (3) (4) (5) (6) (7)Promotion -0.014 -0.015 -0.015 -0.009 -0.006 -0.010 -0.002
Opportunities (0.004)** (0.004)** (0.004)** (0.004)* (0.004) (0.004)* (0.006)married -0.011 -0.007 -0.009 -0.009 -0.008 -0.002 0.003
(0.004)* (0.005) (0.005)* (0.004)* (0.004)* (0.004) (0.006)number of -0.000 0.001 0.000 0.000 0.000 0.003 0.000overtime (0.000) (0.000) (0.000) (0.000) (0.000) (0.010) (0.000)
number of 0.000 0.001 0.001 0.000 0.000 -0.018 0.000working hours (0.000) (0.000)* (0.000)* (0.000) (0.000) (0.005)** (0.000)
TU -0.018 -0.016 -0.016 -0.018 -0.019 0.001 -0.027(0.005)** (0.005)** (0.005)** (0.005)** (0.005)** (0.007) (0.007)**
Travel to 0.006 0.011 0.009 0.006 0.005 0.004work time (0.004) (0.005)* (0.004)* (0.004) (0.004) (0.005)log wage -0.030 -0.015 -0.025 -0.024
(0.005)** (0.006)** (0.005)** (0.007)**wage -0.057
residuals (0.011)**satisfaction -0.011 -0.008with money (0.001)** (0.001)**satisfaction -0.002 -0.004 -0.005with hours (0.001) (0.001)** (0.002)**satisfaction -0.007 -0.009 -0.006
with initiative (0.001)** (0.002)** (0.002)**satisfaction -0.002 -0.004 -0.005
with work itself (0.001) (0.002)** (0.002)**more hours (0.009)
…nancial situation 0.005better now (0.005)
…nancial situation 0.012worse now (0.006)*
…nancial exp 0.021better than now (0.004)**…nancial exp -0.004
worse than now (0.008)fam. Commit. -0.025
work fewer hours (0.015)fam. Commit. -0.004
prevent job search (0.019)Observations 13124 11938 11938 13114 13084 11829 6348
Notes. Robust standard errors reported, corrected for within group correlation (clustering).Controls for occupation, industry, region and time dummies included in all regressions. Standard errors reportedbelow the marginal e¤ects.
49
8.2 Graphs
Figure 1: Proportion of men, women who do not change jobs
year
men women
1990 1995 2000
80
85
90
Figure 2: Proportion of men, women with kids who do not change jobs
year
men women
1990 1995 2000
80
85
90
50
Figure 3: Proportion of men, women without kids who do not change jobs
year
men women
1990 1995 2000
75
80
85
90
Figure 4: Proportion of men, women who get promoted
year
men women
1990 1995 2000
6
8
10
12
Figure 5: Proportion of men, women with kids who get promoted
year
men women
1990 1995 2000
6
8
10
12
51
Figure 6: Proportion of men, women without kids who get promoted
year
men women
1990 1995 2000
6
8
10
12
Figure 7: Proportion of men, women who quit
year
men women
1990 1995 2000
0
5
10
Figure 8: Proportion of men, women with kids who quit
year
men women
1990 1995 2000
4
6
8
10
12
52
Figure 9: Proportion of men, women without kids who quit
year
men women
1990 1995 2000
0
5
10
15
53