DISCUSSION PAPER SERIES
IZA DP No. 1184
Gender Differences in Job Assignment and
Promotion on a Complexity Ladder of Jobs
Tuomas Pekkarinen
Juhana Vartiainen
June 2004
Forschungsinstitut
zur Zukunft der Arbeit
Institute for the Study
of Labor
Gender Differences in Job
Assignment and Promotion on
a Complexity Ladder of Jobs
Tuomas Pekkarinen
Nuffield College, Oxford
and IZA Bonn
Juhana Vartiainen
FIEF, Stockholm
Discussion Paper No. 1184
June 2004
IZA
P.O. Box 7240
53072 Bonn
Germany
Phone: +49-228-3894-0
Fax: +49-228-3894-180
Email: iza@iza.org
This paper can be downloaded without charge at:
http://ssrn.com/abstract=562451
An index to IZA Discussion Papers is located at:
http://www.iza.org/publications/dps/
Any opinions expressed here are those of the author(s) and not those of the institute. Research
disseminated by IZA may include views on policy, but the institute itself takes no institutional policy
positions.
The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center
and a place of communication between science, politics and business. IZA is an independent nonprofit
company supported by Deutsche Post World Net. The center is associated with the University of Bonn
and offers a stimulating research environment through its research networks, research support, and
visitors and doctoral programs. IZA engages in (i) original and internationally competitive research in
all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research
results and concepts to the interested public.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be
available on the IZA website (www.iza.org) or directly from the author.
IZA Discussion Paper No. 1184
June 2004
ABSTRACT
Gender Differences in Job Assignment and Promotion
on a Complexity Ladder of Jobs∗
This paper studies gender differences in the allocation of workers across jobs of different
complexity using panel data on Finnish metalworkers. These data provide a measure for the
complexity of the workers' tasks that can be used to construct a complexity ladder of jobs.
We study whether women have to meet higher productivity requirements than men in order to
be assigned to more complex tasks. Gender differences in the promotion rates are examined.
We use productivity measures that are based on the supervisors' performance evaluations
and examine gender differences in the productivity of promoted and non-promoted workers. It
is found that women start their careers in less complex tasks than men and that they are also
less likely to get promoted than men who start in similar tasks. When we compare the
productivity of men and women, both at the initial assignment and when some of these
individuals have been promoted, we find that there is no gender-related productivity
differential at the time of the initial assignment, but women become on average more
productive than men afterwards, both in promoted and non-promoted subsets. The most
plausible interpretation of these results is that women face a higher promotion threshold than
men.
JEL Classification:
Keywords:
J0, J7
promotions, gender wage gap, discrimination
Corresponding author:
Tuomas Pekkarinen
Nuffield College
Oxford, OX1 1NF
United Kingdom
Email: tuomas.pekkarinen@nuffield.oxford.ac.uk
∗
We would like to thank David Autor, Andrea Ichino, Ian Jewitt, Markus Jäntti, Frank Vella and Rudolf
Winter-Ebmer for helpful comments. The usual disclaimer applies. We are grateful to the
Confederation of Finnish Industry and Employers (TT) for the permission to use their wage records.
The data are archived at the Labour Institute for Economic Research, Helsinki, and the permission to
use the data are controlled by the Confederation of Finnish Industry and Employers. Pekkarinen also
thanks the Academy of Finland, Finnish Work Environment Fund and Yrjō Jahnsson Foundation for
financial support.
1 Introduction
The gender wage gap is a persistent phenomenon. Despite the fact that in many countries the female participation rates have been high for decades and the experience on
anti-discriminatory legislation is long, women tend to earn lower wages than men. An
important part of this wage gap is explained by occupational segregation. It is common to
…nd, that when su¢ ciently narrow occupational categories are controlled for, the gender
wage gap is considerably reduced.1 Hence, one of the key elements in understanding the
gender wage gap is the asymmetric allocation of men and women across occupations.
This paper uses panel data on Finnish metalworkers to study gender di¤erences in the
allocation of workers across jobs of di¤erent complexity. In the Finnish metal industry,
as in many other industries, women typically work on less complex jobs than men. Our
aim is to …nd out, whether this happens because women need to meet higher productivity
requirements than men to be assigned to more complex jobs. We address this question
by examining gender di¤erences in the initial job assignments and in the probability of
promotions from the initial jobs. Furthermore, we use a measure of individual productivity that is based on supervisors’performance evaluations to study how promoted and
non-promoted women perform with respect to their male counterparts. We then use this
information to infer whether the male and female productivity thresholds of assignment
are di¤erent.
Very common explanation for gender di¤erences in the assignment thresholds is tastebased discrimination. But there are also theoretical arguments that do not rely on discriminatory behavior on the part of the employers. Most notably, in the model by Lazear
and Rosen (1990), the comparative advantage of women in non-market activities makes
them more likely to quit than men. Consequently, women have to be more productive
than men to be assigned to complex jobs.
Previous empirical literature on gender di¤erences in job allocation has almost exclusively concentrated on estimating gender di¤erences in the probability of promotion.2
There are studies that use data from large surveys such as Winter-Ebmer and Zweimüller
(1997) using Austrian census data, McCue (1996) using the PSID, and Booth et al (2001)
using the BHPS. Other studies have used data from a single industry or …rm: Granqvist
and Persson (2002) analyze gender di¤erences in the career mobility of workers in the
Swedish retail trade industry, Hersch and Viscusi (1996) focus on one US public utility,
and Jones and Makepeace (1996) look at workers in a British …nancial institution. A
special branch of the literature are the studies on the career advancement of academics
like Ginther and Hayes (1999, 2003) and McDowell et al (1999). The common …nding is
that women are less likely to get promoted than men.
But gender di¤erences in the assignment thresholds have implications that go beyond
promotion probabilities. The promotion process will also improve the relative productivity
of women within jobs. Since the promoted women have passed a higher threshold than
promoted men, they will, on average, be more productive than men who are promoted to
the new job. On the other hand, the men that remain in the less complex jobs have failed
to meet a lower productivity threshold and are therefore less productive than women who
remain in the same job.
Here, we will directly address the question of whether women have to be more productive than men to be assigned to complex jobs. We focus on the early careers of workers
who are initially assigned to jobs of similar complexity. The idea is that the productivity
di¤erences between men and women who are initially assigned to the same jobs should be
1 See
for example Blau and Ferber (1987) or Groshen (1991).
are Booth et al (2001) and Hersch and Viscusi (1996) who also estimate gender di¤erences
in the wage growth upon promotion. Both …nd that women gain less from promotions than men.
2 Exceptions
2
so small that we can draw conclusions about the assignment thresholds by examining the
gender di¤erences in the probability of promotions and in productivity among promoted
and non-promoted workers. If women are at the same time less likely to be promoted
than men who start in similar tasks but also more productive than men among promoted
and non-promoted workers, we conclude that the promotion threshold must be higher
for women than for men. Finally, we use the estimated male and female quit rates to
assess whether the gender di¤erences in reservation wages could plausibly explain the
asymmetry in the job assignment.
We believe that the Finnish metal industry data are suitable for this kind of analysis.
First of all, they provide a measure for the complexity of the jobs. This complexity measure is valid for both within- and between-…rm comparisons. In this industry, all the jobs
are evaluated according to their complexity, and on the basis of this evaluation a minimum
wage is attached to each job. We use these minimum wages to construct a complexity
ladder of jobs. Second, the panel nature of the data allows us to distinguish between
the initial job assignments and subsequent promotions. We can therefore compare the
careers of men and women who start in jobs of similar complexity. Finally, the data
provide information on bonuses that are based on individual performance evaluations, so
that we do not have to rely on …nal wages when measuring individual productivity.
Using these features of the data, we …nd that women are initially assigned to less
complex jobs than men and that they are also less likely to get promoted than men who
start in the same jobs. When we measure individual productivity with bonuses, we …nd
that there are no clear gender di¤erences in productivity at the initial job assignment.
However, gender di¤erences in productivity become clear as some of the workers are selected for promotion. Women are consistently more productive in the groups of promoted
and non-promoted workers. We interpret these results as supportive of the asymmetric
threshold hypothesis.
In this industry, women are also more likely to quit than men. These di¤erences
in quit rates are especially signi…cant among younger workers, who are the most likely
candidates for promotion. However, as the di¤erences in quit rates may also be a result
of the di¤erent promotion prospects, we can only say that the quit behaviour of men and
women in this industry is consistent with, but not necessarily exclusively supportive of,
the reservation wage explanation of gender di¤erences in assignment thresholds.
The rest of the paper is organized as follows. In the following section, we summarize
the familiar Lazear and Rosen (1990) model and discuss the implications regarding the
job assignments and gender productivity di¤erences within jobs. We also point out what
di¤erent assumptions about the underlying male and female productivity distributions
imply for the inference on the assignment thresholds. In the third section, we discuss the
data and the fourth section explains the construction of the complexity ladder of jobs.
We then study the movement of workers on the complexity ladder. In the …fth section, we
examine the allocation of workers across levels of complexity at the initial job assignment
and replicate the standard analysis of gender di¤erences in the probability of promotion.
The most original analysis is reported in section 6, in which we use our productivity
proxies to evaluate the relative performance of a given group of newly recruited men and
women, both before and after the …rst promotion decisions have taken place. In section
7, we brie‡y examine the quit behavior of the workers and discuss whether the gender
di¤erences are large enough to explain the di¤erences in the job allocation. Section 8
concludes.
3
2 Theoretical background
The model by Lazear and Rosen (1990) is a popular framework in which to think about
gender di¤erences in job assignment. In this model, the workers who are assigned to more
demanding jobs undergo costly training to learn the tasks involved. Once the training is
…nished, the workers cannot commit to staying with the employer but leave the …rm, if
the value of their reservation wage exceeds the wage paid by the …rm.
Women are assumed to have higher reservation wages than men because of their
comparative advantage in non-market activities. Hence, women are more likely to leave
the …rm and the risk of loosing the training investment is higher when a woman is assigned
to the demanding job. Thus, the productivity threshold that determines whether the
worker is assigned to the demanding job is higher for women than for men.
2.1 Job assignments
Consider workers who enter the …rm with some level of innate ability and who gradually
acquire seniority that increases their e¤ective ability. Denote the innate ability of worker i
with i and his/her seniority prior to period t with xit . Following Gibbons and Waldman
(1999), assume that the e¤ective ability of the worker, it , is a function of the innate
ability and seniority: it = i f (xit ), where f 0 > 0; f 00 0.
There are two jobs, the more demanding job A and the less demanding job B. Consider
the employer’s job assignment decisions at time t. If the worker is assigned to job A,
he/she will undergo costly training during the fraction of the period t and reaches the
full productive capacity for the fraction (1
) of the period. One way to interpret is
to view it as the time spent in job speci…c monitoring, as in Lazear (1986). There is no
training in task B:
The output per worker in job B during the whole period t is equal to his/her e¤ective
ability it = i f (xit ). If the worker is assigned to job A, his/her output is reduced to
), the output of the worker
1 it = 1 i f (xit ) during , where 0 < 1 < 1. During (1
in job A reaches 2 it = 2 i f (xit ), where 2 > 1.
Assume that the workers are certain to stay in the …rm during but that they cannot
commit to staying during (1
). Workers will leave the …rm if the value of their reservation wage is higher than the wage paid by the …rm. Lazear and Rosen assume that the
worker’s reservation wage is a random variable ! and that at the beginning of period t
only the the cdf, F (!), of this variable is known.
An e¢ cient assignment rule should induce the workers to stay in the …rm, if the market
value of their output exceeds their reservation wage. Furthermore, the rule should assign
only those workers to job A, whose social output is higher in job A than in job B.
Social output of the workers who are assigned to job A is equal to:
!
Z
Z
1
2 it
1 it
+ (1
)
dF +
2 it
0
!dF
Correspondingly, the social output of the workers in job B is:
!
Z
Z
1
it
it + (1
)
dF +
it
0
!dF
(2)
it
The di¤erence between (1) and (2) can be written as:
Z 2 it
D=
(1
)
+
(1
)
F (!)d!
it
1
it
4
(1)
2 it
(3)
Workers are assigned to task A if D > 0 and to task B if D < 0. Hence, there is an
assignment threshold D( ) = 0.
The employers do not observe the innate abilities, when making the initial assignment
decisions. The e¤ective ability of the worker is revealed after these assignments and the
subsequent promotion decisions are based on the observed e¤ective ability. Thus, for each
level of seniority there is an innate ability threshold of promotion, , and for each level
of innate ability there is a seniority threshold of promotion, xt
2.2 Gender di¤erences in reservation wages
Lazear and Rosen assume that men and women have di¤erent outside options. More specifically, female distribution of reservation wages, Ff (!), …rst-order stochastically dominates
that of men, Fm (!). To introduce this assumption in the setting above, write the distribution of reservation wages as F (!; ), where is a shifter, such that @F=@ > 0.
De…ne Ff (!)
F (!; f ) and Fm (!)
F (!; m ) so that m > f , which implies that
Ff (!) < Fm (!).
Di¤erentiating D( ) = 0 with respect to yields:
d
d
=
R
@F (!; )
d!
@
2
@D( )=@
it
which is negative since @D( )=@ it > 0. Hence, decreases the threshold value of
e¤ective ability
and because m > f we have that m < f . This means that in
order to be assigned to the complex job women need to have a higher innate ability than
men with the same level of seniority. Similarly, women need to acquire more seniority
than men of equal innate ability to get promoted to the complex job.
2.3 Implications of the gender di¤erences in reservation wages
Gender di¤erences in the assignment thresholds have implications for the job assignments
and productivity di¤erences within jobs. The direction in which these impacts work
depends on the underlying ability distributions of men and women.
Denote the male ability distribution with Gm ( ) and the female distribution with
Gf ( ). Women will be less likely to be assigned to complex job, if:
Pr(
m)
=1
Gm (
m)
1
Gf (
f)
= Pr(
f)
Furthermore, women will be on average more productive within jobs if:
Z 1
Z 1
E[ j
]
=
dG
(
j
)
dGf ( j
j
m
m
m
f) = E
m
(4)
f
(5)
f
Below, we will use gender di¤erences in promotion rates and productivity within
groups of promoted and non-promoted workers to infer whether the assignment thresholds
are di¤erent for men and women. In order to do this, we need to know under which
assumptions the fact that (4) and (5) hold simultaneously necessarily implies m
f.
It is easy to see that if Gm ( ) and Gf ( ) are identical, both (4) and (5) can only hold
if m
f . When the male ability distribution …rst-order stochastically dominates the
female distribution, 1 Gm ( m ) 1 Gf ( f ), the condition (4) can also hold for some
values m > f but the condition (5) holds only when m
f . On the other hand,
if the female distribution dominates, 1 Gm ( m )
1 Gf ( f ), the condition (4) can
only hold if m
f , whereas the condition (5) also holds for some values such that
>
.
Thus,
in
these
cases women can be simultaneously less likely to be assigned to
m
f
5
the complex job and more productive within jobs only if the female assignment threshold
is higher than the male assignment threshold.
Naturally, one cannot draw such conclusions if male and female ability distributions
intersect. It might appear to be a very strong assumption to rule out the intersecting
ability distributions. After all, it is often argued that the variance of male ability is
higher than that of the female ability. However, in our case we are not comparing the
"global" male and female ability distributions. Instead, we focus on "ability brackets" of
men and women who have been initially assigned to similar tasks by their employers. It
seems very unlikely that the male and female ability distributions would systematically
intersect within all of these ability subsets. Indeed, in our empirical analysis we will show
that the male and female distributions of measured productivity seem to be remarkably
similar within the initial jobs. Hence, in these circumstances, observing that women are
both less likely to be promoted and more productive than promoted and non-promoted
men implies that the female assignment threshold is higher than the male one.
3 The data
The data used in this paper come from the wage records of the Confederation of Finnish
Industry and Employers (Teollisuus ja työnantajat). The wage records contain detailed
information on the wages and working hours of all the workers who are a¢ liated with the
confederation. In the Finnish metal industry, this covers practically 100% of the …rms.
The wage records’data on wages and working hours can be considered as exceptionally
reliable since the information comes directly from the …rms’wage accounts. However, the
information on the individual characteristics is rather scarce. Basically only age, gender,
and seniority can be identi…ed from the raw data. For the objectives of this paper, the
most disturbing piece of missing information are the variables concerning marital status
of the worker and the number of dependent children.
In this paper, we use all the workers who start their careers in the Finnish metal
industry between 1990-1995 and whom we can follow for at least …ve years up to year
2000 when our data end. We chose to restrict the sample like this because for our purposes
it is essential to observe the workers at their initial job assignments and follow them for
a reasonable number of years. Furthermore, restricting the analysis to workers who stay
for more than four years selects the workers with a strong labour market attachment.
This panel of newcomers to the metal industry has 83 474 employee/year observations
on 11 661 workers of whom 2 705 (23 %) are women. We have 64 198 episodes where
the both current and next year’s jobs are observed and the worker stays within the same
…rm.
In table 1, we present the descriptive statistics on this sample at the …rst year of
seniority and compare them with the full cross-section of workers in the metal industry
in 1990. It is clear that this is a strongly male-dominated industry. This fact should be
taken into account, when interpreting the results reported below.
3.1 Wage determination in the Finnish metal industry
The sample was restricted to include only workers from the metal industry because the
peculiar wage determination process in this industry provides particularly interesting information on the complexity of the jobs. According to the industry’s collective agreement,
wages should be determined by the complexity of the job, the individual performance of
the worker, and by various individual and …rm-speci…c factors.
The complexity of the job speci…es a minimum wage for each job. This minimum wage
is called the occupation-related wage. Worker’s individual performance on the job a¤ects
6
the wage outcome through a personal bonus of 2 to 17% of the occupation-related wage.3
In this paper, we will use the occupation-related wages as a measure of the complexity of
the job and personal bonuses as a proxy for the productivity of the individual.
3.2 Job complexity
The complexity of the job is evaluated with a grading system that is similar to the ones
used in some large establishments in the US. The evaluation is carried out by a group of
experts, that considers various aspects of the jobs and assigns them points according to
their complexity. The complexity level is based on three criteria: 1) how long does it take
to learn the tasks, 2) the degree of responsibility, and 3) the working conditions. The
evaluation should be independent of the workers performing the job and it should make
jobs comparable both between and within …rms.
Based on the evaluation of jobs, an occupation-related wage is determined for each job
in the collective agreement. The more demanding the job, that is the more complexity
points it gets, the higher is the corresponding occupation-related wage. Basically, there
should be a one-to-one mapping from the occupation-related wages to the complexity
points. The occupation-related wages can therefore be interpreted as a continuous variable
that measures the complexity of the job. There are typically around 50 di¤erent levels of
occupation-related wages per year.
3.3 Individual performance
The individual performance of the worker is evaluated by his/her immediate supervisor.
The performance evaluation is based on three criteria: 1) how well the tasks are carried
out, 2) the worker’s output relative to what is considered normal in the job, and 3)
how well the worker follows the instructions and regulations. The performance of the
worker is always evaluated relative to what is considered normal on the job in question.
Hence, if, for example, the output of all the workers in the …rm increases, the performance
evaluations of the individual workers should not be a¤ected.
Based on the supervisor’s evaluation each worker is paid a personal bonus that should
amount to 2-17% of the worker’s occupation-related wage. The collective agreement
states, that the bonuses should be symmetrically distributed around the mean of 9.5%
within each complexity tercile in the …rm.
In order to check whether these principles of the collective agreement are also followed
in practice, we examined the variation of bonuses across complexity levels for groups of
workers with a given level of seniority. There was some indication that bonuses were
increasing in the level of complexity. Furthermore, straightforward analysis of variance
revealed that …rm dummies explain approximately 5% of the variation in bonuses. In order
to account for this variation across complexity levels and …rms, we measure individual
performance using the deviation of the worker’s personal bonus from the …rm - complexity
level cell mean. In the following, this productivity measure will be called personal bonus
deviation.
4 Complexity ladder of jobs
As was explained above, the occupation-related wages order the jobs according to their
complexity. In this paper, we use this ordering of jobs as a job ladder where the within3 It is important to note that the sum of the occupation-related wage and personal bonus is not the
…nal wage outcome. The …rms are free to set their …nal wages as long as they are above the minimum
levels set at the collective agreement. Thus, there is considerable within and between …rm variation in
wages even among workers in similar tasks.
7
…rm upwards movement on the ladder is interpreted as a promotion and downwards
movement as a demotion.
The fact that the occupation-related wages are a component of the …nal wages makes
their use as a complexity measure somewhat problematic. There is some year-to-year
variation in the occupation-related wages that is not related to changes in the complexity
of the jobs. For example, in certain years all the occupation-related wages are increased to
account for the e¤ects of in‡ation. This means that the scale with which the complexity
of the jobs is measured is not constant in time.
We corrected for these changes by descaling the occupation-related wages in the following way. We …rst grouped the workers according to their occupation-related wages within
each year and examined the within group distributions of changes in the occupationrelated wages. This analysis revealed that for most of the workers in these groups
the year-to-year changes in occupation-related wages were identical. We interpreted the
group mode changes of the occupation-related wages as increases that were not related
to changes in the complexity of the jobs. The occupation-related wages were then corrected by substracting the annual mode changes from occupation-related wages. After
this correction, the occupation-related wages form a consistent ladder for all the years
1990-2000.
Table 2 is a transition matrix of movements between complexity levels in 1990-1991.
It shows all the within-…rms movement between complexity levels, including entries, exits
and stays as percentages of the number of workers on each level. The table can be
interpreted in the same way as the job-to-job transition matrices in Baker et al (1994) or
Treble et al (2001). For the purposes of this table, the complexity levels were aggregated
into integers.
If complexity levels really are a true job ladder, the movements on this ladder should
resemble the stylized facts about promotions and demotions. Basically, there should be
more movement upwards than downwards. In table 2, there are 62 304 workers who
stayed with the same employer between 1990-1991. Approximately 9% of these workers
worked in a more complex and about 5% in a less complex job in 1991 than in 1990.
Shaded areas in the table indicate the levels that were the most frequent destinations of
movement between complexity levels. There is a considerable amount of movement, most
of which is upwards and rarely leaping over many levels. In our view, the information
in table 2 is in line with the stylized facts about promotions and demotions. It seems
appropriate to interpret the complexity axis as a job ladder.
5 Gender di¤erences in job assignments
How are men and women then allocated on this ladder? Figure 1 plots the percentages of
male and female workers across the same complexity levels as in table 2 in the 1990 crosssection. The …gure reveals that women are concentrated at the low end of the complexity
axis while most men work on more complex jobs. This pattern is repeated throughout
the years 1990-2000.
Why are women concentrated at the low-end of the complexity axis as in …gure 1?
Perhaps the most common approach to studying gender di¤erences in job allocation, is
to examine movement of workers between jobs. If the assignment thresholds are identical
for men and women, we shouldn’t see any di¤erences in the movement of men and women
once the productive characteristics are controlled for. In this section, we study the gender
di¤erences in the allocation of workers across jobs of di¤erent complexity at the initial
job assignment and in the promotion process.
8
Figure 1: Percentages of male and female workers across tasks of di¤erent complexity,
1990 cross-section
5.1 Initial job assignments
In our sample, we observe the jobs where the workers started their careers and their
subsequent promotions and demotions. In …gure 2, we have plotted the distributions of
the …rst job assignments of the men and women in our sample.
The asymmetry in …gure 2 is striking. Women seem to start their careers in clearly
less complex jobs than men. The Duncan and Duncan index of dissimilarity for the initial
job assignments gives a value 49:3. That is, nearly half of the women in our sample should
change their initial job assignment to arrive to the male allocation of jobs.
However, …gure 2 should not be interpreted as evidence on gender di¤erences in the
assignment thresholds. After all, it is possible that the asymmetry in …gure 2 only re‡ects
di¤erences in the ability distributions of men and women who choose to work in the metal
industry. If the underlying productivity of women in this industry is lower than that of
men, it is not surprising that women start their careers in less complex jobs than men.
Still, whatever the reason for the asymmetry in …gure 2, it is clear that the initial job
assignment has to be taken into account in the analysis of gender di¤erences of promotion
and productivity.
5.2 Promotions
Are women then less likely to move upwards on the complexity ladder than men? In
table 3, we report the gender di¤erences in the change of complexity. We take advantage
of the fact that the occupation-related wage is a continuous variable. Thus, the changes
of occupation-related wages conveniently summarize both the extent and the direction of
the change in the complexity of the jobs that workers are performing.
The …rst row of table 3 reports the sample means of changes in log occupation-related
wages for men and women in our sample. The male and female sample mean of changes
in the complexity of the jobs are virtually identical. If anything, the female mean is
9
Figure 2: Percentages of male and female workers across jobs of di¤erent complexity at
the initial job assignment
slightly higher than the male mean, although the di¤erence is not statistically signi…cant.
However, once we split the workers according to the complexity of their initial job assignments, the gender di¤erences become clear. In the following rows of table 3, we report
the male and female means of changes in log occupation-related wages in the complexity
groups of initial job assignment. In most groups, the mean change in log complexity is
clearly lower for women than for men.
The numbers in table 3 are just cell means which do not control for anything else than
the complexity of the initial job assignment. In order to control for observable productive
characteristics, we ran a following regression on the pooled data:
0
cit = Fi + Xit
+ Iit + "it
(6)
where cit = ci;t+1 cit is the di¤erence in log occupation related wages, Fi is the
female dummy, and Xit is a set of productive characteristics including age and seniority
and their squares. In Iit we include a full (15) set of initial job complexity dummies to
control for the initial job assignment.
The regression results are reported in table 4. The coe¢ cient on the female dummy is
again negative, 0:008, and clearly signi…cant. Thus, the results in tables 3 and 4 clearly
indicate that women take smaller steps on the complexity ladder than men who start in
jobs of similar complexity.
Another way of looking at gender di¤erences in the promotion patterns is to examine
duration to promotion. In order to do this, we de…ned a promotion indicator that takes
value one if the individual experienced a positive change in the occupation-related wage
within the same …rm and zero otherwise. In table 5, we report the Kaplan-Meier estimates
of the survivor function for men and women. The …rst column shows the estimates for
the whole sample. Although the female survivor function is consistently above the male
one, the di¤erences are relatively small.
10
However, the gender di¤erences become clear once we focus on workers who start their
careers in jobs of more or less similar complexity. This is done in columns 2-4 of table 5,
where we report the estimates of the survivor function within subsamples of workers. The
groups were formed by dividing the workers into "low-complexity", "medium-complexity",
and "high-complexity" groups according to the complexity of the initial job assignment.
Female survivor functions are clearly above the male ones in the low and medium groups
of initial job complexity. A straightforward log-rank test for the equality of the KaplanMeier estimates, results of which are reported in the last row of table 5, rejects the null
in all the cases expect the high complexity jobs.
In order to control for observable productive characteristics in the duration analysis,
we ran a discrete-time proportional hazards model of promotion, using the log-likelihood
suggested by Jenkins (1995):
log L =
N X
T
X
i=1 t=1
N
yit
T
XX
hit
+
log(1
(1 hit )
i=1 t=1
hi;t )
(7)
where N is the number of individuals in our sample, T is the number of years, and
yit is a dummy that takes the value 1 for individual i at t if he/she is promoted at that
period and zero otherwise. We use a complementary loglog speci…cation for the hazard
rate:
0
hit = 1 exp f exp [ (t) + Fi + Xit
+ Iit ]g
(8)
where the baseline hazard (t) is left unspeci…ed and the rest of the variables are as in
(6). The results are presented in table 6. The estimated coe¢ cient of the female dummy
is negative ( 0:4) and clearly signi…cant.
We interpret the results in tables 3-6 as implying that women are initially assigned to
jobs that are less complex but where the promotions are also more frequent. However,
compared to the male workers who start their careers on those same jobs, women are
much less likely to move to more complex tasks. Thus, women tend to "get stuck" in
the initial jobs of low complexity whereas men are more successful in advancing to more
complex jobs.
6 Gender di¤erences in productivity
The regressions in the previous section controlled for productive characteristics in a very
crude way. An alternative approach to study gender di¤erences in the assignment thresholds is to compare the relative productivity of men and women before and after the
promotion decisions have taken place. If the threshold of promotion is higher for women
than for men, the promoted and non-promoted women should be, on average, more productive than their male counterparts.
We now use personal bonus deviations to measure individual productivity and examine the productivity of the workers in our newcomer sample. We …rst measure the
productivity of each worker at the initial job assignment and then partition the workers
into two sets of workers: those who have got promoted up to some speci…c year (the
"promoted" group) and those who have stayed at the initial assignment until that year
(the "stagnant" group).
In table 7, we report the median personal bonus deviations for workers in each of these
groups. In the …rst row, we display the personal bonus deviations of the new entrants
during their …rst year in the industry. The next row depicts the personal bonus deviations
of men and women as measured in the second year of the career, separately for those who
were promoted after the initial year and those who were not. The following row depicts
11
the median personal bonus deviations two years after the initial assignment, similarly
di¤erentiated between those who had been promoted up to their third year of the career
and those who were not; and analogously for the fourth and the …fth row.
The medians reported in table 7 indicate that there are hardly any gender di¤erences
in productivity at the initial task assignments. However, once the workers are split into
promoted and non-promoted groups, the gender di¤erences become clear. Women tend
to dominate in both groups. Among the non-promoted workers the gender di¤erence
reaches 0:009 and among the promoted workers 0:011 in favour of women.
Table 7 only reports the di¤erences in medians. In order to have a better view of
what happens to the productivity distributions, we have plotted the kernel estimators of
the cumulative distributions of personal bonus deviations for men and women separately
for stagnant and promoted workers in the …rst four years of seniority in …gure 3.4 The
male and female distributions of personal bonus deviations are almost indistinguishable
at the initial job assignment. After the …rst year, the workers are split into stagnant and
promoted groups. In both groups, the female distributions clearly dominate the male
distributions and the di¤erences seem to become clearer with seniority.
Whether the results presented above, together with the results on gender di¤erences
in promotion probabilities, imply that the female assignment thresholds are higher than
the male ones depends on the underlying cumulative productivity distributions of men
and women who choose to work in the metal industry. This implication only follows if the
underlying cumulative productivity distributions of men and women do not intersect. As
we argued in section 2.3, it seems plausible to assume that this is the case here. First of
all, we focus on men and women who start their careers within jobs of similar complexity.
The employers have already selected the workers that they judge to be appropriate for
given tasks. Any di¤erences in the underlying distributions of productivity of men and
women within these groups should be small. Furthermore, it seems very unlikely that
distributions would intersect within all these groups. Finally, the results in table 7 and
in …gure 3 indicate that there are no visible gender di¤erences in the personal bonus
deviations within initial jobs. Hence, in our view, it is justi…ed to conclude that these
results as indicating that women need to be more productive than men to be assigned to
complex jobs.
Finally, personal bonus deviation comparisons also tell us something about the behaviour of the employers. First of all, it is clear that the workers that the employers choose
to promote tend to be more productive already in their initial jobs. In terms of Lazear
and Rosen story, this implies that training costs do play a role in these …rms. After all,
if the assignment to more complex jobs was costless, the employer would assign all the
workers to more complex tasks. However, it seems that women need to be more productive at their initial job than men to get promoted. For promoted women, the average
personal bonus deviation in the initial job is :021 while for promoted men it is only
:025. Thus, promoted women are approximately 18% more productive at the initial job
than promoted men. Since the Lazear and Rosen model does not assume any di¤erences
in the training costs of men and women, the only way to explain these di¤erences in terms
of their model are gender di¤erences in the quit behaviour.
7 Gender di¤erences in quit behavior
According to the Lazear and Rosen model, gender di¤erences in reservation wages give rise
to di¤erences in the assignment thresholds. We will now examine whether the observed
4 The estimation was done using the Epanechnikov kernel with a bandwidth of .005. Di¤erent bandwidths were tried out without relevant changes in the qualitative results.
12
Figure 3: Kernel density estimates of the cumulative distributions of personal bonus
deviations
13
Figure 4: Average quit rates by age and gender in the whole metal industry population,
1990-2000
quit behaviour of men and women in the Finnish metal industry is consistent with the
Lazear and Rosen argument. However, we emphasize that, in our view, these data do
not really allow for any exclusive "testing" of the Lazear and Rosen argument against all
alternative models such as taste-based discrimination in promotions. After all, if there
are other forces that hamper women’s promotion prospects, women’s higher exit rates
will also re‡ect these other mechanisms. That is, any gender di¤erences in quit behaviour
that we observe may be caused by gender di¤erences in the assignment thresholds which
themselves could be a result of taste-based discrimination. Hence, this exercise should be
seen more as a simple check of whether gender di¤erences in quit behaviour could be the
underlying cause for the di¤erences in the assignment thresholds in the most favourable
case where the quit rates are exogenous.
We do not observe the actual reasons of job separations in our data. Hence, it is
not possible to distinguish layo¤s and voluntary quits directly. However, since voluntary
quits are what we are interested in, we tried to tackle this problem by de…ning as a "quit"
those separations, where more than 75% of the employees remain in the departure …rm.
This de…nition rules out the situations where a large proportion of the …rm’s workforce
is laid o¤ at the same time, and it is likely to catch more truly voluntary separations.
We do not distinguish between job-to-job turnover and job-to-nonemployment turnover,
since from the point of the view of the employer these are equivalent.
In the whole metalworker population, the average quit rate of men was 7.8% and that
of women 9.6%. Figure 4 plots the average quit rates for men and women by age in the
whole metalworker population. As is clear from the …gure, the male and female quit rates
di¤er signi…cantly among younger workers, but tend to converge among older workers
Table 8 reports the marginal e¤ects from a probit model of quits that uses the whole
population of metalworkers. In this model, we allow for di¤erent intercepts and age pro…les for men and women. The e¤ects are calculated at the mean values of the continuous
14
independent variables. According to these results, women are …ve percentage points more
likely to leave the …rm.
However, from the point of view of the employer who is considering whether to promote one of the workers in our newcomer sample, the most relevant di¤erences are the
di¤erences in the quit rates of the younger workers. In table 9, we report the predicted
quit rates of men and women in di¤erent age groups that are calculated using the coef…cients of the probit model in table 8. The predicted quit rates of men and women are
substantially di¤erent among workers who are younger than 30 years old. In fact, in both
groups 25 and below as well 25-30 years old, the female quit rate is approximately 40%
higher than the male quit rate. In our newcomer sample, nearly 50% of the observations
come from these two age groups.
Thus, we …nd a pattern of exit rates that is certainly consistent with the Lazear and
Rosen story but, of course, not exclusively supportive of that model only. The women are
more likely to quit the …rm precisely in the group of workers where one would expect the
Lazear and Rosen argument about gender di¤erences in the reservation wages to apply,
i.e. young workers. Interestingly, when the stochastic dominance analysis of …gure 3 is
replicated for the population of over 40 years old workers, one does not …nd a similar
pattern of gender productivity di¤erences among promoted and non-promoted workers as
in …gure 3.
8 Conclusions
The asymmetric allocation of men and women across jobs is one of the most common
explanations for the persistence of gender wage gaps. It is often argued that women
have to meet higher productivity requirements than men to be assigned to demanding
jobs. Because of these di¤erences in the assignment thresholds women tend to start their
careers in less demanding jobs than men and …nd it more di¢ cult to get promoted than
men.
In this paper, we examine gender di¤erences in the job allocations using panel data
on Finnish metalworkers. These data are exceptionally suitable for the analysis of the
assignment thresholds. They provide a measure for the job complexity that can be used to
construct a complexity ladder of jobs. Thus, we are able to compare men and women, who
start their careers in similar jobs. Furthermore, the data contain information on personal
bonuses that can used to study the productivity implications of the gender di¤erences in
the assignment thresholds.
In the Finnish metal industry, women work on clearly less complex jobs than men.
This gender di¤erence in the job allocation is already visible at the initial job assignment.
Nearly 50% of the women would have to change their initial jobs to arrive at the male allocation of initial jobs. Furthermore, women …nd it more di¢ cult to move to more complex
tasks than men: Women are clearly less likely to get promoted than men who start their
careers in the same jobs, even after controlling for observable productive characteristics.
Using personal bonuses as a measure of individual productivity, we also …nd that
there are no clear productivity di¤erences between men and women within the jobs where
the workers are initially assigned to. However, when the workers are split into groups
of promoted and non-promoted workers, the gender di¤erences in productivity become
clear. Promoted and non-promoted women are consistently more productive then their
male counterparts.
Whether one can interpret these results as evidence on gender di¤erences in the assignment thresholds, depends on the underlying productivity distributions of male and
female workers in this industry. We argue that since we do not observe any productivity
di¤erences within the initial jobs, we can interpret the combination of the results that
15
women are less likely to be promoted and that women are more productive among the
promoted and non-promoted workers as evidence on higher female thresholds. Whereas
the result that women are less likely to get promoted than men has been reported often,
it hasn’t been combined with the results on gender di¤erences in productivity that would
match the asymmetric assignment threshold hypothesis before.
Finally, the observed quit behaviour of men and women in this industry is broadly
consistent with the Lazear and Rosen explanation of the gender di¤erences in the promotion thresholds. Namely, the fact that women are more likely to quit the …rm than men,
may imply that women face higher thresholds because their promotion is more costly to
the employer. However, since the quit behaviour may be a result of the di¤erences in
the promotion thresholds, it is not possible to draw de…nite conclusions on the causes of
these threshold di¤erences with these data.
References
[1]Baker, G. P., M. Gibbs & B. Holmström, (1994): ”The internal economics of the …rm:
Evidence from personnel data”, Quarterly Journal of Economics, 109: 921-955.
[2]Blau, F. D, & M. A. Ferber, (1987): "Discrimination: Empirical evidence from the
United States", American Economic Review, 77, 316-20.
[3]Booth, A. L., M. Francesconi & J. Frank, (2003): ”A sticky ‡oors model of promotion,
pay, and gender”, forthcoming European Economic Review, 47(2), 295-322..
[4]Gibbons, R. & M. Waldman, (1999): ”A theory of wage and promotion dynamics
inside …rms”, Quarterly Journal of Economics, 114 (4), 1321-1358.
[5]Ginther, D. K. & K. J. Hayes, (1999): ”Gender di¤erences in salary and promotion in
the humanities”, American Economic Review, Papers and Proceedings, 89, 397-402.
[6]Ginther, D. K. and K. J. Hayes, (2003): "Gender di¤erences in salary and promotion
for faculty in the humanities 1977-1995", Journal of Human Resources, 38(1), 34-73.
[7]Granqvist, L. & H. Persson, (2002): ”Gender di¤erences in career mobility - an example from the wholesale and retail trade”, mimeo, Swedish Institute for Social Research,
Stockholm University.
[8]Hersch, J. & W.K. Viscusi, (1996): ”Gender di¤erences in promotions and wages”,
Industrial Relations, 35, 461-72.
[9]Jenkins, S.P., (1995): "Easy estimation methods for discrete-time duration models",
Oxford Bulletin of Economics and Statistics, 57, 129-138.
[10]Jones, D. R. & G. H. Makepeace, (1996): ”Equal worth, equal opportunities: Pay and
promotion in an internal labour market”, Economic Journal, 106, 401-409.
[11]Lazear, E.P. & E. Rosen, (1990): ”Male-female wage di¤erentials in job ladders”,
Journal of Labor Economics, 13 (3), 426-471.
[12]McCue, K. (1996): ”Promotions and wage growth”, Journal of Labor Economics, 14
(2), 175-209.
[13]McDowell, J.M, L.D. Singell & J.P. Ziliak, (1999): ”Cracks in the glass ceiling: Gender
and promotion in the economics profession”, American Economic Review, Papers and
Proceedings, 89, 392-396.
16
[14]Treble, J. & E. van Gameren & S. Bridges & T. Barmby, (2001): ”The internal
economics of the …rm: Further evidence from personnel data”, Labour Economics, 8,
531-552.
[15]Winter-Ebmer, R. & J. Zweimüller, (1997): ”Unequal assignment and unequal promotion in job ladders”, Journal of Labor Economics, 15:1, 43-71.
17
Table 1 Descriptive statistics
Variable
Men in our sample
Mean
Standard
Women in our
Men in 1990
Women in 1990
sample
cross-section
cross-section
Mean
deviation
Standard
Mean
deviation
Standard
Mean
deviation
Standard
deviation
Age
28.43
8.35
31.43
9.22
37.23
10.40
40.19
10.37
Complexity
33.03
3.24
29.55
2.35
35.21
3.17
30.62
2.62
Bonus
.05
.04
.06
.04
.09
.04
.09
.04
Average
41.35
8.23
34.92
4.61
46.79
7.64
38.48
5.07
hourly
earnings
Individuals
8 956
2 705
(23%)
Ind/year
52377
14 607
(22%)
83474
observations
Our sample consists of all the observations during 1990-2000 on workers who enters the metal industry
during 1990-1995 and stays for at least five years. The descriptive statistics are 1990 means and
standard deviations. 1990 cross-section consists of all the workers in the metal industry population in
1990. Complexity refers to the occupation-related wage in FIM 1990. Bonus is reported as a proportion
of occupation-related wage. Average hourly earnings consist of worker’s earnings in different hourly
wage schemes in FIM 1990.
Table2 Transition matrix between jobs for newcomers in the metal industry, 1990-1991
Exit
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
Total
N
36.5
45.1
18.2
19.5
18.4
14.3
12.8
15.4
8.1
7.7
6.3
4.3
4.9
2.8
3.7
7405
Entry
38.9
44.3
0.1
9.4
2.9
.
1.1
1.3
0.1
1.6
0.3
.
.
0.1
.
.
100
1762
26
39.9
0.1
45.3
3.0
0.1
4.6
3.4
.
1.4
0.3
1.3
0.4
.
0.1
.
.
100
704
27
30.0
1.2
0.4
59.9
2.9
1.0
2.3
0.9
0.5
0.4
0.5
0.2
.
.
.
.
100
3358
28
28.6
0.6
.
2.0
60.1
0.1
5.5
2.2
.
0.7
0.1
.
0.1
.
.
.
100
3475
29
28.6
.
0.7
1.6
.
56.3
5.5
0.2
3.1
0.4
2.1
0.6
0.9
0.1
.
.
100
1344
30
24.9
0.1
.
0.8
1.0
0.4
62.0
4.8
2.2
1.3
1.4
0.7
0.1
0.1
0.1
.
100
6796
31
25.7
0.1
.
0.3
0.5
.
2.3
62.2
0.1
6.0
2.1
.
0.5
.
0.2
.
100
6457
32
22.6
.
0.1
0.3
.
0.5
1.5
.
64.6
0.2
5.5
3.5
.
0.9
.
0.3
100
3303
33
20.9
.
.
0.1
0.2
.
0.7
3.3
.
63.8
9.1
0.3
1.3
.
0.4
.
100
6529
34
16.6
.
.
0.1
.
0.1
0.3
0.7
0.9
2.5
68.9
2.9
5.7
0.5
0.6
0.2
100
10382
35
13.8
.
.
.
.
0.1
0.2
.
0.7
.
4.3
74.0
0.1
5.3
0.1
1.6
100
3754
36
12.5
.
.
.
0.1
.
0.2
0.4
.
1.2
6.5
.
74.3
0.1
4.8
.
100
5677
37
13.1
.
.
.
.
.
0.1
.
0.3
.
0.9
4.0
0.7
76.7
0.1
4.1
100
2199
38
11.8
.
.
.
.
.
.
0.3
.
0.4
1.5
.
4.3
0.1
81.6
0.2
100
3964
39
12.2
.
.
.
.
.
.
.
0.3
.
0.5
2.4
.
2.7
1.7
80.1
100
2600
40
20.6
1.4
0.6
3.8
3.8
1.4
7.9
7.7
4.0
8.1
14.2
5.5
8.3
3.3
5.9
3.6
100
69709
Total
Shows all transitions between complexity levels, including entry, exit, and stays from 1990 to 1991, as percentage of movements from a complexity level to another.
Aggregating occupation-related wages into integers creates complexity levels. Shaded cells indicate the level that was the most frequent destination of the complexity level
moves. Numbers in boxed cells indicate stays within a complexity level. Zeros denote nonempty cells that round up to zero and “.”s denote empty cells.
Table 3 Change in complexity – Gender differences
Total
Male
.010
(.000)
Female
.011
(.000)
Difference
.000
(.000)
.026
.018
-.008**
(.001) (.001)
(.001)
Complexity group 27 .035
.018
-.017**
(.002) (.002)
(.002)
Complexity group 28 .026
.013
-.013**
(.001) (.001)
(.001)
Complexity group 29 .017
.009
-.008**
(.001) (.001)
(.001)
Complexity group 30 .022
.013
-.009**
(.001) (.001)
(.001)
Complexity group 31 .014
.005
-.008**
(.000) (.001)
(.001)
Complexity group 32 .010
.003
-.007**
(.000) (.001)
(.001)
Complexity group 33 .010
.005
-.005**
(.000) (.001)
(.001)
Complexity group 34 .006
.004
-.002
(.000) (.002)
(.002)
Complexity group 35 .005
.001
-.005**
(.000) (.001)
(.001)
Complexity group 36 .005
-.002
-.007**
(.001) (.003)
(.001)
Complexity group 37 .000
.004
.004
(.001) (.011)
(.003)
Complexity group 38 .001
.002
.001
(.001) (.003)
(.001)
The numbers in the second and third columns are the sample mean changes of log occupation-related
wages for men and women. The fourth column gives the female-male difference. The groups were
constructed by aggregating the occupation-related wages of the workers’ initial task assignments to
integers. The numbers in parentheses are standard errors. ** denotes significane at 5%-level and * at 10level.
Complexity group 26
Table 4 Change in complexity – regression results
Variable
(3)
Female
-.008**
(.000)
Age/10
-.011**
(.001)
(Age/10)2
.001**
(.000)
Seniority/10
-.086**
(.003)
.066**
(Seniority/10)2
(.002)
Initial task complexity dummies Yes (15)
R2
.135
N
64 198
Dependent variable is the difference between log of the occupation-related wage in the next period and
the log of the occupation-related wage at the current period. Seniority measured as number of years that
individual has been present in the metal industry. Regression also includes a full set of initial task
complexity group and firm dummies. Numbers in parenthesis are robust standard errors. ** denotes
significane at 5%-level and * at 10-level.
Table 5 Kaplan-Meier estimates of the survivor function by gender
Seniority
1
2
3
4
5
Log-rank test
(1)
The whole sample
Men
Women
0.709
0.746
0.564
0.628
0.473
0.531
0.412
0.468
0.371
0.424
(2)
Low complexity tasks
Men
Women
0.596
0.707
0.410
0.580
0.327
0.478
0.274
0.413
0.237
0.371
(3)
Medium complexity tasks
Men
Women
0.679
0.808
0.536
0.709
0.441
0.624
0.385
0.567
0.345
0.521
(4)
High complexity tasks
Men
Women
0.859
0.943
0.770
0.862
0.696
0.813
0.624
0.766
0.583
0.723
Χ2(1)=21.72
(p=0.000)
Χ2(1)=104.68
(p=0.000)
Χ2(1)=116.85
(p=0.000)
Χ2(1)=2.59
(p=0.108)
Table reports the Kaplan-Meier estimates of the survivor function, where hazard is the probability of
promotion conditional on not being promoted up to the year of seniority. Low-complexity refers to
tasks that have occupation-related wages less than or equal to 30. Medium complexity refers to tasks
that have occupation-related wages larger than 30 but less than or equal to 35. High complexity refers
to tasks that have occupation-related wages higher than 35. The last row reports the log-rank test
statistic for the hypothesis that the male and female survivor functions are equal. ** denotes significane
at 5%-level and * at 10-level.
Table 6 Duration to promotion – Discrete-time complementary log-log proportional hazards estimates.
Variable
Female
Age/10
(Age/10)2
-.357**
(.026)
-.191**
(.087)
-.000
(.012)
Yes (15)
Initial task complexity
dummies
N
64 198
Estimates from a discrete-time proportional-hazards model of promotion. Complementary log-log
specification was chosen for the hazard of promotion. In addition to the variables listed in the table, 11
duration dummies were included in the estimation Standard errors are reported in the parentheses. **
denotes significane at 5%-level and * at 10-level.
Table 7 Median personal bonus deviations in the groups of promoted and not-promoted workers
Stagnant
Stagnant
Difference Promoted
Promoted
Difference
men
women
men
women
-.027
-.027
.000
First
assignment
(.000)
(.001)
(.001)
N=8 956
N=2 705
After 1 year
-.014
-.003
.011**
-.010
-.001
.009**
(.000)
(.001)
(.001)
(.001)
(.001)
(.001)
N=6 345
N=2 017
N=2 611
N=688
-.005
.001
.007**
After 2 years
-.010
.000
.011**
(.001)
(.001)
(.001)
(.001)
(.001)
(.001)
N=5 057
N=1 675
N=3 899
N=1 030
-.002
.005
.007**
After 3 years
-.007
.001
.008**
(.001)
(.001)
(.001)
(.000)
(.001)
(.001)
N=4 229
N=1 431
N=4 727
N=1 274
.000
.007
.007**
After 4 years
-.005
.003
.008**
(.001)
(.001)
(.001)
(.001)
(.001)
(.001)
N=3 687
N=1 255
N=5 269
N=1 450
Numbers in cells are medians of bonuses as deviations from firm- and task-specific means. Stagnant
refers to the group of workers who remain in their first assignment job. Promoted refers to the groups
of workers who move to more complex jobs. Medians of bonus deviations are reported for each group.
The numbers in parentheses are standard errors. N refers to the sample size of the cell. ** denotes
significane at 5%-level and * at 10-level.
Table 8 Probit model of voluntary quits, marginal effects
Variable
Female dummy
Age/10
Age squared /100
(Female x age)/10
(Female x age squared)/100
Seniority/10
Seniority squared /100
Firm size/1000
14 complexity dummies
Marginal effect
.050**
-.216**
.029**
-.008*
-.001
-.044**
.005**
.000**
Yes
Standard error
(.011)
(.002)
(.000)
(.004)
(.001)
(.001)
(.000)
(.000)
N
630 839
% correctly predicted
91.8
Log-likelihood
-165709.14
.072
Pseudo R2
Marginal effects of the probit model of quit behaviour using the whole population of metalworkers
1990-2000. Marginal effects of the female dummy reports the effect of 0 to 1 change. Quit is defined as
a separation where at least 75% of the employees remain in the firm. ** denotes significane at 5%-level
and * at 10-level.
Table 9 Predicted probabilities of quitting in different age groups
Age group
Male Female
25 and below .173 .242
25-30
.086 .124
30-35
.054 .079
35-40
.041 .059
40-45
.041 .054
45-50
.052 .059
50-55
.081 .079
55 and above .166 .141
Predicted probabilities of quitting by age groups. The probabilities are calculated using the coefficients
of the probit model in table 8. Quit is defined as a separation where at least 75% of the employees
remain in the firm. ** denotes significane at 5%-level and * at 10-level.
View publication stats