DP 2222

IAB-DISCUSSION PAPER
Articles on labour market issues
22|2022 Gender-Specific Application Behavior,

Matching, and the Residual Gender Earnings Gap
Benjamin Lochner, Christian Merkl
ISSN 2195-2663
Gender-Specific Application Behavior,
Matching, and the Residual Gender Earnings
Gap
Benjamin Lochner (Institute for Employment Research (IAB) and

Friedrich-Alexander-University Erlangen-Nürnberg)
Christian Merkl (Friedrich-Alexander-University Erlangen-Nürnberg and Institute of Labor
Economics (IZA))
Mit der Reihe „IAB-Discussion Paper“ will das Forschungsinstitut der Bundesagentur für
Arbeit den Dialog mit der externen Wissenschaft intensivieren. Durch die rasche Verbreitung
von Forschungsergebnissen über das Internet soll noch vor Drucklegung Kritik angeregt
und Qualität gesichert werden.
The “IAB-Discussion Paper” is published by the research institute of the German Federal
Employment Agency in order to intensify the dialogue with the scientific community. The
prompt publication of the latest research results via the internet intends to stimulate
criticism and to ensure research quality at an early stage before printing.
Contents
1 Introduction 7
2 Theory 13
2.1 Model Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.1 Application Decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.2 Hiring Decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.3 Wage Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.4 Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.5 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Model Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.1 Taste-Based Discrimination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.2 (Non)Linear Production Functions and Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.3 Model and Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3 Data 24
3.1 Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Administrative Data Linkages and Imputations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Final Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4 Empirical Results 29
4.1 Application and Selection Patterns at the Firm Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Application Behavior and Firm-Sided Flexibility Requirements . . . . . . . . . . . . . . . . . . . . 32
4.3 Residual Gender Earnings Gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4 Evidence for Nonlinear Jobs on the Person Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5 Conclusion 42
References 44
A.1 Descriptive Details 48
A.2 Additional Empirical Results 50

A.2.1 Age Cohorts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
IAB-Discussion Paper 22|2022 3

A.2.2 Alternative Selection Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
A.2.3 Application and Selection Behavior within Task Complexities. . . . . . . . . . . . . . . . . . . . . . 51
A.2.4 Application and Selection Behavior with Alternative Firm Fixed Effects . . . . . . . . . . . 54
A.2.5 Application and Selection Behavior and Bargaining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
A.3 Alternative Sample Restriction 58

Abstract
This paper opens up the black box of gender-specific application and hiring behavior and its
implications for the residual gender earnings gap. To understand the underlying
mechanisms, we propose a two-stage matching model with testable implications. Using the
German IAB Job Vacancy Survey, we show that the patterns in the data are in line with linear
and nonlinear production functions at different jobs. Women’s application probability at
high-wage firms is much lower than at low-wage firms. By contrast, women have the same
probability of being hired as men when they apply at high-wage firms. These patterns are
not in line with taste-based discrimination, but they can be rationalized by high-wage firms
that ask for more employer-sided flexibility. We show that the share of male applicants
increases in various dimensions of employer-sided flexibility requirements. Adding the
share of male applicants as a proxy for flexibility requirements to Mincer wage regressions
reduces the residual earnings gap by around 50 to 60 percent. Women who match at jobs
with a high share of male applicants earn substantially more than women at comparable
jobs with only females in the application pool (due to compensating differentials). By
contrast, when women with children match at these jobs, they face large earnings discounts
relative to men.
Zusammenfassung
In diesem Papier untersuchen wir das geschlechterspezifische Bewerbungs- und
Einstellungsverhalten sowie deren Implikationen für die Lohnlücke. Um die
zugrundeliegenden Mechanismen besser zu verstehen, leiten wir aus einem zweistufigen
Matching-Modell Implikationen her, die wir auf Basis der IAB Stellenerhebung überprüfen.
Es zeigt sich, dass die Muster in den Daten konsistent mit der Theorie von linearen und
nichtlinearen Produktionsfunktionen sind. Wir dokumentieren, dass sich Frauen seltener
bei Hochlohn- als bei Niedriglohnfirmen bewerben. Dahingegen finden wir keinen
statistisch signifikanten Unterschied in der Einstellungswahrscheinlichkeit zwischen
Männern und Frauen, wenn sie sich bei Hochlohnfirmen bewerben. Diese Muster sprechen
gegen die Diskriminierungshypothese aus dem theoretischen Modell, lassen sich aber
dadurch erklären, dass Jobs bei Hochlohnfirmen höhere arbeitgeberseitige

Flexibilitätsanforderungen haben. Wir zeigen, dass der Anteil der männlichen Bewerber in
verschiedenen, beobachtbaren Flexibilitätsanforderungen ansteigt. Wenn wir den Anteil der
männlichen Bewerber in Mincer-Lohnregressionen als zusätzliche erklärende Variable für
Flexibilitätsanforderungen aufnehmen, sinkt der Geschlechterunterschied im
Einstellungslohn zwischen Männern und Frauen um etwa 50-60 Prozent. Frauen, die in Jobs
eingestellt werden, bei denen sich viele Männer bewerben, verdienen signifikant mehr als
Frauen, die in vergleichbaren Jobs eingestellt werden, bei denen sich ausschließlich Frauen
bewerben. Wenn Frauen mit Kindern Jobs mit hohen Flexibilitätsanforderungen annehmen,
müssen sie dahingegen große Lohneinbußen im Vergleich zu Männern hinnehmen.
JEL
E24, J16, J31
Keywords
Job Search, Application Behavior, Gender Earnings Gap
Acknowledements
We thank Jesper Bagger, Steven Davis, Jason Faberman, Simon Jäger, Lisa Kahn, Philipp
Kircher, Andreas Müller, Johannes Schmieder, Bastian Schulz, Rune Vejlin, and seminar
participants from the Chair of Macroeconomics, the IAB, the NBER Summer Institute 2022,
the Gender Gaps 2021 conference, the NUEdialog, the 12th ifo Conference on
Macroeconomics and Survey Data, the DTMC Workshop on Interactions between Labor and
Marriage Markets, the Warsaw International Economic Meeting and the APB/IOS Conference
on Gender Inequality in Central and Eastern Europe for valuable feedback. We thank Linas
Tarasonis, Ipek Yükselen, and Olga Zajkowska for discussing our paper.

1 Introduction
After several decades of gender convergence, substantial differences in earnings between

men and women remain. Part of this gap can be explained by men and women working in
different occupations and industry sectors (Blau/Kahn, 2017), or in firms with different wage
premiums (Card/Cardoso/Kline, 2016). However, even within narrowly defined sectors and
occupations, a substantial gender earning gap remains. A recent strand of the literature
analyzes the role of gender-specific search behavior of workers for the gender earnings
gaps, combining search theory and newly available microeconomic datasets (Cortés et al.,
2021; Faberman et al., 2017; Fluchtmann et al., 2020).
Our paper contributes to this stream of the literature, by using data from the German IAB
Job Vacancy Survey that we link to administrative employment records. This unique
combination allows us to observe important dimensions of the search and matching
process such as characteristics of the hiring firm (e.g., wage premium), the hired worker
(e.g., whether a woman is a mother), and the recruitment process itself (e.g., the gender
distribution in the application pool).1 Guided by our two-stage search and matching model,
we show that men and women tend to apply at different firms2 and for different jobs. These
differences can explain a large part of the residual gender earnings gap. Specifically, we
show that women in Germany have a lower probability to apply for jobs at firms with high
wage premiums from a two-way fixed effect regression (Abowd/Kramarz/Margolis, 1999;
Card/Heining/Kline, 2013). At the same time, the probability of being hired at these
high-wage firms conditional on application is similar for males and females. We argue
through the lens of our theoretical model that these patterns are not reconcilable with
taste-based discrimination at the hiring stage. By contrast, they can be explained by
different job characteristics (Goldin, 2014), namely more employer-sided flexibility
requirements at high-wage firms. We show that the share of male applicants3 increases with
1
To the best of our knowledge, we are the first to use data that has information on the pool of gender-specific
applicants for a particular job in a particular firm.
2
Although we refer to firms, the IAB data rather identifies plants/establishments, i.e., individual production
units. We use these terms interchangeably throughout the paper.
3
We residualize the share of male applicants by controlling for occupation, sector, and firm size.

various employer-sided flexibility requirements (such as working irregular hours or at
various locations). Adding the share of male applicants as a proxy for multidimensional
flexibility requirements to standard Mincer earnings regressions leads to a drop in the
residual gender-earnings gap of around 50 to 60 percent. Women who match at jobs with a
high share of male applicants earn substantially more than women at comparable jobs with
only females in the application pool (netting out worker, firm, and job characteristics).
These patterns are in line with Goldin (2014)’s idea of nonlinear jobs that pay a
disproportional premium for providing flexibility. We show that the discount in earnings is
particularly strong for mothers with children in jobs with high employer-sided flexibility
requirements. In line with our model, if mothers match at these nonlinear jobs, they are
more likely unable to provide the desired flexibility requirements and thereby have lower
earnings.
We motivate and structure our empirical exercise with a simple two-stage search and
matching model. In the first stage, searching workers have to decide whether they want to
apply for a particular job profile. Facing heterogeneous application costs, they will apply
whenever the expected returns from the application are larger than application costs. In the
second stage, only those worker-firm pairs with a positive surplus will form a match.
Worker-firm pairs draw an idiosyncratic match-specific training cost shock. Only a certain
fraction of workers will be selected in the model (see Chugh/Merkl (2016), or
Carrillo-Tudela/Gartner/Kaas (2020) for selection models). In our model, the male and
female application behavior is a function of the expected match surplus. Thus, a high share
of male applicants shows that men (on average) perceive a higher surplus for certain job
types. We analyze two scenarios. In the first scenario, we assume taste-based discrimination
at the hiring stage. Employers only recruit women if they are compensated by higher profits
for their distaste. This scenario leads to lower female application rates at discriminating
firms and lower selection rates at discriminating employers. In the second scenario, we
assume nonlinear and linear jobs as proposed by Goldin (2014). In nonlinear jobs, higher
input (e.g., in terms of providing more working hours or more employer-sided flexibility)
leads to a more than proportional increase in output. We assume that the desired input
level among men and women is heterogeneous. If there is a larger fraction of women who is

willing/able to provide lower input, this will generate a sorting equilibrium with more
women applying for linear jobs and more men applying for nonlinear jobs. Under strong
sorting (i.e., workers who are unable to provide a large input apply predominantly for linear
jobs), firms with nonlinear production functions would predominantly receive applications
from workers that are willing and able to provide a high input. Thus, those men and women
who apply at these nonlinear firms would have similar selection rates and wages.
In the first step of the empirical analysis, we sort different hiring firms along AKM firm wage
effect deciles, which we obtained from a two-way fixed effects regressions
(Abowd/Kramarz/Margolis, 1999; Card/Heining/Kline, 2013). We find that the probability for
women to apply for a job decreases almost monotonically in the firm wage premium. After
taking into account differences in sectors, occupations, and firm size, women have a 10
percentage points higher probability to apply in the lowest AKM firm decile and a 6
percentage points lower probability to apply in the highest AKM decile.4 Interestingly, we
find no evidence for taste-based discrimination, as we observe indistinguishable male vs.
female selection rates in the second stage of the application process (after controlling for
sectors, occupations, and firm size).
In the second step, we show that the (residualized) share of male applicants increases in
various indicators for employer-sided flexibility (e.g., longer working hours, changes in
working hours, mobility). Therefore, we argue that the share of male applicants is a suitable
encompassing proxy for employer-sided flexibility requirements, which we use in
subsequent regressions.
In the third step of the empirical analysis, we estimate standard Mincer earnings regressions
controlling for detailed worker, firm, and job characteristics. Next, we add the share of male
applicants as proxy for employer-sided flexibility requirements. We find that this proxy has
significant explanatory power beyond standard observables. The residual gender earnings
gap drops significantly in all specifications. It falls from around 14-15 percentage points to
4
Moreover, this pattern holds qualitatively within well-defined occupational task complexities. See Appendix
A.2.3.

around 6-7 percentage points, i.e., by around 50-60 percent. Importantly, the share of male
applicants is also relevant for the level of earnings when we look at female matches only.
Women who match in a pool with a large share of male applicants have up to 9 percentage
points higher earnings compared to jobs with a medium share of male applicants
(controlling for a large set of worker and job observables). Women who match in jobs with
no male applicants at all have up to 10 percentages points lower earnings, again compared
to comparable jobs with a medium share of male applicants.
Finally, we analyze characteristics of the matched workers. We show that workers who
match in a pool with a larger share of male applicants have on average larger AKM worker
fixed effects. This is in line with what we expect based on the model. If a certain group of
workers matches at firms with nonlinear production functions (proxied by a larger share of
male applicants in the data), they will produce more on average and part of this larger
production will be handed on to workers as higher wages. Thus, these patterns in the data
provide further support for the sorting hypothesis from the theoretical model.
We show that the residual gender earning gaps is significantly larger for mothers than for
women without children and that there is a strong interaction with flexibility requirements.
If mothers match at high-flexibility jobs, they face much larger discounts both relative to
men and relative to women without children. Again, this is in line with our hypothesis of
nonlinear production functions. Women with children tend to be less flexible. If they match
at nonlinear jobs, they produce significantly less and thereby face particularly large wage
discounts.
Our findings are complementary to a recent strand of the literature that analyzes
gender-wage gaps for specific industries or firms (Azmat/Ferrer, 2017; Bolotnyy/Emanuel,
2019; Cook et al., 2021). These authors find that once they control for the detailed working
behavior (e.g. working longer hours or working night shifts), the gender wage gap shrinks
considerably. While these studies have very detailed information on gender-specific
behavior of workers within certain industries, we have a dataset that represents the entire
economy, but at the same time contains information on application behavior and flexibility

requirements that are both typically absent in standard datasets.
Our work is most closely related to another recent strand of literature that analyzes gender
issues combining insights from search and matching theory with rich microeconomic data.
Faberman et al. (2017) document men and women’s job search behavior and the
implications for the gender wage gap using US survey data for workers. Cortés et al. (2021)
show a substantial difference between men and women in terms of the timing of their job
acceptances based on a sample of (former) undergraduate students. Xiao (2021) analyzes
the gender wage gap from a life cycle perspective and finds that both statistical
discrimination based on fertility concerns and different labor force attachments play an
important role in explaining the gender wage gap in Finland. While these studies are similar
in spirit to our paper, the unique combination between the tractable model and the IAB Job
Vacancy Survey with its linkages to administrative data allows us to shed light on the
intertwining of the gender-specific application of workers and the selection behavior of
firms. Specifically, the data allows us to explore the role of job characteristics such as
employer-sided flexibility requirements while simultaneously controlling for important
worker and firm characteristics. Due to the cross-sectional nature of our data, we have less
to say on the life cycle component. However, in Appendix A.2.1, we show that the residual
gender earnings gap is particularly large for women who match in their 30s and 40s (when
childcare considerations may matter most). In addition, we directly show that women with
children face the largest earnings discount in male-dominated jobs. This observation is in
line with Illing/Schmieder/Trenkle (2021) who show that having children sharply increases
the gender gap in earnings losses after displacement. Fluchtmann et al. (2020) are probably
closest to our paper. They use Danish unemployment insurance (UI) recipients data to
empirically show that the differences in the application behavior between males and
females can explain large parts of the traditional gender wage gap. The data are very
similar, however, ours are not limited to UI recipients as they contain all hires. In addition,
we have specific information about the exact gender distribution of the pool of applicants
for each specific recruitment process, which allows us, as we show below, to calculate
important measures derived from our model which help explaining the gender wage gap.

Our paper also contributes to the recent literature on compensating differentials. Sorkin
(2018) shows for the United States that compensating differentials can explain around
two-thirds of the variance of firm-level earnings. Taber/Vejlin (2020) show for Denmark that
preferences for non-pecuniary aspects are very important for job choices. Our empirical
findings are in line with these findings. Women have a higher probability to apply for
low-wage jobs and to get compensated in terms of low employer-sided flexibility
requirements. Consistently, Budig/Hodges (2010) show that mothers are more willing than
women without children to trade wages for family friendly employment.
Based on experimental data, Wiswall/Zafar (2017) show that women have a higher
willingness to pay for non-wage job characteristics. In the same vein,
Le Barbanchon/Rathelot/Roulet (2020) analyze gender differences in willingness to
commute. They show for France that women commute much shorter than men. Based on
their search model, they find that 14 percent of the residualized gender wage gap can be
explained by this mechanism. While the IAB Job Vacancy Survey does not contain any
commuting times, we believe that this mechanism is included in our regressions when we
use the share of male applicants in our Mincer type regressions. Matches that require longer
commuting times can be expected to be disliked by women (in particular those with care
responsibilities). In our view, this is another dimension of employer-sided flexibility
requirements that is not directly measurable in our data. Thus, our proxy for
employer-sided flexibility is more encompassing than pure commuting times. Against this
background, it makes sense that the residual gender-earnings gap is reduced by a lot more
in our regressions (by 50 to 60 percent) than in Le Barbanchon/Rathelot/Roulet (2020).
Our paper is also highly relevant from an economic policy perspective. In particular, the
Covid-19 episode with work-from-home arrangement provided a test-laboratory whether
more flexibility from the employee side is possible. Barrero/Bloom/Davis (2020) argue that
these working from home arrangements have boosted productivity. To the extent that these
arrangements have changed the production process and that they will stick permanently,
the results from our paper imply that this leads to a decline in the residual earnings gap, as
this would make certain jobs more accessible and attractive for women.

The rest of the paper proceeds as follows. Section 2 describes the model framework and
derives theoretical implications for taste-based discrimination at the hiring stage and for
different production functions. Section 3 provides details on the used datasets. Section 4
contains the empirical analysis on gender-specific application behavior, the estimated
gender earnings gap, differences between male and female dominated jobs, and how
flexibility requirements and being a woman with children interact. Section 5 briefly
concludes.
2 Theory
We derive a theoretical model that allows us to interpret the patterns in the IAB Job Vacancy
Survey through a gender-specific labor market flow perspective.5 In the data, we observe
the application behavior of males and females for particular jobs (both in terms of pay and
flexibility requirements) and the hiring behavior of firms for particular jobs. Accordingly, our
model assumes a two-stage decision problem (i.e., application and hiring/selection). In the
first stage, workers have to decide whether they apply for a particular job or not. In the
second stage, only those worker-firm pairs with a positive match surplus will form a match,
i.e. only a certain fraction of workers will be selected by firms.6 We analyze the implications
of two specific scenarios and compare them to patterns in the data. First, some firms may
do taste-based discrimination at the hiring stage, i.e. they may dislike hiring women.
Second, following Goldin (2014), we assume that there are jobs with nonlinear and others
with linear production functions. At nonlinear jobs, the output increases more than
proportionally with input. Working hours are certainly one important dimension of input.
However, we define input in a multidimensional sense (e.g., including the ability to do
business travels or to be available on short notice).
5
In line with the cross-sectional dataset, the model is completely silent on some potentially important
dimensions (e.g. the intertemporal life cycle perspective).
6
For other selection models, see Brown/Merkl/Snower (2016), Chugh/Merkl (2016), or
Carrillo-Tudela/Gartner/Kaas (2020)

2.1 Model Environment
We assume that there are different job profiles, where yp,j denotes the output level when
worker j matches at a certain job profile p. For simplicity but without loss of generality, we
derive a static model and we exclude the possibility of multiple vacant jobs for one worker,
i.e. one random job is visible for each searching worker. We assume that workers learn
about one particular job profile. In the first stage, they have to decide whether to apply for
this particular job or not. They will do so if application costs e are smaller than the expected
return from this application. The ex-ante application costs e are drawn from a stable density
function, g (e). The application costs are sunk at the time of application, i.e. they will not
play any role for the surplus in the second stage.
In the second stage, workers j that decided to apply for a particular job profile p, draw a
match-specific training cost shock upon contact with a firm. We denote this shock by εp,j .
The ex-post shock is drawn from a stable density function, f (ε). Only those worker-firm
pairs with a positive joint surplus will create a match.
2.1.1 Application Decision
Worker j will apply for a particular job p whenever the expected returns from a match are
larger than the application costs:
Eηp,j w̄ (ε̃p,j ) − ξj > ep,j . (1)
The left-hand side of the equation shows the expected returns from a match, where ηp,j is
the hiring rate in the second stage, w̄ (ε̃p,j ) is the expected wage conditional on being hired
that will be defined below (which is a function of the cutoff point in the second stage, ε̃p,j ),
and E is the expectations operator.7 ξ is the worker’s value of unemployment (e.g., home
7
In the first stage, workers do not know their shock realization in the second stage yet. However, they know
the output level of the job, yp,j , and the properties of the training costs distribution. Therefore, under

production and benefits). e are application costs that are drawn from a stable density
function.
Thus, there is a certain cutoff point level, ẽp,j , up to which workers will apply for job type p:
ẽp,j = Eηp,j w̄ (ε̃p,j ) − ξj . (2)
Above ẽp,j , application costs are larger than returns. Below this threshold, workers will
apply for job p. The application rate of group j for a particular job p is the integral from the
lower support of the distribution (emin
p ) up to the cutoff point:
∫ ẽp,f
αp,j = g (e) de. (3)
emin
p
2.1.2 Hiring Decision
Upon contact, each worker-firm pair draws an idiosyncratic match-specific cost shock, εp,j ,
which we interpret as training costs. Some workers require little training, others require a
lot of training to do the same job. Once a match is formed, each job profile produces a
certain output level yp,j , which may be dependent on the willingness of the worker to
provide input (to be discussed and specified below). In addition, there may be taste-based
discrimination of employers at the hiring stage against certain worker groups. This means
that the firm will only hire from this group if there is a compensation in the amount of tp,j
for the distaste. The joint match surplus between workers and firms is defined as:
Πp,j = yp,j − εp,j − tp,j − ξj > 0. (4)
rational expectations, they know the average expected hiring probability and the average expected wage
conditional on being hired.

The next two equations define the worker and firm surplus separately. Both surpluses have
to be positive for a match to take place:
w (Πp,j ) − ξj ≥ 0, (5)
yp,j − w (Πp,j ) − εp,j − tp,j ≥ 0. (6)
Equation (6) defines the condition under which the employer is willing to hire a worker and
to produce. Under a bilaterally efficient wage formation process, there will be production
whenever there is a non-negative joint surplus Πp,j ≥ 0. At the cutoff point for training
costs, the joint surplus equals zero. Thus, imposing bilateral efficiency, we can calculate the
cutoff point for idiosyncratic match-specific costs up to which workers and firms are willing
to produce:8
ε̃p,j = yp,j − tp,j − ξj . (7)
The selection rate of a worker from group j at job p is the integral from the lower support of
the idiosyncratic cost function (εmin
p ) up to the cutoff point:
∫ ε̃p,j
ηp,j = f (ε) dε. (8)
εmin
p
2.1.3 Wage Formation
In order to be able to define the wage and the application rate, we need to take a stance on
wage formation. Without loss of generality, we assume Nash bargaining between workers
and firms. This leads to the plausible outcome that wages are a function of firm-specific
output, the idiosyncratic training costs realization and workers’ fallback options.
8
Note that the wage does not show up in equation (7) because of the imposed bilateral efficiency.

Under Nash bargaining, workers and firms maximize their joint Nash product, Λ, with
respect to the wage:
Λ = (w (yp,j , εp,j , ξj ) − ξj )α (yp,j − w (yp,j , εp,j , ξj ) − εp,j − tp,j )1−α , (9)
where α is workers’ bargaining power.
This yields the following wage:
w (yp,j , εp,j , ξj ) = α (yp,j − εp,j − tp,j ) + (1 − α) ξj . (10)
Equations (5) and (6) establish conditions under which wage formation is bilaterally
efficient. They hold under Nash bargaining.
Based on the wage formation mechanism, we calculate the expected wage conditional on
being hired for a particular job that we require for the first stage of the decision process:
∫ ε̃p,j
εmin
w (ε) f (ε) dε
p,j
w̄ (ε̃p,j ) = . (11)
ηp,j
2.1.4 Production
We allow for two scenarios in terms of production. Either there is a fixed production level for
each job profile, yp , or there may be two types of production functions. The second case will
be derived below.
Following Goldin (2014), we assume that there may be firms with different production
functions and that workers can choose the amount of input provided, λj .9 Input may be
working hours, but it may also be other employer-sided flexibility requirements such as
working in different locations or such as being available on short notice.
9
As we focus on workers’ application behavior in a partial setting, we abstract from the question under which

Figure 1: Nonlinear and Linear Jobs
Note: The figure illustrates the output as a function of input for a linear and a nonlinear production function. It illustrates the input-output
connection for a worker who is willing to provide a high input (type 1) and for a worker who is willing to provide a lower input (type 2).
Source: Own representation based on Goldin (2014).
For jobs with a nonlinear production function, nl, output is defined as:
ynl,j = λj anl if λj > λ∗ (12)
ynl,j = λj anl (1 − δ) if λj ≤ λ∗ (13)
In addition, there are other jobs where the output is linear, l:
yl.j = λj al (14)
As in Goldin (2014), we assume that λj anl > λj al for λj > λ∗ and λj anl < λj al for λj < λ∗ .
Figure (1) illustrates the nature of the two production functions. If a worker is willing to
provide working hours/flexibility beyond the minimum threshold λ∗ , this leads to more
output at nonlinear firms than at linear firms. If not, there is more production at linear
firms.
circumstances these nonlinear and linear firms coexist in a full general equilibrium setting.

The underlying idea is that certain job profiles require a large degree of flexibility in order to
deliver high output levels (nonlinear jobs). A surgeon in a hospital may for example have to
be available on short request, while he/she may have more reliable working times in a
doctor’s office. A sales manager at an internationally operating firm may have to travel long
distances, while this may not be the case for a sales manager at a locally operating firm. As
we will be controlling for occupation, sector, and firm size in our empirical specification, we
have in mind different jobs in similar occupations or sectors.
2.1.5 Equilibrium
The labor market equilibrium is described by the application cutoff point in equation (2),
the application rate (3), the cutoff point for the idiosyncratic match-specific cost shock (7),
the corresponding selection rate (8), and the wage expectations conditional on being hired
(11). Output per job is either exogenous. Alternatively, production may be governed by
different types of (non)linear production functions and the willingness of applicants to
provide certain input levels.
2.2 Model Implications
Our model allows us to analyze how different scenarios affect the application rate, the
selection rate and the wage for different worker groups j . Therefore, we now look at two
scenarios. First, we analyze what happens if there is taste-based discrimination against
women in high-productivity jobs. The empirical observation that women earn
systematically less than men (controlling for observables) may be driven by taste-based
discrimination at firms that produce a large output level per worker. Second, we analyze the
implications of our model with nonlinear and linear jobs.10
10
As a third potential mechanism, we could analyze different bargaining powers of men and women. However,
we do not have any direct proxy for the level of bargaining power in our dataset. In addition, we show in
Appendix A.2.5 that our empirical results are very similar at firms with and without an institutionalized
bargaining agreement (e.g., collective bargaining).

2.2.1 Taste-Based Discrimination
Let’s start by assuming that workers are ex-ante homogeneous and production per job is
exogenous, yp . Applicants only differ in terms of their gender. For the sake of the argument,
assume further that employers at certain firms/jobs discriminate against women at the
hiring stage (tp,f > 0, tp,m = 0, where f stands for female and m for male).
Taste-based discrimination of females would reduce the joint surplus in case of a female
match and thereby reduce the cutoff point for the idiosyncratic shock realization:
ε̃p,f = yp,f − tp,f − ξf . (15)
This leads to a lower selection rate in the second stage of the application process.
As women anticipate the selection behavior and the wage in the second stage, only a
smaller fraction of them will send an application to these firms in the first place, i.e. the
cutoff for application is lower. This can be seen best, by substituting the wage conditional
on hiring (equation (11)) into the application cutoff point condition (equation (2)):
∫ ε̃p,j
ẽp,f = E w (ε) f (ε) dε − ξf . (16)
εmin
p,j
Overall, taste-based discrimination at the hiring stage leads to lower female application
rates and lower female selection rates. These implications can be tested in the data.
2.2.2 (Non)Linear Production Functions and Sorting
Next, we analyze the implications of two types of production function (linear and
nonlinear). Let’s assume for illustration purposes that there are two types of workers (see
also Figure (1)). Type 1 workers are willing/able to provide a larger input, λj , than type 2

workers. In addition, we assume that type 1 workers are above the threshold, λ1 > λ∗ ,
while type 2 workers are below, λ2 < λ∗ .
Under these assumptions, we obtain four different cutoff points:
ε̃nl,1 = λ1 anl − ξ1 , (17)
ε̃l,1 = λ1 al − ξ1 , (18)
ε̃nl,2 = λ2 (1 − δ) anl − ξ2 , (19)
ε̃l,2 = λ2 al − ξ2 . (20)
And under our assumptions, the following ranking holds:
ε̃nl,1 > ε̃l,1 , (21)
and
ε̃l,2 > ε̃nl,2 . (22)
Thus:
ηnl,1 > ηl,1 , (23)
ηl,2 > ηnl,2 . (24)
Intuitively, type 1 workers generate the largest output at nonlinear production firms and
thereby face the largest selection rate at these firms. By contrast, type 2 workers generate
the largest output at firms with linear production functions. The same ranking is true for
wages and thereby the probability to apply at the respective firms.

Under certain parametrizations (large differences in production between linear and
nonlinear jobs and small dispersion of idiosyncratic application costs), our model generates
a complete sorting equilibrium of the following sort:
ηnl,1 > ηl,1 = 0 (25)
ηl,2 > ηnl,2 = 0 (26)
In this case, type 1 workers would have no surplus at linear jobs and type 2 workers would
have no surplus at nonlinear jobs. As a consequence, type 1 workers would not apply at
linear jobs and type 2 workers would not apply at nonlinear jobs. Although this example
appears to be extreme, it is very useful for illustration purposes.
How could different production functions and input provisions interact with gender? Even
nowadays women bear a larger responsibility in terms of childcare and other family-related
responsibilities. Thereby, a larger fraction of women may be less flexible in terms of input
provision than men (i.e. they may have more trouble working long hours, being available on
short notice, or doing business travel). Assume that a larger share of men is type 1 workers
(compared to women). In this case, we would observe that the average application rate of
females at high-wage firms (those with nonlinear production function) is lower. Note that
under complete sorting those women who match at nonlinear firms (only type 1 females)
would have the same selection rate and the same wage as males.
We are unable to observe type 1 and type 2 persons in the data directly. However, one of the
key data innovations is that we have proxies for the required flexibility at specific job
vacancies (e.g., hours worked or other flexibility requirements) and proxies for the flexibility
that can be provided on the worker side (e.g., whether women are mothers or not).

2.2.3 Model and Data
Although our theoretical model is too simple to be used for structural model estimations, it
provides useful guidance at which outcome variables we should look at. The model
provides a roadmap for the empirical analysis.
As we have AKM firm-fixed effects for each firm and we observe the exact number of
applicants for each job, we can calculate the share of female applicants and the probability
of being selected (upon application) for jobs with different wage premiums. In a first step,
we will test our hypothesis of taste-based discrimination at the hiring stage by checking
whether hiring probabilities for women (upon application) are generally lower than for men
(controlling for observables). In addition, we will check whether such a pattern is prevalent
in high-wage premium firms. If high-wage firms discriminate more than low-wage firms, this
would lead to a gender earnings gap, as women would apply at these firms with lower
probability and as they would be selected with lower probability at these firms. Overall, this
would depress the share of women in firms with the highest earnings.
In a second step, we will analyze the connection between female application behavior and
employer-sided flexibility requirements at the job level. This will help us to understand
whether these flexibility requirements (potentially driven by nonlinear production
functions) may be an important driver for gender differences. In addition, it will help us to
understand whether the share of male applicants may be a suitable proxy for these
flexibility requirements.
In a third step, we will analyze whether the share of male applicants matters for the realized
earnings. We will analyze whether females who match in a pool with a larger share of male
applicants earn more than females who match in a pool with a large share of female
applicants (controlling for observables).
Finally, we move to the person level and analyze how the share of male applicants is

correlated with worker-fixed effects. Under sorting, we expect them to be larger with a larger
share of male applicants, as workers at more demanding workplaces generate more output
and thereby larger earnings. In addition, we will directly check whether having children
affects certain outcomes for women. This provides a direct test for the question whether
nonlinear production functions and inflexibility for women with children interact.
3 Data
3.1 Data Sources
We use the IAB-Stellenerhebung (IAB Job Vacancy Survey, JVS, see Moczall et al., 2015) as
our primary source of data. The JVS covers up to 14,000 establishments per year and is a
representative survey among establishments in Germany from all sectors and from all
establishment size classes. Each year, the survey collects information on the hiring process
of German establishments.11
An important component of the JVS is an array of questions about the recruitment process
of their most recent new hire.12 These questions gather information on job characteristics
such as the exact job requirements, search channels, search duration, the exact hiring date,
individual hire attributes such as gender, age, as well as match-specific characteristics like
educational qualification, wage bargaining, and, in some waves, the hourly wage. Crucial
for our purposes, the JVS asks about details on the pool of applications for the most recent
hire. Specifically, employers report the number of applicants, the (self-assessed) number of
suitable applicants, the number of invited individuals, and their gender composition.
We complement the JVS data with information from the German social security system.
Specifically, we use the method developed by Lochner (2019) to identify establishments’
11
We use the information from the ’main’ survey, which is conducted in each fourth quarter. For a subset of
establishments, there are follow-up questionnaires in the three next quarters.
12
Specifically, establishments are asked to report their most recent hire (regular part- or full-time worker, no
marginal employed or apprentices) within the last 12 months.

last hires in the administrative records, the Integrated Employment Biographies (IEB). The
identification is based on overlapping information such as the hiring date, workers’ age,
gender, and occupational codes. Using a deterministic matching algorithm, around 70
percent of the last hires from the JVS can be found in the administrative records. Table 2 in
Lochner (2019) shows that identified JVS-hires are similar to new hires in the admin data in
terms of observable worker characteristics.13 The IEB encompasses labor market
information for the majority of workers in Germany.14 Combining the survey data with the
administrative records, hence allows us to observe workers’ entire employment and
earnings history.
In our baseline specifications in the main part, we restrict the sample to full-time jobs,
which we define as a job with more than 25 contractual hours. In Appendix A.3, we show
that all our results are robust when giving up this restriction and taking also part-time jobs
into account.
3.2 Administrative Data Linkages and Imputations
The social security data reports the total wage sum over workers’ employment spell. These
sums are right-censored at the contribution assessment ceiling
(“Beitragsbemessungsgrenze”), given by the statutory pension fund. We follow
Dustmann/Ludsteck/Schönberg (2009) and fit a series of Tobit regression to impute the
censored part of the wage distribution.15
For workers’ educational attainment, we construct a variable from information on both

schooling and education in terms of the German vocational system. First, we correct for
13
The algorithm performs several plausibility checks with respect to deviations in the overlapping
information. Note that hires with missing information in the key variables are not taken into account.
14
The IEB covers around 80 percent of the German working population, only excluding civil servants and the
self-employed.
15
First, wages are deflated. Then, Tobit regressions are performed separately for East and West Germany as
well as males and females. All regressions control for age and education categories, and all possible
interactions. The administrative data lacks details on hours worked, so only wages for full-time workers can
be estimated. However, the share of part-time observations with censored wages is negligibly small (less
than 1 percent).

misreporting and inconsistencies using the procedure proposed by
Fitzenberger/Osikominu/Völter (2006). Then, we build a categorical variable with five
distinct values: 1) intermediate school leaving certificate without vocational training, 2)
intermediate school leaving certificate with vocational training, 3) upper secondary school
leaving certificate without vocational training, 4) upper secondary with vocational training,
5) college or university degree.
To exploit the role of children, we will use established proxies for motherhood
(Mueller/Strauch, 2017).16 The proxy uses family-related breaks in the employment
biography of females to identify childbirth in the administrative data. For identification, the
approach uses either employment notifications (maternity allowance payments by the
statutory health insurance provider during paid maternal leave) or detailed process data of
the Federal Employment Agency (e.g., withdrawal into maternity allowance) about
unemployment and benefits. Since the procedure is suitable for all of the administrative
data, we can run it on our linked JVS-IEB sample and hence identify women with children
among the identified JVS hires.
3.3 Final Sample
For our analysis, we use the JVS from 2010–2016.17 We then link the administrative data to
the survey information. In the end, our estimation sample consists of 21,694 distinct new
hires for which we have further information on the recruitment process such as the pool of
applicants. Furthermore, we can link workers’ full employment history to the new hires
data. Table 1 shows descriptive statistics for our main variables, separately for females and
males.
16
The administrative data also allows to use a proxy for marriage (Baechmann et al., 2021). We have
experimented with this proxy. However, motherhood appears to be the more meaningful variable to use.
17
Due to legal reasons, we can only link individual information from the administrative sources to the JVS from
2010 onward.

Table 1: Main variables by gender
females males
individual characteristics mean std. dev. mean std. dev.
age 35.86 10.75 36.46 10.91
education (scale 1-5)* 2.39 1.73 2.05 1.66
experience (years) 8.19 8.19 9.67 8.38
match characteristics
contract hours 34.40 7.69 38.85 4.20
job requirements (scale 1-4) ** 2.17 0.61 2.12 0.63
firm size decile 5.47 2.92 5.44 2.88
firm wage premium decile 5.47 2.89 5.58 2.84
log daily earnings 4.13 0.47 4.36 0.44
log daily earnings if full-time 4.22 0.42 4.37 0.42
Note: *1) intermediate school leaving certificate without vocational training, 2) as 1) but with vocational training, 3) upper secondary
school leaving certificate without vocational training, 4) as 3) but with vocational training, 5) College or university degree; ** 1)
missing 2) unskilled 3) vocational training, 4) college or university; Source: JVS, IEB;
On average, at the time of being hired, males are about half of a year older and about 1.4
years more experienced than females in our sample. Females are somewhat more
educated. Males work on average around 4 hours longer. Females and males are not
different with respect to the formal job requirements that are linked to the positions in
which they get hired. The same is true with respect to firm size. However, when we look at
earnings outcomes, we observe large differences. The unconditional difference in daily
hiring earnings amounts to 23 log points on average for all jobs in our sample and to 15 log
points for full-time jobs.18 Figure 2 shows the distributions of the hiring earnings for females
and males in full-time jobs.
In contrast to most other datasets, the IAB Job Vacancy Survey contains information on the
pool of applicants for a particular hire. Specifically, firms report the number of male and
female applicants for their most recent hire. Hence, we can calculate the share of
male/female applications. Table 2 shows the distribution of the share of male applications
for different occupations.19 Women are for example more likely to apply in health care
related occupations than men, while the opposite is the case in occupations related to
construction and architecture. Table A.1 in the Appendix shows similarly distinct
18
We define the hiring earnings as earnings within the first employment spell in the administrative data that
refers to the new hire.
19
Note that the shares of female and male applications always sum up to one for each hire and thereby also
for each occupation.

Figure 2: Hiring earnings distribution by gender
1
.8
.6
Density
.4
.2
full-time working males

full-time working females
0
2 3 4 5 6 7
log daily earnings
Note: Kernel density estimates for full-time workers using an epanechnikov kernel with bandwidth of 0.1. Source: JVS, IEB.
Table 2: Share of male/female hires and applicants across occupations
share of hires share of applicants

Occupation in (KldB2010 1-digit) total hires males (%) females (%) males (%) females (%)
1 agriculture, 701 68.47 31.53 66.46 33.54
forestry, farming, etc.
2 production of raw 4,785 84.91 15.09 82.65 17.35
materials, manufacturing etc.
3 construction, architecture, 1,601 90.82 9.18 88.90 11.10
techn. building services etc.
4 natural sciences, 894 77.85 22.15 75.80 24.20
geography, informatics etc.
5 traffic, logistics, etc. 1,910 80.00 20.00 76.16 23.84
6 commercial services, 1,814 40.24 59.76 40.26 59.74
trading, sales, hotels, etc.
7 business organisation, 5,643 30.96 69.04 34.82 65.18
accounting, law, etc.
8 health care, the social sector, 3,679 17.75 82.25 19.44 80.56
teaching, education etc.
9 philology, humanities, 574 41.99 58.01 43.48 56.52
soc. sciences, media, etc.
Total 21,604 53.67 46.33 58.66 41.34
Source: JVS, IEB.
application patterns across industry sectors. The share of male applicants is for example
much larger in manufacturing than in certain service sectors (e.g. related to education).20
20
In line with results by Gomes/Kuhn (2019), female application rates are much larger in the public sector than
in the rest of the economy. See Appendix.

4 Empirical Results
4.1 Application and Selection Patterns at the Firm Level
We start by looking at the application and selection behavior at particular firms through the
lens of our theoretical model. For this purpose, we use the information on the pool of
applicants for different jobs from the IAB-Job Vacancy Survey. We know the gender
composition of application pools, i.e., the number of male and female applicants. However,
we do not know any further characteristics of these applicants. In later steps, we will also
use information on the characteristics of the person who was actually hired and the
characteristics of the job.
In the theoretical model, higher firm-specific wages may either be driven by a larger output
per worker or wage formation.21 As we do not have any value added or sales information in
the IAB Job Vacancy Survey, we analyze how the gender-specific application behavior
differs across firm fixed effects from a two-way fixed effects regressions as described in
Bellmann et al. (2020) and Lochner/Seth/Wolter (2020).22
Panel (a) of Figure 3 shows the share of male and female applicants for each of these firms,
ranked according to AKM firm-fixed effect deciles (with the firms that pay the largest
average discount on the left-hand side, and the firms with the largest premium on the
right-hand side). At the highest earnings premiums, the share of male applicants is more
than 20 percentage point larger than the share of female applicants. At the bottom of the
earnings premium distribution the opposite is true, with a 10 percentage point larger
female application share at firms that pay the lowest premiums.
21
We do not model different wage formation mechanisms. However, in the Appendix we show that our key
results on the application and selection behavior are robust for different wage formation regimes.
22
These authors run an AKM wage regression on the universe of German administrative data for 2010-2017 in
the spirit of Abowd/Kramarz/Margolis (1999). These effects imply firm-specific wage premiums (or
discounts), often associated with rent-sharing, efficiency wages, or strategic wage posting behavior (see
among others Card/Heining/Kline, 2013; Postel-Vinay/Robin, 2002; Burdett/Mortensen, 1998)

Figure 3: Application and selection rate by gender and AKM firm effect deciles
(a) Share of female/male applications (b) Residualized share of female/male applications
.7 .1
.6 .05
.5 0
.4 -.05
99% CIs 99% CIs

share male applicants male application rate
share female applicants female application rate
.3 -.1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
AKM Firm Effect Decile AKM Firm Effect Decile
(c) Female/male selection rate (d) Residualized female/male selection rate

.4
.06
.35
.04
.3 .02
.25 0
-.02
.2
99% CIs 99% CIs
male selection rate male selection rate
female selection rate -.04 female selection rate
.15
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Note: Full-time jobs only. Variables are defined as follows: a) and b) share male appl.=number of male appl./ number of all appl., share
female appl.=number of female appl./ number of all appl. c) and d) male selection rate=1/number of male appl. if male hired, in this case
female selection rate equals zero, female selection rate=1/number of female appl. if female hired, in this case male selection rate equals
zero; Control variables: Industry categories (Nace Rev 2), firm size categories, occupation categories (5 digits); Source: JVS, IEB.
A sizeable part of these patterns may be driven by women and men applying in different
sectors and occupations, as visible in Table 2 and Table A.1. Therefore, we control for
occupation, industry, and firm size in panel (b) of Figure 3, as these are variables that are
typically included in Mincer-type wage regressions. Although the differences between male
and female application behavior are quantitatively less pronounced when adding controls,
the striking insight is that a substantial gap in the application behavior remains. There is a
roughly 7 percentage points larger probability for men to apply at the highest-wage firms
and a 10 percentage points larger probability for women to apply at the lowest-wage firms.
Through the lens of our model, this large difference in the gender-specific application
behavior may either be driven by taste-based discrimination at the hiring stage or by
different production functions at different jobs.

In the Appendix, we show that higher female application rates at low-paying firms and
lower female application rates at high-paying firms are a very robust result (both for the raw
data and the residualized version). This is true within different task complexity groups (see
Appendix A.2.3), when firm fixed effects are estimated separately for men and women (see
Appendix A.2.4), for different wage formation regimes (see Appendix A.2.5), or when giving
up the full-time restriction (see Appendix A.3).
To analyze the second stage of the matching process, we propose a proxy for the
gender-specific selection rate of firms conditional on application (in line with our model).
We define the gender-specific selection rate as follows (in analogy with the selection rate
from the model, see equation (8)): If a female (male) was hired, the female (male) selection
rate is 1 over the number of female (male) applicants and 0 for the gender that was not hired
(if there are applicants from this gender). Assume a firm had 5 applicants, two females and
three males. Assume further that a female (male) is hired. In this case, the probability of a
female to be selected from the pool of females is 50 (0) percent and the male selection rate
is 0 (33) percent. Our selection measure follows the proposition by Hochmuth et al. (2021)
and Lochner et al. (2021) to define the selection rate as the inverse of the number of
applicants based on the JVS. 23
Panel (c) of Figure 3 shows that the (uncontrolled) selection rate for men and women is
remarkably similar across AKM-deciles. Most importantly, at firms with the highest wage
premiums, the probability of men and women getting hired/selected (conditional on
applying) is almost the same (with confidence bands overlapping). When we control for
sector, occupation, and firm size in panel (d), male and female selection rates are almost the
same in all deciles. The confidence bands overlap in all deciles.
In the Appendix, we show that the indistinguishable female and male selection rates at
different AKM deciles are a very robust result (after controlling for observables). In Appendix
23
This definition of the selection rate yields several realistic properties that are in line with model predictions.
Hochmuth et al. (2021) show the the aggregate selection rate is procyclical over the business cycle (i.e., firms
get less selective in booms).Lochner et al. (2021) show that the selection rate is positively correlated with the
employment growth distribution (for growing firms). In different words, growing firms are less selective than
firms with a constant workforce. In addition, firms that do a lot of replacement hiring are less selective.

A.2.3, we show that our results also hold for other selection measures. Furthermore, our
results are robust within different task complexity groups (see Appendix A.2.3), when firm
fixed effects are estimated separately for men and women (see Appendix A.2.4), for different
wage formation regimes (see Appendix A.2.5), or when giving up the full-time restriction
(see Appendix A.3).
Given the stark differences in gender-specific application rates and the strong similarities in
selection rates across AKM-deciles, the model mechanism that high-paying firms
discriminate more strongly against women than low-paying firms (and thereby drive up the
earnings gap) is not supported by the empirical gender-specific selection patterns. By
contrast, the patterns are reconcilable with the second hypothesis that high-paying firms
offer different jobs (namely, nonlinear jobs) and predominantly attract workers that are
willing to provide the necessary flexibility. Thereby, women who sort into these high-paying
firms may have the same probability as men to be selected. We will analyze this hypothesis
in more detail in the next subsections.
4.2 Application Behavior and Firm-Sided Flexibility Requirements
While our previous analysis was at the firm level, we now move to the job level. The IAB Job
Vacancy Survey offers several proxies for firm-sided flexibility requirements. They serve as
proxy for Goldin (2014)’s hypothesis of different production functions. All the information
we use is available at the job level. Thus, we do not have to rely on flexibility definition
based on occupations codes and we can use the variation within occupations (by adding
fixed effects).
We use four different flexibility requirements from the IAB Job Vacancy Survey that are
asked for the last hire, namely the number of hours worked, the necessity to work overtime,
the necessity to change working hours on short notice, and the necessity to be mobile in
terms of the work place (e.g., due to business traveling).24 In Figure 4, we plot these four
24
Employers answer whether these flexibility requirements happen "often," "rarely," or "never" for the last
hire. We experimented with further questions from the survey. These four selected dimensions seem to
reflect the flexibility dimension best.

Figure 4: The share of male applicants and flexibility requirements
(a) Number of hours (b) Overtime
.65
Share of male applicants

.6
.6
.55
.55
.5
.5
.45
.45
.4
30 35 40 45 1 1.5 2 2.5 3
hours overtime: never to frequently
(c) Change in working hours (d) Mobility

.7
.65
.6 .6
.55 .55
.5 .5
.45
1 1.5 2 2.5 3 .5 1 1.5 2 2.5 3

change schedule: never to frequently mobility: never to frequently
Note: Figures show binscatters with 50 bins and quadratic fit lines. To residualize the x-variable and y-variables, we regress each variable
on the controls, generate the residuals, and add the sample mean of each variable back to its residuals. We then group the x-axis variable
into equal-sized bins, compute the mean of the x-axis and y-axis variables within each bin, and create a scatterplot of these data points.
Control variables: Industry categories (Nace Rev 2), firm size categories, occupation categories (5 digits); Full-time jobs only; Source: JVS,
IEB.
employer-sided flexibility requirements against the (residualized) share of male

applicants.25 In line with the second model hypothesis that there are different types of jobs,
all four flexibility requirements comove positively with the share of male applicants for
these particular jobs. Thus, these figures show that higher employer-sided flexibility
requirements are associated with a larger share of male applicants.
In reality, flexibility requirements are multidimensional. Although the survey questions in

the IAB Vacancy are much more detailed in this dimension than in many other surveys, we
believe that employer-sided flexibility requirements can only be captured partially.26 Given
25
Both the horizontal and the vertical axis are residualized by sector, occupation, and firm size.
26
This concept follows the idea by Goldin (2014: p.1104): "By job flexibility I mean a multitude of temporal
matters including the number of hours, precise times, predictability and ability to schedule ones own hours."

the strong connection between observed flexibility requirements and the share of male
applicants, we consider the share of male applicants as a suitable proxy for
multidimensional flexibility requirements. We will use this proxy for our further empirical
analysis. In the next step, we will analyze how the residual gender earnings gap is affected
by the gender-specific application behavior.
4.3 Residual Gender Earnings Gap
We start by estimating standard Mincer-type regressions, where we control for a rich set of
observables. In addition, we add a dummy for females to estimate the size of the residual
gender earnings gap. Recall that we observe new hires, hence we estimate the gap in hiring
earnings without potential gender-specific tenure effects. In a second step, we add our
proxy for firm-sided flexibility requirements, namely the share of male applicants. This
variable is absent in standard datasets. Thereby, we can check how much of the residual
gender earnings gap is due to an omitted variable bias.
Our benchmark Mincer-type regression looks as follows:
Log wagei,t = α genderi,t + γ controlsi,t + errori,t , (27)
where i is the recruitment from the cross-sectional JVS in year t (2010 to 2016), and gender
is a dummy for female hires (with male as the reference group). In our benchmark
specification, the set of controls includes the total number of applicants, worker age fully
interacted with education attainment (measured by five categories), experience in years as
well as its squared term, an indicator variable for the previous labor market status
(non-employed, unemployed, employed), the contractual hours of the new job, formal job
requirements (four categories), and year dummies. We estimate equation (27) for various
specifications, which include additional controls. Specifically, we subsequently add a full
set of dummies for industries, occupations, establishment size deciles, and all dummies at
the same time.

The left-hand side of Figure 5 shows the estimated α-coefficients for different regression
specifications as laid out in the Figure legend. The estimated gender gap in the hiring
earnings is 15 percent in our benchmark specification. Including a set of industry or
occupation categories, or establishment size dummies to the control variables barely
changes this pattern. Even if we add all these additional controls at once to the benchmark
specification, the gender-gap in the hiring earnings is in the same ballpark. This is the same
order of magnitude as in the existing literature for Germany (see for example Fuchs et al.,
2019).
Figure 5: The gender hiring earnings gap
Note: The Figure shows the estimates for the gender gap (α) in the hiring earnings as specified in equation 28. Dependent variable:
imputed log daily earnings. Default independent variables: gender dummy, the total number of applicants, worker age fully interacted
with education attainment (measured by five categories), experience in years as well as its squared term, an indicator variable for the
previous labor market status (non-employed, unemployed, employed), the contractual hours of the new job, formal job requirements
(four categories), and year dummies. Estimates for full-time workers only. Source: JVS, IEB;
In the second step, we add the share of male applicants as an additional explanatory
variable to control for the flexibility requirements of different jobs:
Log wagei,t = α genderi,t + β share male appl.i,t + γ controlsi,t + errori,t . (28)

The right-hand side of Figure 5 shows that adding the gender-share of applicants reduces
the gap in the hiring earnings to 6.1 percent (a reduction of 59 percent) in the benchmark
specification. The same pattern holds in all other specifications. When adding the share of
male applicants to the regressions, the residual earnings gap drops substantially.27
Figure 6: Coefficients for categories of share of male applicants, male hires
-.07
-.051
% male appl.=low -.055
-.061
-.042
.072
.063
% male appl.=high .085
.059
.072
.049
Mincer
.038
Mincer + (10) industry FEs
% male appl.=1 .082
Mincer + (5 digit) occupation FEs
.058
Mincer + (10) size FEs
.085 Mincer + all
-.1 -.05 0 .05 .1
Note: The Figure shows the coefficients for the share of male applicants (β ) as specified in equation 28. Dependent variable: imputed log
daily earnings; Default independent variables: the total number of applicants, worker age fully interacted with education attainment
(measured by five categories), experience in years as well as its squared term, an indicator variable for the previous labor market status
(non-employed, unemployed, employed), the contractual hours of the new job, formal job requirements (four categories), and year
dummies. Five categories for the number of male appl. (only females, low male share, medium male share (reference), high male share,
only males); Estimates for full-time male workers only. Source: JVS, IEB.
Under the second theoretical hypothesis, jobs with a high share of male applicants are
different from those with a lower share of male applicants. Both men and women (and not
only men) should earn more than men and women with comparable observable
characteristics. To be able to test this further, we construct a categorical variable instead of
the continuous share of male applicants. We distinguish five categories: one if a vacancy
has only female applications, five if there are only male applications, and two, three, and
27
In further robustness checks, we restricted our sample to only female-dominated jobs and used an
alternative occupational classification. The pattern that the residual gender earning gap drops substantially
when adding the share of male applicants holds in all specifications. Results are available on request.

four in between.28 Two refers to a low, three to a medium, and four to a high share of male
applicants. We choose a medium share of male applicants as the reference group, which
allows us to compare the coefficients across genders.
Figure 6 shows the estimated coefficients for the categorical variable for hired men only.29
Men who match at a job with a high share of male applicants earn 5.9 to 8.5 percentage
points higher earnings compared to those who match with a medium share.
Figure 7: Coefficients for categories of share of male applicants, female hires
only males); Estimates for full-time female workers only. Source: JVS, IEB.
Figure 7 shows the estimated coefficients for the categorical variable from regressions for
hired females only. In line with the second hypothesis in our theoretical model, the
coefficients are increasing in the share of male applicants. We observe large effects in all our
regressions. For instance, in our benchmark specification, a female recruitment with a zero
28
Figure A.1 in the Appendix shows the categories. We divide the inner part of the distribution into three parts.
In the first part the mean of male applicants is 21 percent, in the second it is 48 percent, and in the third it is
80 percent)
29
We again focus on full-time workers. Figure B.12 and B.13 show that all our findings are qualitatively
unaltered once we include part-time workers.

share of male applicants on average results in 7.0 to 9.8 percentage points lower earnings as
compared to a female recruitment where there was a medium share of male applicants. On
the other hand, depending on the exact specification, a female recruitment with high share
of male applicants on average results in 7.4 to 9.4 percentage points higher earnings as
compared to a female recruitment where there was a medium share of male applicants.
These numbers show that females earn substantially higher earnings if they match in
comparable jobs with a high share of male applicants compared to zero male applicants.
These patterns in the data provide further evidence for the hypothesis that jobs with a
larger share of male applicants are different from those with a low share of male applicants.
Employers appear to provide compensating differentials for the higher degree of
employer-sided flexibility requirements.
4.4 Evidence for Nonlinear Jobs on the Person Level
In our final step, we analyze the interaction of the share of male applicants with
characteristics of the person who matched. More precisely, we analyze the connection
between the share of male applicants and the worker fixed effect from the two-way fixed
effects regressions.30 In addition, we check how being a mother affects the residual gender
earnings gap and how this interacts with the share of male applicants.
Figure 8a shows the residualized share of male applicants and the residualized worker fixed
effect of the hired workers from the AKM two-way fixed effects regression. A larger share of
male applicants is associated with a larger AKM worker fixed effect. Through the lens of our
model, workers that are willing/able to provide a high input and who are hired in a
nonlinear job will produce more than hires in linear jobs. A certain fraction of this higher
production will be passed on in form of higher wages (under Nash bargaining or any other
wage formation where wages depends on produced output) and show up as larger
30
As our data is a cross-section of hires, we cannot estimate person fixed effects directly. However, we can use
the worker fixed effects that were estimated on the universe of German administrative data and link it to our
cross section.

worker-specific wage premiums. Figure 8b shows the relation between the AKM person
effects of hired workers and the share of male applicants separately for hired males and
females. A higher share of male applicants is associated with higher AKM worker fixed
effects both for men and women. Thus, higher flexibility requirements at certain jobs are
associated with higher worker fixed effects for both genders.
Figure 8: AKM Person effects and the gender distribution of the application pool
Note: Figures show binscatters with 50 bins and linear fit lines. To residualize the x-variable and y-variables, we regress each variable on
the controls, generate the residuals, and add the sample mean of each variable back to its residuals. We then group the x-axis variable into
equal-sized bins, compute the mean of the x-axis and y-axis variables within each bin, and create a scatterplot of these data points. Control
variables: Industry categories (Nace Rev 2), firm size categories, occupation categories (5 digits); Full-time jobs only; Source: JVS, IEB.
Obviously, the connection between the share of male applicants and AKM worker fixed
effects cannot be interpreted causally. The worker fixed effects capture unobserved worker
heterogeneity and differences in the worker fixed effect in Figure 8a may therefore (partly)
be driven by ex-ante worker ability. Against this background, the positive correlation is in
line with the result by Lamadon/Mogstad/Setzler (2022) who show that compensating
differentials are larger for high-ability workers and smaller for low-ability workers. As
shown before, applicant pools with a larger share of male applicants can be found at firms
with higher firm fixed effects and are thus associated with higher pay (i.e., a compensating
differential for higher employer-sided flexibility requirements).31
The quantitative difference between these two estimated curves in Figure 8b is relatively
small (i.e. an order of magnitude smaller than the gender earnings gap when not controlling
for the share of male applicants). Thus, the connection between the share of male
31
Note that worker ex-ante heterogeneity is absent in our model and therefore the model is silent on this issue.

applicants and AKM worker fixed effects is on average very similar for men and women.
So far, our empirical results suggest that high-flexibility jobs (i.e., those with a larger share
of male applicants) are associated with a disamenity and thereby pay compensating
differentials. At the person level, we can also test the hypothesis whether these patterns are
driven by different production functions. Assume that a person that is unable to provide
high-flexibility matches at a firm with nonlinear production function. In this case, our model
would predict low output at this job and a particularly large earnings discount for the
matched person. Although we do not have any information about the degree of flexibility
that a person can provide, we consider motherhood as a suitable proxy. Mothers in Germany
still bear a larger fraction of childcare than fathers and thereby tend to be less flexible.
Therefore, we use the established proxy for being a mother in the administrative data
(Mueller/Strauch, 2017). Based on this proxy, we estimate the residual gender earnings gap
relative to males for female mothers and for childless females. Column (1) of Table 3 shows
that the residual gender earnings gap is about 6 percentage points larger (-20 vs. -14
percent) for mothers compared to childless women. When we add our proxy for firm-sided
flexibility requirements (i.e., the share of male applicants) to the regression in column (2),
the gap between female mothers and childless females remains similarly large (-12 vs. -6
percent). Overall, this exercise shows that mothers face a larger hiring earnings discount in
the labor market than women without children.
Table 3: Gender earnings gap for mothers and childless females
(1) (2)
log earnings log earnings
mother -0.2024∗∗∗ -0.1231∗∗∗
(male=reference) (0.0142) (0.0158)
childless female -0.1389∗∗∗ -0.0637∗∗∗
(male=reference) (0.0071) (0.0093)
Observations 12,945 11,631
Adjusted R2 0.6038 0.6126
Notes: Estimates for full-time workers only; Standard errors in parentheses; Controls: total number of applicants, a set of worker age
dummies fully interacted with education dummies, experience in years as well as its squared term, a dummy for the previous labor
market status (non-employed, unemployed, employed), the hours of the new contract, dummies for formal job requirements, year
dummies, industry categories, occupation categories, and establishment size deciles; Columns (2) additionally control for the share
of male applicants; ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01; Source: JVS, IEB.

Finally, we interact the share of male applicants with dummies for mothers and women
without children. Figure 9 shows the (predicted) earnings discount for mothers and women
without children (relative to men) split up according to the shares of male applicants at the
respective jobs (from 0.1 to 0.9, as share of 0 and 1 have to be excluded as only one gender
matches at those jobs).32 When mothers match at a job with a 90 percent share of male
applicants, they face a more than 20 percent residual gender earnings gap relative to men,
while this number is very small at low shares of male applicants. Note that the weighted
average of these estimates corresponds to the point estimates in column (2) of Table 3. This
is in line with our interpretation that jobs with high share of male applicants tend to be
nonlinear jobs. As mothers are unable to provide the employer-sided (desired) flexibility,
they produce less and thereby face a large earnings discount. It is also striking that the
earnings discount differential between mothers and childless females increases with the
share of male applicants. While the differences in the point estimates are economically very
small for matches with small shares of male applicants, it is more than 15 percentage points
for matches with 90 percent male applicants.33
It is also worthwhile discussing that the earnings discount for childless females also
increases in the share of male applicants. First, the economic differences between the
highest and lowest share of male applicants are relatively small. Second, having children is
an incomplete proxy for the ability and willingness of women to provide flexibility. It is for
example well known that women also bear a larger burden of care responsibilities that are
not related to children (e.g., elderly care). Thus, even women without children may on
average be less flexible than men.
32
We include an interaction term of the share of male applicants as a continuous variable with a dummy
variable which has distinct values for mothers and women without children relative to males in our
regression. Based on this regression, we then calculate marginal effects over a grid of values of the share of
male applicants.
33
The confidence bands are larger for larger share of male applicants as the number of observations is small.
This is due to two reasons. First, due to the matching of the IAB Job Vacancy and administrative data, the
sample size is reduced. Second, by definition at jobs with a larger share of male applicants, the absolute
number of females and even more so for mothers is small.

Figure 9: Mothers and Women without Children
0
Effects on linear prediction
-.1
-.2
mothers
-.3 childless females
.1 .2 .3 .4 .5 .6 .7 .8 .9
share of male applicants
Note: Figure shows the earnings gap (marginal effects) for mothers and childless females compared to males as a reference group at various
levels of the share of male applicants. Controls: the share of male applicants interacted with a dummy for mothers and childless females
(male=reference), the total number of applicants, a set of worker age dummies fully interacted with education dummies, experience in years
as well as its squared term, a dummy for the previous labor market status (non-employed, unemployed, employed), the hours of the new
contract, dummies for formal job requirements, year dummies, industry categories, occupation categories, and establishment size deciles;
Full-time jobs only; Source: JVS, IEB.
5 Conclusion
This paper shows that gender-specific application behavior is key for understanding hiring
earnings differences. Even within industries, firm size categories, and occupations, women
are 10 percentage points more likely to apply at the lowest-wage firms than men. Our
theoretical labor market flow model rationalizes this behavior based on different
production functions at different jobs, where the highest paying jobs are nonlinear in input,
as defined by Goldin (2014).
We show that the share of applicants is a positive function of various measurable

dimensions of employer-sided flexibility requirements. Therefore, we consider it as a

suitable proxy for multidimensional flexibility requirements at the job-level. Once we
include this proxy into standard Mincer regressions (beyond standard observable variables
such as occupations, sectors, and worker characteristics), the residual gender-earnings gap
drops by 50-60 percent. This illustrates that the gender-specific application behavior is an
important explanatory variable that is typically omitted in Mincer-type wage regressions, as
it is not contained in standard datasets.
Our paper combines information from the IAB Job Vacancy Survey with administrative
information on the last hire. This combination allows us to use a proxy whether women
have children. We show that earnings discounts are particularly large for women with
children. This earnings discount increases in our proxy for employer-sided flexibility. Again,
this is in line with the nonlinear jobs hypothesis. When women with children match at
nonlinear jobs, they are less able to provide a high-degree of employer-sided flexibility and
thereby face a large earnings discount.
Our paper offers variable policy-relevant lessons. Policy interventions that allow women to
get access to jobs with high-flexibility requirements (such as better access to childcare or
incentives for different intra-family sharing of care responsibilities) will change their
application behavior and thereby can reduce the gender-earnings gap. Furthermore, the
Covid-19 pandemic has shown that a different organization of work is possible (e.g., more
working from home arrangements). Only future research will show whether this new work
environment will stick and whether it will boost women’s possibilities to get better access to
jobs with high-flexibility requirements.

References
Abowd, John M.; Kramarz, Francis; Margolis, David N. (1999): High Wage Workers and High
Wage Firms. In: Econometrica, Vol. 67, No. 2, p. 251–333.
Azmat, Ghazala; Ferrer, Rosa (2017): Gender Gaps in Performance: Evidence from Young
Lawyers. In: Journal of Political Economy, Vol. 125, No. 25, p. 1306–1355.
Baechmann, Ann-Christin; Frodermann, Corinna; Lochner, Benjamin; Oberfichtner, Michael;

Trenkle, Simon (2021): Identifying Couples in Administrative Data for the years
20012014. FDZ-Methodenreport 202103 (en), Institute for Employment Research,
Nuremberg, Germany.
Barrero, Jose Maria; Bloom, Nicholas; Davis, Steven (2020): Why Working From Home Will
Stick. Discussion Paper 174, Becker-Friedman-Institute, University of Chicago.
Bellmann, Lisa; Lochner, Benjamin; Seth, Stefan; Wolter, Stefanie (2020): AKM Effects for
German Labour Market Data, fDZ Methodenreport 01/2020, IAB Nuremberg.
Blau, Francine D.; Kahn, Lawrence H. (2017): The Gender Wage Gap: Extent, Trends, and
Explanations. In: Journal of Economic Literature, Vol. 55, No. 3, p. 789–865.
Bolotnyy, Valentin; Emanuel, Natalia (2019): Why Do Women Earn Less Than Men? Evidence
from Bus and Train Operators. In: Harvard University, mimeo.
Brown, Alessio; Merkl, Christian; Snower, Dennis (2016): An Incentive Theory of Matching. In:
Macroeconomic Dynamics, Vol. 131, No. 2, p. 633–686.
Budig, Michelle J.; Hodges, Melissa J. (2010): Differences in Disadvantage: Variation in the
Motherhood Penalty across White Womens Earnings Distribution. In: American
Sociological Review, Vol. 75, No. 5, p. 705–728.
Burdett, Kenneth; Mortensen, Dale T. (1998): Wage Differentials, Employer Size, and
Unemployment. In: International Economic Review, Vol. 39, No. 2, p. 257–273.

Card, David; Cardoso, Ana Rute; Kline, Patrick (2016): Bargaining, Sorting, and the Gender
Wage Gap: Quantifying the Impact of Firms on the Relative Pay of Women. In: Quarterly
Journal of Economics, Vol. 19, No. 3, p. 643–686.
Card, David; Heining, Jörg; Kline, Patrick (2013): Workplace Heterogeneity and the Rise of
West German Wage Inequality. In: The Quarterly Journal of Economics, Vol. 128, No. 3,
p. 967–1015.
Carrillo-Tudela, Carlos; Gartner, Hermann; Kaas, Leo (2020): Recruitment Policies,

Job-Filling Rates and Matching Efficiency. Discussion Paper 13240, IZA.
Chugh, Sanjay K.; Merkl, Christian (2016): Efficiency and Labor Markt Dynamics in a Model of
Labor Selection. In: International Economic Review, Vol. 57, No. 4, p. 1371–1404.
Cook, Cody; Diamond, Rebecca; Hall, Jonathan; List, John A.; Oyer, Paul (2021): The Gender
Earnings Gap in the Gig Economy: Evidence from over a Million Rideshare Drivers. In:
Review of Economic Studies, Vol. 88, No. 5, p. 2210–2238.
Cortés, Patricia; Pan, Jessica; Pilossoph, Laura; Zafar, Basit (2021): Gender Differences in
Job Search and the Earnings Gap: Evidence from Business Majors. Working Paper
28820, National Bureau of Economic Research.
Dustmann, Christian; Ludsteck, Johannes; Schönberg, Uta (2009): Revisiting the German
Wage Structure. In: The Quarterly Journal of Economics, Vol. 124, No. 2, p. 843–881.
Faberman, R. Jason; Mueller, Andreas I.; ahin, Ayegül; Topa, Giorgio (2017): Job Search and
the Gender Wage Gap. In: Society for Economic Dynamics, Meeting Papers.
Fitzenberger, Bernd; Osikominu, Aderonke; Völter, Robert (2006): Imputation Rules to

Improve the Education Variable in the IAB Employment Subsample. In: Schmollers
Jahrbuch : Journal of Applied Social Science Studies / Zeitschrift für Wirtschafts- und
Sozialwissenschaften, Vol. 126, No. 3, p. 405–436.
Fluchtmann, Jonas; Glenny, Anita M.; Harmon, Nikolaj; Maibom, Jonas (2020): The Gender
Application Gap: Do men and women apply for the same jobs? Unpublished
manuscript.

Fuchs, Michaela; Rossen, Anja; Weyh, Antje; Wydra-Somaggio, Gabriele (2019):
Unterschiede in der Lohnlücke erklären sich vor allem durch die Betriebslandschaft vor
Ort. In: IAB Kurzbericht, Vol. 10.
Goldin, Claudia (2014): A Grand Gender Convergence: Its Last Chapter. In: American
Economic Review, Vol. 104, No. 4, p. 1091–1119.
Gomes, Pedro; Kuhn, Zoe (2019): You’re the One that I Want! Public Employment and
Women’s Labor Market Outcomes. IZA Discussion Paper 12702.
Hochmuth, Brigitte; Kohlbrecher, Britta; Merkl, Christian; Gartner, Hermann (2021): Hartz IV
and the Decline of German Unemployment: A Macroeconomic Evaluation. In: Journal of
Economic Dynamics and Control, , No. 127, p. 104 114.
Illing, Hannah; Schmieder, Johannes F; Trenkle, Simon (2021): The Gender Gap in Earnings
Losses after Job Displacement. Working Paper 29251, National Bureau of Economic
Research.
Lamadon, Thibaut; Mogstad, Magne; Setzler, Bradley (2022): Imperfect Competition,

Compensating Differentials, and Rent Sharing in the US Labor Market. In: American
Economic Review, Vol. 112, No. 1, p. 169–212.
Le Barbanchon, Thomas; Rathelot, Roland; Roulet, Alexandra (2020): Gender Differences in

Job Search: Trading off Commute against Wage. In: The Quarterly Journal of
Economics, Vol. 136, No. 1, p. 381–426.
Lochner, Benjamin (2019): A Simple Algorithm to Link "Last Hires" from the Job Vacancy
Survey to Administrative Records. In: FDZ-Methodenreport, Vol. 6.
Lochner, Benjamin; Merkl, Christian; Stüber, Heiko; Gürtzgen, Nicole (2021): Recruiting
Intensity and Hiring Practices: Cross-sectional and Time-series Evidence. In: Labour
Economics, Vol. 68, p. 101 939.
Lochner, Benjamin; Seth, Stefan; Wolter, Stefanie (2020): Decomposing the Large Firm Wage
Premium in Germany. In: Economics Letters, Vol. 194, p. 109 368.

Moczall, Andreas; Müller, Anne; Rebien, Martina; Vogler-Ludwig, Kurt (2015): The IAB Job
Vacancy Survey. Establishment Survey on Job Vacancies and Recruitment Processes.
Waves 2000 to 2013 and Subsequent Quarters from 2006. , FDZ Datenreport 04/2015,
Institut für Arbeitsmarkt- und Berufsforschung (IAB), Nürnberg.
Mueller, Dana; Strauch, Katharina (2017): Identifying mothers in administrative data.

FDZ-Methodenreport 201713 (en), Institute for Employment Research, Nuremberg,
Germany.
Postel-Vinay, Fabien; Robin, Jean-Marc (2002): Equilibrium Wage Dispersion with Worker
and Employer Heterogeneity. In: Econometrica, Vol. 70, No. 6, p. 2295–2350.
Sorkin, Isaac (2018): Ranking Firms Using Revealed Preference. In: The Quarterly Journal of
Economics, Vol. 133, No. 3, p. 1331–1393.
Taber, Christopher; Vejlin, Rune (2020): Estimation of a Roy/Search/Compensating

Differential Model of the Labor Market. In: Econometrica, Vol. 88, No. 3, p. 1031–1069.
Wiswall, Matthew; Zafar, Basit (2017): Preference for the Workplace, Investment in Human
Capital, and Gender. In: The Quarterly Journal of Economics, Vol. 133, No. 1, p. 457–507.
Xiao, Pengpeng (2021): Wage and Employment Discrimination by Gender in Labor Market
Equilibrium. In: VATT Working Papers, 144.

A.1 Descriptive Details
Table A.1: Share of male/female hires and applicants across industries
Share of hires Share of applicants

NACE Rev. 2 total hired males (%) females (%) males (%) females (%)
A - Agriculture, forestry and fishing 941 67.59 32.41 66.13 33.87
B - Mining and quarrying
C - Manufacturing 4,952 72.70 27.30 70.72 29.28
D - Electricity, gas, etc. 1,579 68.84 31.16 69.65 30.35
E - Water supply, sewerage, etc.
F - Construction 826 87.89 12.11 85.00 15.00
G - Wholesale and retails trade, etc. 1,613 68.20 31.80 65.36 34.64
H - Transportation and storage
I - Accommodation and food 664 41.27 58.73 39.03 60.97
J - Information and communication 4,470 52.24 47.76 52.99 47.01
K - Financial and insurance
L - Real estate
M - Professional, scientific and tech-
nical
N - Administrative and support ser-
vice
O - Public administration 1,860 34.68 65.32 37.17 62.83
P - Education 4,789 26.12 73.88 27.87 72.13
Q - Human health and social work
R - Arts, entertainment and recre-
ation
S - Other services
T - Households as employers
U - Extraterritorial organisations
Total 21,694 53.72 46.28 57.10 42.90
Source: JVS, IEB.

Figure A.1: Share of male applicants: categories
4
3
Density
2 1
0
1 2 3 4 5
Share of male applications (categories)
Source: JVS, IEB.

A.2 Additional Empirical Results
A.2.1 Age Cohorts
Figure B.1: GWG estimates by 5 year cohorts

(a) Full- and part-time workers (b) Full-time workers only
Note: Figure shows the estimates for the gender gap in the hiring earnings by age groups as laid out on the x-axis. Dependent variable:
(four categories), and year dummies. Source: JVS, IEB.
A.2.2 Alternative Selection Measures
Figure B.2 shows differently defined selection rates. Version 1 defines the selection rate as 1
divided by the overall number of applicants (instead of the gender-specific number of
applicants). Thus, it represents the probability of being selected from the overall pool of
applicants. Version 2 uses the number of gender-specific suitable applicants instead of all
applicants. Version 3 uses the measure proposed by Carrillo-Tudela/Gartner/Kaas (2020),
namely the number of suitable (gender-specific) applicants divided by the overall number
of (gender-specific) applicants. Firms may endogenously change their definition of which
candidate is suitable (i.e., more candidates are defined as suitable when firms want to hire
more). Interestingly, in all three cases, once we control for observables, there are no
meaningful differences between males and females selection rates. This confirms our
results from the main part.

Figure B.2: Alternative Selection Measures
(a) Overall appl. pool
(b) Residualized version (a)
.35
.06
.3
.04
.25
.02
.2
0
.15 -.02
99% CIs 99% CIs
female selection rate female selection rate
.1 -.04
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
(c) Gender-specific suitable appl. pool (d) Residualized version (c)

.6 .1
.5 .05
.4
0
99% CIs 99% CIs

.3 female selection rate female selection rate
-.05
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
(e) Gender-specific suitable appl. share (f) Residualized version (e)

.5
.06
.45 .04
.02
.4
.35
-.02
99% CIs 99% CIs
.3 male selection rate male selection rate
-.04
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Note: Full-time jobs only. Variables are defined as follows: a) and b) male selection rate=1/number of all appl. if male hired, in this case
female selection rate equals zero, female selection rate=1/number of all appl. if female hired, in this case male selection rate equals zero;
c) and d) male selection rate=1/number of male suitable appl. if male hired, in this case female selection rate equals zero, female selection
rate=1/number of female suitable appl. if female hired, in this case male selection rate equals zero; e) and f) male selection rate=number
of male suitable appl./number of male appl. if male hired, in this case female selection rate equals zero, female selection rate=number
female suitable appl./number of female appl. if female hired, in this case male selection rate equals zero; Control variables: Industry
categories (Nace Rev 2), firm size categories, occupation categories (5 digits); Source: JVS, IEB.
A.2.3 Application and Selection Behavior within Task Complexities
Figures B.3 and B.4 show the gender-specific residualized application and selection rates
within different task complexity groups (unskilled, trained, expert, specialist). They are
defined based on the fifth digit of the occupational code (KldB2010).

Figure B.3: Residualized share of male applicants over grid of AKM firm effect deciles by task complex-
ity
(a) Helper (b) Trained
.1
.05 .05
0 0
-.05 -.05
99% CIs 99% CIs

male application rate male application rate
female application rate female application rate
-.1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
(c) Specialist (d) Expert

.1 .1
.05 .05
0 0
-.05 -.05
99% CIs 99% CIs

male application rate male application rate
female application rate female application rate
-.1 -.1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Note: Full-time jobs only; Variables are defined as follows: a)-d) share male appl.=number of male appl./ number of all appl., share female
appl.=number of female appl./ number of all appl.; Control variables: Industry categories (Nace Rev 2), firm size categories, occupation
categories (5 digits); Source: JVS, IEB.

Figure B.4: Residualized selection rates over grid of AKM firm effect deciles by job level
(a) Helper (b) Trained
.2 .1
.1
.05
-.1
99% CIs 99% CIs

-.2 -.05
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
(c) Specialist (d) Expert

.1
.1
.05
.05
0
0
-.05 -.05
99% CIs 99% CIs

-.1 -.1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Note: Full-time jobs only; Variables are defined as follows: a)-d) male selection rate=1/number of male appl. if male hired, in this case
female selection rate equals zero, female selection rate=1/number of female appl. if female hired, in this case male selection rate equals
zero; Control variables: Industry categories (Nace Rev 2), firm size categories, occupation categories (5 digits); Source: JVS, IEB.

A.2.4 Application and Selection Behavior with Alternative Firm Fixed Effects
Figures B.5 and B.6 show the patterns in the data with differently estimated firm-fixed
effects. In this case, the firm-fixed effect is estimated separately for men and women (i.e.,
each firm has two wage premia: one for men and one for women).
Figure B.5: Application and selection rate by gender and AKM firm effect deciles (estimated from
males only)
.7
.1
.6
.05
.5 0
-.05
.4
99% CIs 99% CIs

share male applicants -.1 male application rate
.3
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

.1
.35
.3
.05
.25
.2
99% CIs 99% CIs

.15
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Note: Full-time jobs only. Firm effects estimates for males only. Variables are defined as follows: a) and b) share male appl.=number of
male appl./ number of all appl., share female appl.=number of female appl./ number of all appl. c) and d) male selection rate=1/number
of male appl. if male hired, in this case female selection rate equals zero, female selection rate=1/number of female appl. if female hired,
in this case male selection rate equals zero; Control variables: Industry categories (Nace Rev 2), firm size categories, occupation categories
(5 digits); Source: JVS, IEB.

Figure B.6: Application and selection rate by gender and AKM firm effect deciles (estimated from fe-
males only)
.1
.6
.05
.55
.5 0
.45
-.05
99% CIs 99% CIs

.4
-.1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

.4 .1
.35
.05
.3
.25
0
.2
99% CIs 99% CIs
.15
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Note: Full-time jobs only. Firm effects estimates for females only. Variables are defined as follows: a) and b) share male appl.=number of
male appl./ number of all appl., share female appl.=number of female appl./ number of all appl. c) and d) male selection rate=1/number
of male appl. if male hired, in this case female selection rate equals zero, female selection rate=1/number of female appl. if female hired,
in this case male selection rate equals zero; Control variables: Industry categories (Nace Rev 2), firm size categories, occupation categories
(5 digits); Source: JVS, IEB.

A.2.5 Application and Selection Behavior and Bargaining
Figures B.7 and B.8 show the application and selection behavior across AKM firm effect
deciles, separately for firms that are inside a collective or firm-level bargaining agreement
(denoted by organized bargaining) and those that are not, respectively. Although the
application rates differ somewhat in the raw data, once we control for our full set of
controls, the quantitative results are very similar to our baseline sample.
Figure B.7: Application and selection rate by gender and AKM firm effect deciles, with organized bar-
gaining
.1
.7
.6 .05
.5 0
.4 -.05
99% CIs 99% CIs

.3 share female applicants female application rate
-.1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

.1
.4
.35
.05
.3
.25
-.05
.2
99% CIs 99% CIs

.15
-.1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Note: Full-time jobs with organized bargaining only. Firm effects estimates for females only. Variables are defined as follows: a) and b)
share male appl.=number of male appl./ number of all appl., share female appl.=number of female appl./ number of all appl. c) and d)
male selection rate=1/number of male appl. if male hired, in this case female selection rate equals zero, female selection rate=1/number
of female appl. if female hired, in this case male selection rate equals zero; Control variables: Industry categories (Nace Rev 2), firm size
categories, occupation categories (5 digits); Source: JVS, IEB.

Figure B.8: Application and selection rate by gender and AKM firm effect deciles, without organized
bargaining
.7
.6 .05
.5 0
.4 -.05
99% CIs 99% CIs

.3
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

.1
.4
.35
.05
.3
0
.25
99% CIs 99% CIs

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Note: Full-time jobs without organized bargaining only. Firm effects estimates for females only. Variables are defined as follows: a) and b)
share male appl.=number of male appl./ number of all appl., share female appl.=number of female appl./ number of all appl. c) and d)
male selection rate=1/number of male appl. if male hired, in this case female selection rate equals zero, female selection rate=1/number
of female appl. if female hired, in this case male selection rate equals zero; Control variables: Industry categories (Nace Rev 2), firm size
categories, occupation categories (5 digits); Source: JVS, IEB.

A.3 Alternative Sample Restriction
This Appendix replicates all main results, without imposing the full-time restriction (i.e.,
only workers with more than 25 hours working time). All our key insights are unaffected by
the chosen sample restrictions, although the quantitative numbers differ somewhat.
Figure B.9: Application and selection rate by gender and AKM firm effect deciles
(a) Share of female/male applications (b) Resiudalized share of female/male applications
.7
.6 .05
.5 0
.4 -.05
99% CIs 99% CIs

.3
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
(c) Female/male selection rate (d) Resiudalized female/male selection rate

.4
.06
.35 .04
.02
.3
.25
-.02
99% CIs 99% CIs
-.04
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Note: Full-time and part-time jobs. Variables are defined as follows: a) and b) share male appl.=number of male appl./ number of all appl.,
share female appl.=number of female appl./ number of all appl. c) and d) male selection rate=1/number of male appl. if male hired, in this
case female selection rate equals zero, female selection rate=1/number of female appl. if female hired, in this case male selection rate
equals zero; Control variables: Industry categories (Nace Rev 2), firm size categories, occupation categories (5 digits); Source: JVS, IEB.

Figure B.10: The share of male applicants and flexibility requirements
(a) Mobility (b) Overtime
.65

.6 .6
.55 .55
.5 .5
.45 .45
.5 1 1.5 2 2.5 3 1 1.5 2 2.5 3

mobility: never to frequently overtime: never to frequently
(c) Change in working hours (d) Number of hours

.7
.6
.6
.55 .5
.5
.4
.45
.3
1 1.5 2 2.5 3 10 20 30 40 50
change schedule: never to frequently hours
Note: Figures show binscatters with 50 bins and quadratic fit lines. To residualize the x-variable and y-variables, we regress each variable
on the controls, generate the residuals, and add the sample mean of each variable back to its residuals. We then group the x-axis variable
into equal-sized bins, compute the mean of the x-axis and y-axis variables within each bin, and create a scatterplot of these data points.
Control variables: Industry categories (Nace Rev 2), firm size categories, occupation categories (5 digits); Full-time and part-time jobs;
Source: JVS, IEB.

Figure B.11: The gender hiring earnings gap
Note: The Figure shows the estimates for the gender gap (α) in the hiring earnings as specified in equation 28. Dependent variable:
(four categories), and year dummies. Estimates for full-time and part-time workers. Source: JVS, IEB.

Figure B.12: Coefficients for categories of share of male applicants, male hires
-.056
-.04
-.049
-.036
.071
.061
.06
.069
.052
Mincer
.04
% male appl.=1 .082
.062
.086 Mincer + all
-.1 -.05 0 .05 .1
only males); Estimates for full-time and part-time male workers. Source: JVS, IEB.
Figure B.13: Coefficients for categories of share of male applicants, female hires
-.078
-.08
% male appl.=0 -.077
-.057
-.059
-.045
-.047
-.037
-.047
.075
Mincer
.067
.074
.053 Mincer + all
-.1 -.05 0 .05 .1
only males); Estimates for full-time and part-time female workers. Sources: JVS, IEB.

Figure B.14: AKM Person effects and the gender distribution of the application pool
(a) Women and men (b) By gender
.8 4.3
.7 4.2
AKM worker effect
AKM worker effect

.6 4.1
.5 4
.4 3.9
3 3.5 4 4.5 5 -.5 0 .5 1 1.5
share of male applicants share of male applicants
Note: Figures show binscatters with 50 bins and linear fit lines. To residualize the x-variable and y-variables, we regress each variable on
the controls, generate the residuals, and add the sample mean of each variable back to its residuals. We then group the x-axis variable into
equal-sized bins, compute the mean of the x-axis and y-axis variables within each bin, and create a scatterplot of these data points.
Control variables: Industry categories (Nace Rev 2), firm size categories, occupation categories (5 digits); Full-time and part-time jobs;
Source: JVS, IEB.
Table B.1: Gender earnings gap for mothers and childless females
(1) (2)
log earnings log earnings
mother -0.1877∗∗∗ -0.1150∗∗∗
(male=reference) (0.0114) (0.0129)
childless female -0.1300∗∗∗ -0.0607∗∗∗
(male=reference) (0.0063) (0.0084)
Observations 18,324 16,390
Adjusted R2 0.6417 0.6498
Notes: Estimates for full-time and part-time workers; Standard errors in parentheses; Controls: total number of applicants, a set of
worker age dummies fully interacted with education dummies, experience in years as well as its squared term, a dummy for the
previous labor market status (non-employed, unemployed, employed), the hours of the new contract, dummies for formal job
requirements, year dummies, industry categories, occupation categories, and establishment size deciles; Columns (2) additionally
control for the share of male applicants; ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01; Source: JVS, IEB.

Figure B.15: Mothers and Childless Females
0
Effects on linear prediction
-.1
-.2
mothers
-.3
childless females
.1 .2 .3 .4 .5 .6 .7 .8 .9
share of male applicants
Note: Figures show the earnings gap (marginal effects) for mothers and childless females compared to males as a reference group at various
levels of the share of male applicants. Controls: the share of male applicants interacted with a dummy for mothers and childless females
(male=reference), the total number of applicants, a set of worker age dummies fully interacted with education dummies, experience in years
as well as its squared term, a dummy for the previous labor market status (non-employed, unemployed, employed), the hours of the new
contract, dummies for formal job requirements, year dummies, industry categories, occupation categories, and establishment size deciles;
Full-time and part-time jobs; Source: JVS, IEB.

List of Figures
Figure 1: Nonlinear and Linear Jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

Figure 2: Hiring earnings distribution by gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Figure 3: Application and selection rate by gender and AKM firm effect deciles. . . . . . . . . 30
Figure 4: The share of male applicants and flexibility requirements . . . . . . . . . . . . . . . . . . . . . 33
Figure 5: The gender hiring earnings gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Figure 6: Coefficients for categories of share of male applicants, male hires . . . . . . . . . . . . 36
Figure 7: Coefficients for categories of share of male applicants, female hires. . . . . . . . . . 37
Figure 8: AKM Person effects and the gender distribution of the application pool . . . . . . 39
Figure 9: Mothers and Women without Children . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Figure A.1: Share of male applicants: categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Figure B.1: GWG estimates by 5 year cohorts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Figure B.2: Alternative Selection Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Figure B.3: Residualized share of male applicants over grid of AKM firm effect deciles
by task complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Figure B.4: Residualized selection rates over grid of AKM firm effect deciles by job level. 53
Figure B.5: Application and selection rate by gender and AKM firm effect deciles (es-
timated from males only) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Figure B.6: Application and selection rate by gender and AKM firm effect deciles (es-
timated from females only) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Figure B.7: Application and selection rate by gender and AKM firm effect deciles, with
organized bargaining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Figure B.8: Application and selection rate by gender and AKM firm effect deciles, with-
out organized bargaining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Figure B.9: Application and selection rate by gender and AKM firm effect deciles. . . . . . . . . 58
Figure B.10: The share of male applicants and flexibility requirements . . . . . . . . . . . . . . . . . . . . . 59
Figure B.11: The gender hiring earnings gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Figure B.12: Coefficients for categories of share of male applicants, male hires . . . . . . . . . . . . 61
Figure B.13: Coefficients for categories of share of male applicants, female hires . . . . . . . . . . 61

Figure B.14: AKM Person effects and the gender distribution of the application pool . . . . . . 62
Figure B.15: Mothers and Childless Females . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
List of Tables
Table 1: Main variables by gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

Table 2: Share of male/female hires and applicants across occupations . . . . . . . . . . . . . . . . . . 28
Table 3: Gender earnings gap for mothers and childless females . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Table A.1: Share of male/female hires and applicants across industries . . . . . . . . . . . . . . . . . . . . 48
Table B.1: Gender earnings gap for mothers and childless females . . . . . . . . . . . . . . . . . . . . . . . . . . 62

Imprint
IAB-Discussion Paper 22|2022
Publication Date
18 August 2022
Publisher
Institute for Employment Research
of the Federal Employment Agency
Regensburger Straße 104
90478 Nürnberg
Germany
Rights of use
This publication is published under the following Creative Commons licence: Attribution -
ShareAlike 4.0 International (CC BY-SA 4.0)
https://creativecommons.org/licenses/by-sa/4.0/deed.de
Download
https://doku.iab.de/discussionpapers/2022/dp2222.pdf
All publications in the series “IAB-Discusssion Paper” can be downloaded from

https://www.iab.de/en/publikationen/discussionpaper.aspx
Website
www.iab.de/en
ISSN
2195-2663
DOI
https://doi.org/10.48720/IAB.DP.2222
Corresponding author
Dr. Benjamin Lochner
+49 911 179 6564
E-Mail [email protected]

DP 2222

Uploaded by

Copyright:

Available Formats

DP 2222

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DP 2222

Uploaded by

Copyright:

Available Formats

IAB-DISCUSSION PAPER

Articles on labour market issues

22|2022 Gender-Speciﬁc Application Behavior,

Benjamin Lochner, Christian Merkl

Benjamin Lochner (Institute for Employment Research (IAB) and

A.1 Descriptive Details 48

A.2 Additional Empirical Results 50

IAB-Discussion Paper 22|2022 3

A.3 Alternative Sample Restriction 58

IAB-Discussion Paper 22|2022 4

IAB-Discussion Paper 22|2022 5

IAB-Discussion Paper 22|2022 6

After several decades of gender convergence, substantial differences in earnings between

IAB-Discussion Paper 22|2022 7

IAB-Discussion Paper 22|2022 8

IAB-Discussion Paper 22|2022 9

IAB-Discussion Paper 22|2022 10

IAB-Discussion Paper 22|2022 11

IAB-Discussion Paper 22|2022 12

IAB-Discussion Paper 22|2022 13

2.1.1 Application Decision

Eηp,j w̄ (ε̃p,j ) − ξj > ep,j . (1)

IAB-Discussion Paper 22|2022 14

ẽp,j = Eηp,j w̄ (ε̃p,j ) − ξj . (2)

2.1.2 Hiring Decision

Πp,j = yp,j − εp,j − tp,j − ξj > 0. (4)

IAB-Discussion Paper 22|2022 15

yp,j − w (Πp,j ) − εp,j − tp,j ≥ 0. (6)

ε̃p,j = yp,j − tp,j − ξj . (7)

2.1.3 Wage Formation

IAB-Discussion Paper 22|2022 16

Λ = (w (yp,j , εp,j , ξj ) − ξj )α (yp,j − w (yp,j , εp,j , ξj ) − εp,j − tp,j )1−α , (9)

where α is workers’ bargaining power.

This yields the following wage:

w (yp,j , εp,j , ξj ) = α (yp,j − εp,j − tp,j ) + (1 − α) ξj . (10)

IAB-Discussion Paper 22|2022 17

ynl,j = λj anl if λj > λ∗ (12)

ynl,j = λj anl (1 − δ) if λj ≤ λ∗ (13)

In addition, there are other jobs where the output is linear, l:

IAB-Discussion Paper 22|2022 18

2.2 Model Implications

IAB-Discussion Paper 22|2022 19

ε̃p,f = yp,f − tp,f − ξf . (15)

2.2.2 (Non)Linear Production Functions and Sorting

IAB-Discussion Paper 22|2022 20

Under these assumptions, we obtain four different cutoff points:

ε̃nl,1 = λ1 anl − ξ1 , (17)

ε̃nl,2 = λ2 (1 − δ) anl − ξ2 , (19)

And under our assumptions, the following ranking holds:

ε̃nl,1 > ε̃l,1 , (21)

ε̃l,2 > ε̃nl,2 . (22)

ηnl,1 > ηl,1 , (23)

ηl,2 > ηnl,2 . (24)

IAB-Discussion Paper 22|2022 21

ηnl,1 > ηl,1 = 0 (25)

ηl,2 > ηnl,2 = 0 (26)

IAB-Discussion Paper 22|2022 22

IAB-Discussion Paper 22|2022 23

3.1 Data Sources