The College Melting Pot:

Peers, Culture and Women’s Job Search

Federica Meluzzi

September 30, 2024

Abstract
Gender norms are widely recognized as key determinants of persistent gender gaps in the labor
market. However, our understanding of the drivers of gender norms, and their implications for prefer-
ences, remain lacking. This paper addresses this gap by examining how cultural assimilation from
college peers influences women’s early-career labor market decisions. For identification of causal
effects, I exploit cross-cohort idiosyncratic variation in peers’ geographical origins within Master’s
programs, combined with unique administrative and survey data covering the universe of college stu-
dents in Italy. The main finding is that exposure to female classmates originating from areas with more
egalitarian gender culture significantly increases women’s labor supply, primarily through increased
uptake of full-time jobs. A one standard deviation increase in peers’ culture increases female earnings
by 3.7%. The estimated peer effects are economically significant, representing more than a third of the
gender earnings gap. Drawing on comprehensive data on students’ job-search preferences and newly
collected data on their beliefs, I shed novel light on two distinct mechanisms driving peer influence:
(i) shifts in preferences for non-pecuniary job attributes, and (ii) social learning, particularly on the
characteristics of the job offer distribution.

JEL classification: J31, J16, J22, R0, Z13.

Keywords: gender gaps, female labor supply, job search, peer effects, biased beliefs, gender norms.

I am grateful to Pierre Cahuc and Arne Uhlendorff for their invaluable guidance, and to Alberto Bisin, Patrick Kline, Camille Landais, Yves Le
Yaouanq, Roland Rathelot, Pauline Rossi and Michela Tincani for their support and suggestions at different stages. For very helpful comments, I
would like to thank Yann Bramoullé, David Card, Pauline Carry, Patricia Cortes, Gordon Dahl, Stefano Della Vigna, Xavier D’Haultfoeuille,
Matthias Doepke, Christina Felfe, Antoine Ferey, Raquel Fernandez, Albrecht Glitz, Paola Giuliano, Claudia Goldin, Libertad Gonzalez, Hilary
Hoynes, Matthew Jackson, Xavier Jaravel, Andreas Kotsadam, Francis Kramarz, Alan Manning, Arnaud Maurel, Guy Michaels, Tatiana Mocanu,
José Montalbán Castilla, Paula Onuchic, Ricardo Perez-Truglia, Barbara Petrongolo, Anna Raute, Alessandro Riboni, Nina Roussille, Nicolas
Salamanca, Kjell G. Salvanes, Francesco Trebbi, Danny Yagan, Alessandra Voena, Christopher Walters, Basit Zafar and Yves Zenou. I thank
audiences at UC Berkeley Labor Lunch Seminar, UC Berkeley IRLE Seminar, CREST Microeconomics Seminar, LSE Labour WiP Seminar,
LSE Public WiP Seminar, Insead/Collège de France Brown Bag Seminar, AMSE PhD Seminar, CEPR Paris Symposium, Cesifo Labor Area
Conference, SOLE Annual Meeting, EALE Annual Conference, IZA Summer School, CEPR-PSE Policy Forum, QMUL PhD Workshop, SEHO
Annual Meeting and Search and Matching Workshop. I am very grateful to the AlmaLaurea consortium, and in particular to Silvia Ghiselli, for
enabling data access. I gratefully acknowledge financial support from the CREST Economics Department, Fondazione Luigi Einaud i, Chaire
sécurisation des parcours professionnels. I thank Alessandra Casarico, Pamela Campa and Paola Profeta for sharing with me data on Italian
provinces from Indagine Excelsior and from the World Value Survey.
CREST, Institut Polytechnique de Paris. E-mail: [email protected]
 Intorno a Napoli è così, tutti lo sanno
(anonymous participant to interview)

1 Introduction

Cultural norms are ubiquitous and shape payoffs from many individual decisions. One
critical area where their influence is particularly strong is in the economic decisions of
men and women. By shaping the beliefs and preferences of both genders, gender norms
are recognized as key determinants of the persistent gender gaps in the labor market, over
and above traditional economic factors such as human capital accumulation, comparative
advantage and discrimination (Cortes and Pan 2023, Bertrand 2020). Prior research has
highlighted that their stickiness over time has prevented gender convergence in the labor
market (Kleven 2024, Fernandez 2013, Fortin 2015).
Understanding the determinants of cultural change is therefore a significant yet insuffi-
ciently understood problem. One hypothesis is that culture evolves through social learning
(Fernandez 2013 and Fogli and Veldkamp 2011). Despite the popularity of these theories,
the empirical evidence is scarce. This is primarily due to the scarcity of natural experi-
ments and of data sources that allow to study empirically how gender norms are formed
and transmitted. In this paper, I address this gap by exploring cultural assimilation from
college classmates. Using comprehensive administrative and survey data that covers 93%
of all college students in Italy, I provide the first quasi-experimental evidence on the effects
of peers’ gender culture on women’s early-career choices. By leveraging comprehensive
data on students’ valuations of job attributes and newly collected data on students’ beliefs
over time, I identify two specific mechanisms through which college peers shape women’s
outcomes.
Owing to features of this setting, university degrees can be thought to reproduce a
melting pot, where students born in places with very different gender culture mix together
within the same programs. Indeed, Italy is a salient example of a country with remarkable
spatial differences in gender culture, as reflected in a variety of indicators, as self-reported
gender attitudes, employment patterns and the magnitudes of child penalties (Campa,
Casarico, and Profeta 2011, Casarico and Lattanzio 2023, Carlana 2019). The magnitude
of these geographical differences is comparable to that of wide cross-country differences.

2
For instance, the share of women (15-64) participating in the labor force ranges from
29% to 67% across provinces (NUTS 3 classification), and the share of firms reporting
hiring preferences for male workers ranges between 46% to 84%. A key feature of this
setting is that a high share of students - around 57% - migrate outside their province of
origin to attend university. This allows the cultural composition of degrees to be very
heterogeneous: in the median degree, half of the students are coming from above-median
FLFP areas and the other half from below-median FLFP areas. Such a setting is therefore
ideal to study how culture evolves when individuals migrate, and whether peers shape
its evolution.
I start by establishing that the childhood environment in which a woman is raised has
a lasting influence on her labor supply decisions at the start of her career—a phenomenon
I refer to as cultural persistence. I substantiate this claim relying on the epidemiological
approach of Fernandez 2007 and re-adapting it to analyze granular within-country vari-
ations in cultural norms through the behavior of movers—individuals who work in a
different province from where they were born1. The effect of early cultural exposure is
estimated based on the relationship between the labor supply decisions of movers and
several measures of gender culture in their place of birth. These measures include standard
indicators, as female labor force participation (FLFP) relative to men across age groups,
as well as indicators of firms’ gender culture and local role models of previous cohorts
of female graduates, drawn from the behavior of stayers in the sample. The findings
reveal that women who move from more gender-egalitarian areas—e.g. reflected in higher
FLFP—have significantly higher labor supply compared to similar peers from less egali-
tarian provinces. This pattern holds even when controlling for factors such as working
in the same local labor market and graduating from the same Master’s program with
similar academic performance. The observed difference is both statistically significant and
economically meaningful, translating into a 7.6% increase in weekly hours worked and a
2.2 percentage point higher likelihood of full-time employment. Leveraging detailed infor-
mation on individual characteristics before entering the Master’s program (e.g., academic
records, family background), I show that this relationship is unlikely driven by differential
selection of movers from different areas. Moreover, the role of local cultural exposure is
1. This approach is similar to recent studies by Kleven 2024, Kerwin, Guryan, and Pan 2024 and Boelmann,
Raute, and Schönberg 2023.

3
both quantitatively larger and not confounded by maternal role models.
What happens to women’s labor market decisions when they get exposed to college
classmates with different gender norms? This paper explores the transmission of gender
norms during college years and their impact on women’s labor market choices. For identi-
fication of causal effects, I exploit idiosyncratic within-Master, between-cohort variation
in students’ geographical origins. The analysis draws from data covering 1,572 two-year
Master’s programs across 71 universities and all fields of study, spanning cohorts enrolled
between 2012 and 2016. This approach, first proposed by Hoxby 2000, rests on the assump-
tion that there exists some variability in the composition of peer groups across adjacent
cohorts within a degree program, which is beyond the control of individual students. I
bolster the validity of this design through a large battery of randomization checks. Im-
portantly, I provide evidence that cross-cohort changes in peers’ cultural composition are
unrelated to pre-determined covariates (ability, family background, socio-demographics)
that explain changes in students’ labor market outcomes.
My main finding is that exposure to peers from provinces with more egalitarian gender
culture increases women’s labor supply along the intensive margin, leading to greater
take-up of full-time jobs and longer weekly working hours. The magnitude of this effect is
large: a one standard deviation increase in the culture of female peers (8.33 pps.) leads
to a 3.3% increase in their weekly hours and in a 1.9 percentage points increase in the
likelihood of fulltime employment one year after graduation, translating to a 3.7% increase
in monthly earnings. These estimated peer effects are economically significant, comparing
to 33% − 41% of the size of the gender differences in the same outcomes. Notably, the
effects are gender-specific, with no influence from male peers. As a placebo, I conduct the
same analysis on male students and find that peers do not affect men’s earnings, labor
supply, or other job characteristics.
Identifying the precise mechanisms through which peer effects operate has long been
a challenge, primarily due to data limitations (Sacerdote 2011). As clearly stated by Barrios
Fernandez 2023, understanding these mechanisms is crucial, as this would allow to design
policies that could replicate the benefits of peer exposure without altering peer group
composition. With unique comprehensive data on students’ valuations of job attributes
and original survey data on students’ beliefs, this paper advances our understanding by

4
identifying two main mechanisms driving peer effects. First, I show that peer effects are
not mediated by changes in academic performance or geographic mobility, allowing me
to rule out local labor market networks as a key factor. Instead, I uncover two channels
of peer influence: (i) shifts in preferences for non-pecuniary job attributes and (ii) social
learning, likely related to the job offer distribution. Using data from a mandatory pre-
graduation survey, I show that exposure to peers from more gender-egalitarian areas
changes women’s preferences, reducing the importance they place on non-pecuniary job
factors such as flexible hours and social utility—consistent with cultural transmission.
Moreover, I find strong asymmetry in peer effects: women from areas with below-median
FLFP are positively influenced by exposure to egalitarian peers, while the reverse does
not occur. This asymmetry, which starkly reject conformism as an underlying mechanism,
aligns with social learning explanations.
To investigate social learning and identify which beliefs are affected, I have designed
an original survey and collected students’ beliefs at two points in time. This includes
beliefs about gender roles, perceptions of employers’ discrimination, beliefs regarding
the distribution of job offers, and expectations of future fertility and child penalties. The
survey also gathers information on the network structure and perceived peer influence. It
was administered to a random sample of current female students across all disciplines at
one large university. With in-person administration and lottery incentives, I achieved a
97% response rate among attending students. I use this data to (1) examine asymmetries in
beliefs based on the gender culture in their province of birth, (2) test whether these beliefs
predict job acceptance decisions, and (3) analyze how beliefs evolve within the social
environment. Analysis of these data reveal that women from low-FLFP areas have more
pessimistic expectations on the arrival rates of job offers, and expect a significantly share
of part-time job offers relative to full-time ones compared to their peers from high-FLFP
areas. Consistent with the predictions of a McCall type model (McCall 1970) incorporating
heterogeneous workers’ beliefs, my results show that students’ beliefs are relevant in the
decision to accept part-time jobs, explaining about one fifth of the part-time acceptance
gap between women from high- and low-FLFP areas. Lastly, I provide evidence of belief
convergence after peer exposure, driven primarily by strong beliefs’ updating concentrated
among women from low-FLFP areas.

5
 Related literature. This article contributes to several strands of literature. First, it
relates to a burgeoning literature that, since the seminal contribution of Fernandez 2007,
has explored the interplay between cultural norms and women’s economic decisions.
Several works have highlighted the role of cultural factors in shaping women’s labor
supply (Fortin 2015, Blau, Kahn, and Papps 2011, Bertrand, Kamenica, and Pan 2015), the
magnitude of child penalties (Kleven 2024, Cortés, Kos¸ar, Pan, and Zafar 2022, Boelman,
Raute, and Schönberg, Forthcoming), fertility choices (Alesina, Giuliano, and Nunn 2013,
Fernandez and Fogli 2006), marriage prospects (Bertrand et al. 2021) and time allocation
(Ichino et al. 2024). In many of these studies, the impact of culture is estimated by com-
paring outcomes among first- or second-generation immigrants within a host country,
with cultural differences stemming from cross-country variations in norms. These works
typically focus on the broader female population and how cultural influences shape
(work) decisions around motherhood or marriage. My contribution to this literature is
twofold. First, I focus on a narrower segment—young, educated women—and exploit
granular within-country variations in gender norms, in a setting where I can rule out
many confounders, such as ability and family background. My findings complement ex-
isting research by showing that childhood exposure to gender norms influences women’s
labor supply decisions at the very start of their careers, which may play a crucial role
in explaining the higher child penalties observed later in life. Second, while existing re-
search often remains agnostic about the mechanisms behind cultural persistence, I uncover
a previously unexplored channel: information asymmetries. With this finding, I extend
the conventional view of cultural persistence beyond its influence on preferences and
attitudes, and show that childhood exposure to local role models also shapes women’s
beliefs about labor demand and employers’ tastes. This finding carries important policy
implications, suggesting that information provision, e.g. on the job offer distribution, can
be an effective tool in reducing cultural persistence and gender disparities.
A second strand focuses on the transmission of gender norms. Theories developed by
Fernandez 2013 and Fogli and Veldkamp 2011 emphasize social learning as a key driver
of cultural evolution. They model a process in which beliefs are shaped both within the
family—via mothers’ behavior (vertical transmission, in the terminology of Cavalli-Sforza
and Feldman 1981)—and through observing the labor force participation of previous

6
generations of women (oblique transmission). Consequently, current shocks to female
labor supply trigger the dynamics of beliefs’ updating. Empirical evidence supporting
the importance of childhood exposure has been provided by Olivetti, Patacchini, and
Zenou 2020 and Mertz, Ronchi, and Salvestrini 2024, who show that women’s career
decisions, such as whether to work after motherhood or their occupational choices, are
influenced not only by their own mothers’ behavior but also by the behavior of other
mothers in their social network, such as their peers from primary or high-school. An
emerging and directly related literature examines the horizontal transmission of norms.
Boelmann, Raute, and Schönberg 2023 and Jessen, Schmitz, and Weinhardt 2024 show that
increased exposure to East Germans, either through coworkers or friendships, influences
West Germans’ behavior by reducing their parental leave or increasing labor supply at
the intensive margin. Similarly, Maurin and Moschion 2009 find that the labor supply
decisions of female neighbors positively affect other women’s labor participation at the
extensive margin in France. My contributions to this literature is twofold. To the best of
my knowledge, my paper provides the first large-scale empirical evidence on the role of
college peers in the transmission of gender norms. Specifically, I exploit quasi-random
variation in the gender norms to which young, educated women are exposed during
college, allowing me to directly compare the influence of horizontal peer effects with
childhood exposure. My findings reveal that the social environment in college plays a
critical role in shaping women’s early-career labor supply decisions, demonstrating that
peers can nearly offset the impact of limited female role models during childhood. This
challenges the view that gender norms are primarily formed in early childhood and
remain sticky over time. Second, I offer direct evidence on the mechanisms behind peer
influence. I show that peers shape women’s preferences for job attributes, using data from
a compulsory survey of all graduates. Newly collected survey data on students’ beliefs
reveal that initial information gaps about job opportunities between women from different
cultural backgrounds narrow substantially, supporting social learning explanations. With
the latter finding, I closely relate to a contemporaneous literature on the role of biases in
beliefs and the stickiness of gender norms (Cortés, Kos¸ar, Pan, and Basit 2022, Bursztyn
et al. 2023).
This paper also contributes to a broad body of work on gender gaps in the labor

7
market. Recent evidence has shown that, in the skilled population, gender differences in
the valuation of temporal flexibility, coupled with increasing returns to the provision of
long hours, largely contribute to earnings inequalities (Cortes and Pan 2019, Wiswall and
Zafar 2018, Blau and Kahn 2017, Azmat and Ferrer 2017, Flabbi and Moro 2012, Bertrand,
Goldin, and Katz 2010). Aligning with previous work, I document large differences in
hours worked and earnings emerge between female and male graduates at labor market
entry. My findings suggest that preferences for job attributes are endogeneous to the social
environment and can explain part of early-career gaps. Specifically, I show that 30% of the
initial gap can be closed through peer influence. Finally, my paper contributes to a rising
literature that has focused on the interplay between biased beliefs, e.g. on job finding
probabilities or on the wage distribution, and job-search behavior (Jäger et al. 2024, Cortes
et al. 2023, Mueller and Spinnewijn 2023, Conlon et al. 2018, Mueller, Spinnewijn, and
Topa 2021, Alfonsi, Namubiru, and Spaziani 2024, Bandiera et al. 2023, Caliendo, Cobb-
Clark, and Uhlendorff 2015). Relative to previous studies, usually focussed on long-term
unemployed individuals, I examine the effects of pessimism on arrival rates of job offers
and workers’ acceptances of part-time jobs.
The rest of the article is organized as follows. Section 2 describes the institutional
setting and the data sources and provides a description of the sample. Section 3 describes
the college melting pot. Section 4 describes the early-career gender earnings gap and
the phenomenon of cultural persistence. Section 5 explains the identification strategy
and discusses its validity. Section 6 presents baseline estimates of peer effects. Section
7 discusses estimates from a battery of robustness exercises. Section 8 explores non-
linearities in peer effects. Section 9 presents evidence on the mechanisms of peer influence.
Section 10 analyzes original data on students’ beliefs and presents a job search model to
illustrate their implications for job search.

2 Institutional Background and Data

2.1 Admission and Structure of Graduate Education in Italy

This subsection provides an overview of the key institutional features governing tertiary
education in Italy. Section 3 will then zoom in on the specific aspects that make this context

8
particularly well-suited for studying cultural assimilation.
Admission to Master’s degrees. Since the early 2000s, degrees are organized as bachelor’s
(three years) and master’s (two years) 2. Most students pursue a two-year master’s pro-
gram after completing their bachelor’s degree. Admission criteria for master’s programs
are determined autonomously by each academic institution. However, certain fields—such
as medicine, health sciences, architecture, psychology, and primary education— are sub-
ject to selective national entry exams under Law 264/1999. A common requirement for
master’s admission is the completion of a bachelor’s degree, along with specific curricular
prerequisites, typically fulfilled through credits in required courses. This system allows
students some flexibility to switch fields between their bachelor’s and master’s studies,
provided they meet the relevant eligibility criteria. Therefore, students have the flexibility
to change fields between the bachelor’s and master’s levels, subject to meeting the eligi-
bility requirements3. Beyond these prerequisites, many programs also include selective
entry exams, bachelor’s grade requirements, and interviews in their admission process.
Admission is often competitive, with applicants ranked based on entry exam scores or
bachelor’s GPA, and additional requirements like English proficiency, motivation letters,
or reference letters may apply.

Tuition fees. Ninety percent of Italian students attend public universities (ISTAT 2016),
with tuition fees varying based on the degree, institution, and family income. Regional
governments set income thresholds for need-based grants, which cover tuition, housing,
and meal vouchers for low-income students (Rattini 2022). On average, 23% of the stu-
dents in my sample receive such grants. For those not eligible, the average annual tuition
fee is €1262 (Commission/EACEA/Eurydice 2016).

2.2 Data Sources and Sample

The empirical analysis relies on two main sources of data. The primary source comprises
comprehensive administrative and survey data, which collectively cover 93% of the uni-
2. Italy adheres to the Bologna process (1999) that ensures comparability in higher education standards
across the European Higher Education Area (EHEA), which comprises 48 European and Central Asian
countries.
3. On average, students must complete 77 constrained credits to qualify for a master’s program, though
requirements vary across fields. For instance, a student entering a master’s program in economics must have
earned at least 53 credits in economics, statistics, or other social sciences (Brandimarti 2023).

9
verse of college students in Italy, obtained from the AlmaLaurea consortium. Specifically,
the dataset encompasses all students enrolled in 1,572 2-year Master’s degree programs
across 71 universities, spanning enrollment cohorts between 2012 and 2016. This database
consists of administrative data from university records, institutional survey data, and
post-graduation follow-up surveys.

1. Administrative student-level information, from university records, for all students.

This source provides information on academic performance during the Master (num-
ber of exams, GPA, final grade), and demographic information (age, immigration
status, municipality of birth and residence), as well as unique identifiers of Master’s
programs within universities and enrollment and graduation dates. Importantly, I
use this source to identify college classmates and construct measures of their gender
culture based on their birth province. Due to the administrative nature of the data,
all information is available for the entire student population, ensuring that I observe
the characteristics of all peers.

2. Institutional pre-graduation survey. Universities administer this survey to all stu-

dents as part of the graduation process. At the end of their final year, students are
required to complete a compulsory survey, with a response rate close to 100%. This
survey collects detailed information on students’ job search intentions and prefer-
ences, including their valuation of several job attributes. Additionally, the survey
gathers data on students’ socio-economic background, including parents’ occupa-
tions and education levels. Furthermore, it collects detailed information on students’
educational histories, such as previous education in high school and Bachelor’s
degree programs, their grades in previous education, and the working activities they
participate in during their studies.

3. Follow-up surveys. Students are contacted by the AlmaLaurea consortium for

follow-up surveys one, three, and five years after graduation. These surveys gather
comprehensive information on realized job characteristics, such as net monthly
earnings, usual weekly hours worked, contract type (part-time vs. full-time), job
security, occupation, industry, sector and location. The responses to the earnings
questions, and usual weekly hours worked, are collected in discrete bins that I

10
 transform into real-valued variables (at the mid-point of each bin)4. Additionally,
they include retrospective information on the job-search process and current job
search activities. While participation in these surveys is voluntary and does not
involve monetary incentives, the response rate remains high (e.g., 74% after one
year).

Original survey on students’ beliefs. To investigate the mechanisms of peer influence,

I have designed an original survey to elicit students’ beliefs regarding gender attitudes
and various future outcomes. This includes perceptions of employers’ discrimination,
beliefs regarding the distribution of job offers, and expectations of future fertility and child
penalties. The survey also gathers information on the network structure and perceived
peer influence. It has been administered to a random sample of students from several
fields at a large university (enrollment cohorts 2022-2023). With in-person administration
and lottery incentives, I achieved a nearly 100% response rate among attending students.
Detailed information on the survey and elicitation methods is provided in Section 10.

2.3 Sample description

In this paper, I use data on students enrolled between 2012 and 2016. Since data are
collected from graduating cohorts, I recostruct enrollment cohorts using students’ enroll-
ment and graduation dates from university records. A student’s classmates, or peers, are
defined as all students who enroll in the same university major, or degree, in the same
cohort, and who remain enrolled for the entire duration of the Master 5. The sample is
composed of students from a panel of Master degrees that (i) count at least one man and
one woman in the same cohort (excludes 3.55% of students) and (ii) exist and respect (i)
for at least 2 consecutive years (excludes 6% of students). The final sample is composed of
316,470 students from 1,572 degrees and 71 universities. Table A.1 and Table A.2 provide
descriptive statistics of the analysis sample. The last column of the tables reports the

4. Possible answers to the earnings question were < €250, €250–€500, €500–€750, €750–€1000, €1000–€1250,
€1250–€1500, €1500–€1750, €1750–€2000, €2000–€2250, €2250–€2500, €2500–€3000, and > €3000. I converted
the answers into a real-valued earnings variable at the mid-point of each earnings bin; I assigned earnings of
€187,5 to those that responded earning less than €250 and earnings of €3750 to those who indicated earning
more than €3000. The response bins for usual weekly hours worked were < 5 hours, 5-9 hours, 10-14 hours,
15-19 hours, 20-24 hours, 25-29 hours, 30-34 hours, . . . , 55-59 hours, and > 60 hours.
5. A drawback is that I lose track of dropouts, which account for 6% of the enrolled students between 2012
and 2016 (ANVUR 2023).

11
p-value of the test of equality of the means across gender.

Background characteristics and academic records. Table A.1 presents summary statistics
on the main background characteristics and academic variables in the sample, disaggre-
gated by student gender. These data are drawn from both administrative records and the
institutional survey. The sample includes 182,792 women and 133,678 men. On average,
women outperform men academically, as evidenced by higher GPAs and final grades dur-
ing their master’s studies, as well as stronger prior academic records, such as bachelor’s
and high school grades. Additionally, the variation in these outcomes is smaller among
women compared to men. Women are also more likely to have attended a general high
school (liceo) than men—84.1% versus 71.3%. In terms of field specialization, women are
underrepresented in scientific tracks and more concentrated in humanities, both in high
school and at university. The largest disparities are in engineering and humanities: 27%
of men study engineering, compared to just 8.2% of women, while 24.7% of women and
only 10.4% of men pursue humanities. Regarding family background, around one-fifth of
students have parents with tertiary education, and roughly one-third come from families
where the father is in a high-SES profession. Additionally, 71% to 73% of students have
mothers in the labor force. Women, however, are less likely to come from wealthier families
compared to men, as indicated by a lower share of parents with tertiary education and a
higher share of parents in low-SES occupations.

Labor market outcomes. Table A.2 presents summary statistics on students’ labor market
outcomes, based on responses from the follow-up survey conducted one year after gradua-
tion. Response rates to this survey are 73.7% and 73.2% in the samples of women and men,
respectively, very high compared to traditional survey data. At the time of the survey,
about 12% of both female and male students are pursuing further education, either at the
master’s or PhD level, while the majority are participating in the labor market. 67.3% of
women and 71.8% of men are employed, either with a standard labor contract or via an
internship. A slightly higher share of women (20.5%) than men (15.1%) are unemployed
and actively searching for a job, while less than 1% of both genders are unemployed
and not looking for work. The observed gender differences in employment patterns can

12
largely be attributed to differences in field specialization, as discussed in Section 4. Among
employed graduates, women experience less favorable outcomes than men. On average,
women earn €1,077.8 per month, compared to €1,324 for men, and are significantly less
likely to hold full-time positions (69.3% vs. 86.2%). Women are also underrepresented
in occupations and industries with above-median earnings or above-median shares of
full-time jobs. A higher proportion of women work in the public sector (16.5% vs. 11% for
men). Regarding the job-search process, around 80% of both men and women are in their
first job after graduation. While men and women begin their job search at similar times,
women tend to search longer before accepting their first job offer.6

3 The College Melting Pot

This section outlines three key features of the Italian context that make it uniquely suited
for studying cultural assimilation from peers on a large scale.

3.1 Spatial Differences in Gender Culture

Studying cultural assimilation requires significant variation in gender norms among stu-
dents. Italy provides an ideal setting, offering granular yet wide geographical variations
in gender culture. Spatial differences in gender culture are striking, comparable in mag-
nitude to wide cross-country differences. These differences are reflected in a variety of
indicators, from self-reported gender attitudes to labor market attachment, as documented
in previous studies (Campa, Casarico, and Profeta 2011, Carlana 2019, Casarico and Lat-
tanzio 2023, Carrer and Masi 2024). A particularly important feature is the substantial
geographical variation in traditional labor market outcomes for women. For example,
female labor force participation rates among those aged 15-64 range from 29% to 67%
across provinces, and for women aged 25-34, the range is even wider, from 38% to 86%. In
contrast, male labor force participation varies much less, ranging from 64% to 82% for the
15-64 age group. These differences in female labor market outcomes are accompanied by
substantial heterogeneity in gender attitudes. For instance, the proportion of individuals
6. This variable takes the value -1 if a student started searching for a job before graduation, 0 if they began
searching at the time of graduation, and 1, 2, 3, etc., if they started their search 1, 2, 3, etc., months after
graduation.

13
who disagree with statements such as "Being a housewife is just as fulfilling as working for
pay" or "Men should be given priority when jobs are scarce" varies from 16% to 67% across
Italian regions (NUTS 2), according to recent waves of the European Values Survey (EVS
1990-2008). Throughout this paper, provinces (the NUTS-3 classification) are used as the
main geographical unit. Italy is partitioned into 103 provinces, which are administrative
divisions of intermediate level between a municipalities and regions 7.
I define the gender culture of a province using several alternative measures:

1. Female labor force participation (FLFP hereafter) of different age groups at the
province level (NUTS 3 level);

2. Ratio of female to male labor force participation (FLFP/MLFP hereafter) of different

age groups at the province level (NUTS 3 level);

3. Indicator of firms’ gender culture, e.g. the share of firms in the private sector without
hiring preferences for male workers, at the province level (NUTS 3 level);

4. Share of female college graduates that are employed full-time at the outset of the
career at the region level (NUTS 2);

5. Share of female vs. male college graduates that are employed full-time at the outset
of the career at the region level (NUTS 2);

Students are assigned to provinces based on their province of residence at the enroll-
ment date, as recorded in university registers. Such province should be interpreted as
the place where the student grew up. Since the objective is to capture the gender culture
and female role models that students were exposed to while growing up, most of these
measures are based on data prior to their university enrollment, particularly during their
adolescence. The first two measures refer to averages of FLFP and FLFP/MLFP from 2004
to 2007. Indicators of firms’ gender culture in (c) are constructed based on answers to
a survey of a nationally representative sample of 100,000 Italian firms in 2003 (Indagine
7. The average population of a province was 551,000 as of 2010, but there is large heterogeneity. The largest
province, Rome, has over 4 million residents and contains 121 different municipalities. The smallest province,
Ogliastra (Sardinia), has less than 60,000 residents and only includes 23 municipalities.

14
Excelsior, Unioncamere8). Each year, firms report their hiring intentions, specifying whether
they prefer to hire male or female workers, or if they have no preference. At the province
level, I define firms’ gender culture as the proportion of firms that either prefer female
workers or are indifferent between male and female hires. Additionally, I introduce novel
measures related to the employment patterns of previous cohorts of female graduates,
who serve as a more direct reference group for the students in the sample. To construct
measures (d) and (e), I map local labor market opportunities for female and male college
graduates, focusing on graduates who remain employed in their province of birth. Using
labor market data from cohorts preceding those analyzed, I calculate the share of female
college graduates employed full-time one year after graduation, both in absolute terms
and relative to male graduates. Due to limited data in smaller provinces, these measures
are aggregated at the regional level (NUTS 2). Table 1 shows descriptive statistics for
these different indicators in the sample, obtained by assigning to each student the value
of the indicator corresponding to the student’s province (or region) of origin. Table A.3
disaggregates these statistics by gender, showing that there are no significant differences
in the geographical origins of female and male graduates. A visual illustration of the
spatial distribution of the FLFP and of the ratio of FLFP/MLFP, for all ages, is presented
in Figure 1. Similar spatial patterns emerge for other indicators of gender culture, such as
participation rates for young women, firms’ gender preferences, and individual gender
attitudes, as illustrated in Figure A.1. All of these measures reveal substantial variation
across provinces. For example, in some areas, the FLFP is as low as 29%, similar to that of
low-income countries, while in other areas, significantly higher shares of women are in the
labor force, aligning with or exceeding the OECD average. These disparities are similarly
pronounced among young women (aged 25-34), where FLFP ranges from 38% to 86%.
Importantly, these differences in labor market opportunities for women are not merely a
reflection of spatial heterogeneity in economic activity, as reflected in the nearly identical
variation in the FLFP/MLFP ratio, whose geographical distribution closely follows that
of FLFP. The ratio shows a wide range: between 43% and 86% of women of all ages, and
8. Starting from 1997, the Excelsior Survey represents one of the main sources of information on the Italian
labor market. The survey is administered by Unioncamere, in partnership with the Ministry of Labor, ANPAL
and the European Union. The survey is conducted on firms operating in the manufacturing and service
sectors with at least 0.5 employees/year. It excludes employers in the agricultural and public sector, as well
as those that are not registered with the Chambers of Commerce. The sample of firms represents roughly one
third of the total in the respective population.

15
 TABLE 1. Summary statistics of measures of gender culture in the sample

Variable Mean SD Min Max Obs

Female labor force participation (age: 15-64) 49.7 11.2 27.3 66.7 316470
Female/Male labor force participation (age: 15-64) 66.7 11.8 43.0 85.7 316470
Female labor force participation (age: 25-34) 65.1 15.2 38.0 86.0 316470
Female/Male labor force participation (age: 25-34) 74.3 13.0 47.0 95.0 316470
Male labor force participation (age: 15-64) 73.8 4.5 62.9 81.9 316470
Male labor force participation (age: 25-34) 86.7 6.5 71.4 97.1 316470
% of female graduates in full-time job 56.4 9.5 40.1 68.9 316470
% of female/male graduates in full-time job 71.8 6.7 55.3 83.5 316470
% of firms without hiring pref. for male workers 34.8 7.8 16.0 54.0 316470
Notes: The Table presents summary statistics f the measures of gender culture 1-5 presented in Section 3 The
unit of observation is a student. Students are assigned to provinces based on their residence province prior to
enrollment in the Master.

between 47% and 95% of young women, participate in the labor force relative to men in
the same age group. Additionally, the ratio of female to male college graduates in full-time
employment also varies considerably by region, from 55% to 83%. Looking at firms’ stated
preferences, between 16% and 54% of firms report having no strict preference for male
workers.

3.2 Students’ Mobility

Within this context, a second key aspect is the high mobility of students who study outside
their home province. Due to historical and institutional factors, the majority of students in
Italy relocate to attend university (ANVUR 2023). This phenomenon has longstanding
origins and has been relatively stable over time. During the period covered by this analysis,
more than 57% of students moved to another province, and about 31% moved to a different
region to pursue higher education. To better understand the strong migration of students
away from their province of origin, it is useful to describe the institutional landscape in
Italy. In 2016, public universities, which account for over 90% of students, offered Master’s
degrees at 89 institutions. However, these universities were concentrated in only 52 of
Italy’s 103 provinces. Furthermore, not all fields of study are available at every institution,
as some universities specialize in specific disciplines. For instance, degrees in information

16
 FIGURE 1. Heatmaps of female labor force participation

(a) Female LFP (%) (b) Ratio of female to male LFP (%)
Notes. The maps present the FLFP (Panel a) and the FLFP/MLFP (Panel b) across provinces in Italy. Indi-
viduals between 15 and 64 years of age are considered. Each geographical partition is a province (NUTS
3 classification) and there are 103 provinces in total. Both measures are constructed as averages of years
2004-2007. Source: Labor Force Survey (Istat).

technology were offered at only 29 universities, and agriculture and veterinary sciences at
just 24 institutions. This uneven distribution of universities results in limited local options
for many students. About 20% of Italians aged 18-19 live in provinces without any higher
education institutions, and only about 77% have access to both STEM and non-STEM
universities in their home province (Braccioli et al. 2023). Moreover, students are free to
apply to any university in the country, regardless of their place of residence. This flexibility,
combined with the uneven availability of university programs and variation in the quality
of these programs, fuels the high levels of student mobility observed across Italy.

Mobility patterns in the sample. Table A.4 present summary statistics of students’ mo-
bility by gender. 58.9% of women 55.4% of men have moved outside their birth provice
for their studies. The table also describes the local gender culture in the provinces where

17
these students are studying. Overall, the mobility destinations for both genders do not
exhibit significant differences, with both groups studying in areas characterized by more
egalitarian gender cultures compared to their provinces of origin. Additionally, Table
A.5 provides summary statistics on mobility patterns to local labor markets for students
employed at the time of the follow-up survey. A large proportion of students work outside
their province of origin, with this share being higher for male students (51.4%, compared to
44% for women), indicating that women are more likely to return to their home provinces
after completing their studies. The majority of students—68.4% of women and 65.2% of
men—secure their first job in the same region where they studied, and approximately 5%
of both men and women move abroad for work. The locations where men and women
work show no significant differences in terms of local gender culture.
Since the focus of the paper will be on gender norms and women’s outcomes, Table
A.6 zooms in on the mobility patterns of women, by contrasting the behavior of those
originating from provinces in the first or the fourth quartile of FLFP. Both groups ex-
hibit high levels of migration outside their provinces of origin, with women from more
egalitarian regions migrating at a higher rate. Women from less gender-egalitarian areas,
however, are significantly more likely to migrate longer distances, such as outside their
home region. Among those who relocate, women from high-FLFP areas tend to move to
regions with higher FLFP levels compared to women from low-FLFP areas. This results
in unequal exposure to peers from high-FLFP regions: on average, women from high-
FLFP provinces attend programs where 67% of their peers are from above-median FLFP
areas, compared to only 27% for women from low-FLFP provinces. In terms of academic
specialization, women from low-FLFP areas are slightly more likely to pursue degrees in
scientific disciplines, engineering, and psychology, while being less likely to choose fields
such as humanities, economics, statistics, and architecture. However, these differences in
specialization are relatively modest in magnitude.

Peers’ composition. A direct consequence of these high mobility rates is that they make
degrees very heterogeneous in terms of students’ geographical origins. Panel (A) of Figure
2 plots the distribution of degrees by the share of students that have moved away from
another province. Less than 1% of programs have no movers. On the contrary, 50% of

18
degrees have more than 56% of students who are movers, as represented by the red line.
Panel (B) offers a visual snapshot of the "melting pot" within degree programs, mapping
each program’s cultural composition through a scatter plot. The y-axis represents the
share of students from provinces in the highest FLFP quartile, while the x-axis captures
those from below-median FLFP areas. Notably, about one-third of degree programs are
entirely filled with students from below-median FLFP provinces, clustered in the right tail.
However, a large portion of programs display significant diversity, with representation
from all geographical groups. A small minority—less than 5% of students—attend pro-
grams comprised solely of peers from above-median FLFP areas, as illustrated in Figure
A.2.
FIGURE 2. Geographical composition of students within degrees

(A) Mobility (B) Cultural background

Notes. Panel (A) represents degrees by the % of movers in 2016. One unit corresponds to a degree (N=1,572).
The red line corresponds to the median % of movers across degrees. In Panel (B), each dot corresponds to
a degree program (N=1,572). For each degree, the figure plots the share of students from provinces in the
highest quartile of FLFP (y-axis), alongside the share of students from provinces with FLFP in the first or
second quartiles (x-axis). Data refer to the enrollment cohort of 2016.

3.3 Size and Relevance of Peer Groups

A third crucial feature of this context makes it particularly well-suited for studying
peer effects. Unlike prior studies that define peer groups broadly as all students within
the same school cohort, I focus on a much narrower and possibly more relevant set:
students enrolled in the same Master’s degree cohort. The typical program is small, with
a median (mean) size of 34 (46) students, as shown in Figure 3 (Panel A), which plots

19
the distribution of programs’ size. This is in a range between 4 and 410. Smaller peer

FIGURE 3. Degree size and gender composition

(A) Size (B) Share of female students

Notes. Panel (A) represents the distribution of degrees by their size in 2016. The red line corresponds to
the median size. Panel (B) represents the distribution of degrees by the share of female students. The red
corresponds to the median share of female students. In both panels, one unit corresponds to a degree
(N=1,572).

groups offer several advantages when studying peer effects. First, in smaller peer groups,
it is very likely that students regularly interact with most of their peers and closely
observe their behaviors. Moreover, the long duration of exposure—at least two years
spent with the same cohort—intensifies these peer interactions. Section 10 provides direct
evidence on the nature and frequence of such interactions, through a survey to a sample of
prospective students. Second, smaller groups reduce the likelihood of endogenous sorting
or self-selection into subgroups, a mechanism often present in larger settings (Carrell,
Sacerdote, and West 2013). Third, the reduced size of degrees is crucial for identifying peer
effects through cross-cohort variations in characteristics. In smaller programs, even minor
idiosyncrasies within a cohort can lead to significant shifts in its composition. Conversely,
as program sizes increase, the law of large numbers indicates that cohort compositions will
tend to the average. I will discuss this point in greater detail in Section 5. Furthermore, note
that 50% of a degree’s course content and credit distribution is fixed, and that students are
free to allocate only about 10% of their total credits (Ministerial Decree 270/2004). This
structure ensures that students are consistently exposed to the same peers throughout
their studies. Additionally, Panel (B) plots the distribution of degrees according to the
share of female students, and descriptive statistics of degrees are presented in Table A.7.

20
4 Two Novel Facts About Early-Career Gender Gaps

4.1 Fact 1: the early-career gender earnings gap

Despite women achieving higher levels of human capital accumulation, as reflected in

their higher college attendance and GPA, they fare significantly worse than men at the
start of their careers. One year after graduation, equally productive women earn 11%
less than their male counterparts from the same Master’s program (Table 2)9. This gap
is both statistically significant and economically meaningful: it represents €1,795 every
year, on average. The earnings gap is primarily driven by differences in the intensive
margin of labor supply: women are 5 percentage points less likely to be employed in
full-time jobs and work 8% fewer hours per week compared to male students with similar
academic performance10. The residual gap in hourly wages, however, is much smaller at
2.9%. These differences in labor supply are not attributable to geographic mobility and
are only mildly related to differences in occupational and industry sorting by gender
(Table A.9). Furthermore, while the results are presented at the mean, gender gaps are
pervasive across types of programs and fields. This is illustrated in Figure 4, which plots
binned degree-specific effects on monthly earnings and full-time employment for women
against the corresponding effects for men. OLS estimates of the slope are presented, after
re-weighting each degree for the share of students it represents across all years. Although
there is a strong correlation between the average premiums for male and female students
within each degree, women across all degrees are systematically employed in jobs with
lower earnings (Panel A) and fewer weekly hours (Panel B). Moreover, these gaps remain
stable and do not shrink within the first five years of entering the labor market. Overall,
these findings provide new and comprehensive evidence of systematic gender differences
in earnings and labor supply at the start of the career among highly skilled individuals
in Italy. Previous research on gender gaps at labor market entry among highly skilled
individuals had mainly focused on the U.S. and narrower groups (such as graduates from
elite universities or specific fields). These studies found little or no gender differences in

9. This is consistent with estimates by Bovini, De Philippis, and Rizzica 2023, who use individual-level
administrative data for all university graduates in Italy, linking education records from the Ministry of
Education with social security data from 2011 to 2018.
10. Note, instead, that there are no differences in the extensive margin of labor supply (Table A.8).

21
 TABLE 2. The gender earnings gap at labor market entry

(1) (2) (3) (4)

Log(monthly earnings) Log(weekly hours) Pr(fulltime) Log(wage)

Female -0.113*** -0.083*** -0.051*** -0.029***

(0.004) (0.003) (0.003) (0.003)

GPA ✓ ✓ ✓ ✓
Degree FEs ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓
Observations 127,153 127,153 127,153 127,153
R-squared 0.294 0.259 0.293 0.089
Notes: The table reports coefficients from regressions of graduates’ labor market outcomes on a female
dummy, after including degree and cohort fixed effects and controlling for GPA. The sample consists of
female and male students who are employed one year post graduation. Standard errors are clustered at the
degree level.

working hours at labor market entry (Cortes et al. 2023) and documented that such differ-
ences typically become significant a few years after entering the labor market (Bertrand,
Goldin, and Katz 2010, Azmat and Ferrer 2017).
Fertility and couple decisions. Importantly, while the focus of a prominent body of lit-
erature has been on the gendered role of parenthood in explaining gender differences in
labor supply11, realized fertility can be ruled out as a major factor in this setting. In the
sample, the average age of women is 24, with only a small fraction having children or
being married or cohabiting (3.7% and 16.1%, respectively). Excluding these groups from
the analysis does not alter estimates of the gender earnings gap, as shown in Table A.10.
Moreover, analysis of newly collected expectations data from the the students’ survey
suggests that anticipated fertility is unlikely to be a major factor in women’s overrepresen-
tation in part-time jobs. On average, women in this sample expect to have their first child
at age 31—well after they have entered the labor market—and only a minority foresee
working part-time or exiting the workforce due to motherhood. A more comprehensive
examination of this potential channel will be provided in Section 10.
11. Examples include: Bertrand, Goldin, and Katz 2010, Kleven, Landais, and Leite-Mariante, Forthcoming,
Angelov, Johansson, and Lindahl 2016, Cortes and Pan 2023, Kleven 2024, Kleven, Landais, and Søgaard 2019,
Bertrand, Goldin, and Katz 2010, Azmat and Ferrer 2017, Casarico and Lattanzio 2023, Altonji and Blank
1999)

22
 FIGURE 4. Gender-specific degree effects on earnings and labor supply

(A) Net monthly earnings (B) Full-time employment

Notes: The figure shows binned scatter plots of estimated degree effects for female students against estimated
firm effects for male students. Each degree is characterized by the average earnings (or full-time employment)
of male students (x-axis) and the average earnings (or fulltime employment) of female students (y-axis), both
computed across all years. The slope is estimated across degrees by OLS, after re-weighting each degree for
the share of students it represents. In the regression, a unit corresponds to a degree. A 45-degree line is shown
in red.

Timing of job offer acceptances. Previous research has highlighted that gender dif-
ferences in job search behavior, such as women accepting job offers earlier due to higher
risk aversion, play a significant role in explaining early-career earnings disparities (Cortes
et al. 2023). However, these data do not support this explanation. In this study, both
women and men start their job search at similar times, yet women actually accept job
offers later than their male counterparts (Table A.2). Furthermore, the gender earnings
gap does not appear to shrink over the course of the job search period.

4.2 Fact 2: Cultural Persistence

A second motivating fact is that women’s labor market opportunities are shaped by the
societal role models they were exposed to during childhood—specifically, the working
behavior of other women. I document that the gender culture in a woman’s province of
origin has a lasting impact on her early career labor supply decisions—a phenomenon I
refer to as cultural persistence. I substantiate this claim, relying on the epidemiological ap-
proach of Fernandez 2007 and re-adapting it to analyze granular within-country variations
in cultural norms through the behavior of movers—individuals who work in a different

23
province from where they were born12. To isolate the influence of culture from other local
factors such markets and institutions, I examine the working behavior of female movers
using information on their province of residence prior to entrance in college and their
province of work. In this context, movers are defined as Italian-born individuals working in
a province different from their birthplace. The role of early cultural exposure is captured
by examining the labor supply of movers and linking it to various measures of gender
culture in their birthplace, as outlined in subsection 3.1. The fundamental idea underlying
this approach is straightforward: movers to the same local labor market share the same
market conditions and institutional settings, but they may not necessarily share the same
information, beliefs or preferences. According to theories of cultural transmission, one
key mechanism through which cultural traits evolve is through learning from societal role
models13. In the models of Fernandez 2013 and Fogli and Veldkamp 2011, one prominent
channel leading to the formation of gender culture, besides maternal influences, is the
observation of other women’s work behavior, a process referred to as oblique transmission
in the terminology of Cavalli-Sforza and Feldman 1981. A key prediction of these theories
is that women’s beliefs, broadly defined, will differ depending on the working behavior of
women in their place of origin. I test this hypothesis in Table 3, which compares the labor
supply of female movers depending on the quartile of FLFP in their province of origin,
controlling for the province where they work, as well as degree and cohort fixed effects
(Columns 2 and 5). The corresponding specification is:

(1) Y = β + α × Q4FLFP + θ + α + γ + ( ∑K β xk )+ε

idcp 0 idcp d c p k idcp idcp
k=1

where i refers to a female graduate in the subsample of movers, in a given cohort c who
has completed a Master’s degree d and is employed in province p. Q4FLFP is an indicator
variable that equals 1 if a graduate was born in a province with FLFP in the top quartile,
and 0 if from the lowest quartile. θd, αc, γp are a set of degree, cohort and province of
employment fixed effects. In some specifications, I add to the baseline model a set of
covariates (GPA and fixed effects for parents’ occupations). The results indicate that fe-
12. This approach is closely related to studies by Kleven 2024, Kerwin, Guryan, and Pan 2024 and Boelmann,
Raute, and Schönberg 2023.
13. See, for instance, Bisin and Verdier 2000 and Cavalli-Sforza and Feldman 1981.

24
male movers originating from high-FLFP provinces have significantly higher labor supply
compared to those from low-FLFP provinces, even when they work in the same local
labor market and graduate from the same Master’s program. This difference is statistically
significant and substantial, translating to a 7.6% increase in weekly hours worked and a 2.2
percentage point higher likelihood of full-time employment. This results in a 6.2% increase
in earnings, equivalent to €802 annually. Moreover, estimates vary little when accounting
for student’s performance and parental background. Robustness checks confirm that these
findings remain consistent across alternative measures of gender culture at the provincial
level.
TABLE 3. Estimates of gender culture on women’s labor supply at labor market entry

(1) (2) (3) (4) (5) (6)

Log(weekly hours) Pr(fulltime)

Q4 vs. Q1 FLFP 0.072∗∗∗ 0.076∗∗∗ 0.074∗∗∗ 0.025∗∗∗ 0.022∗∗∗ 0.021∗∗∗

(0.011) (0.011) (0.011) (0.008) (0.008) (0.008)

Province of job FEs ✓ ✓ ✓ ✓

GPA ✓ ✓
Parental background ✓ ✓
Degree FEs ✓ ✓ ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓ ✓ ✓
Observations 15,838 15,835 15,835 15,838 15,835 15,835
Nb. of degrees 1,218 1,218 1,218 1,218 1,218 1,218
R-squared 0.293 0.302 0.304 0.331 0.348 0.350
Notes: The table reports coefficients from separate regressions of women’s labor market outcomes on a
dummy variable indicating whether the student originates from a province with FLFP in the highest vs.
lowest quartile. All regressions include controls for degree and cohort fixed effects. Controls for parental
background include: FEs for education titles of mother and father (10 classes), FEs for occupations of mother
and father (12 classes). The sample consists of female movers, defined as women working in a different
province from their birth province, who are employed one year post-graduation. Standard errors are clustered
at the degree level.

While these results are striking, one concern in attributing these differences to gender
culture is that movers born in low-FLFP or high-FLFP provinces might differ in other
dimensions that impact their labor market outcomes. To investigate the importance of
such concerns, Table A.11 compares the ability, educational histories, and socio-economic

25
backgrounds of female movers based on the FLFP of their birthplace (top versus bottom
quartile). Specifically, it provides predictions from the empirical model 1 for these two
groups of women. A key advantage of the data is its rich set of pre-Master’s degree
characteristics, which allows to assess potential confounders. The table shows that, con-
ditional on attending the same degree program, movers from high-FLFP provinces are
similar to those from low-FLFP provinces in terms of ability and and socio-economic
background. In particular, the two groups are equivalent in parental education, while
there are some differences in parental occupations: mothers from low-FLFP areas are less
likely to hold low-SES jobs and more likely to hold medium-SES jobs than those from
high-FLFP areas, though there are no significant differences in their representation in
high-SES jobs. A notable exception is the labor market participation of mothers14, which
strongly correlates with local gender culture. Despite these differences in maternal role
models, they do not mediate the relationship between local gender culture and women’s
labor supply. As shown in Table, women’s labor supply has little to no correlation with
their mother’s employment status. Furthermore, women from low-FLFP areas appear
to be better selected in terms of their educational histories: they are more likely to have
attended general high-school tracks, particularly in science and humanities, compared to
their peers from high-FLFP areas. Therefore, evidence presented here does not suggest
that female students from low-FLFP areas have characteristics that explain why they fare
significantly worse in the labor market than women from high-FLFP areas. These findings
thus reduce concerns about differential selection of movers based on birth province and
support the validity of the epidemiological approach. A second concern arises from the
the fact that other local factors, such as economic activity, labor market conditions and
measures of social capital, correlate with spatial disparities in FLFP, as documented in
Acciari, Polo, and Violante 2021. Should these factors influence the beliefs and preferences
of individuals, the estimated relationship would not purely capture the effects of local
gender culture. To explore this possibility, I present an epidemiological analysis focusing
on the subsample of male students in Table A.12. While there is some positive relationship
between the gender culture in males’ province of origin and their working hours, the
magnitude of the coefficients is less than half of those observed for women.
14. This refers to the student’s response about whether their mother is currently employed at the time of the
survey.

26
This main goal of this subsection has been to provide comprehensive evidence that,
within degrees, labor supply differs systematically among women with similar academic
credentials based on the gender culture of their birthplaces. By focusing on a specific
population—young female graduates entering the labor market—and leveraging a setting
where admission rules are typically based on academic performance, this approach min-
imizes potential confounders and differential selection biases related to birth province.
Several factors may explain these observed differences, including differences in prefer-
ences, information disparities, as well as employers’ taste. While this section remains
agnostic about the precise mechanisms behind cultural persistence, Section 10 will provide
direct evidence of significant asymmetries in specific beliefs shaped by early cultural
exposure.

How are gender norms transmitted? What happens to women’s labor supply decisions
when they get exposed to peers raised in a different cultural environment? This setting,
together with unique characteristics of the data, allow to reproduce this experiment at a
large scale and explore the mechanisms of cultural change. Section 5 presents the empirical
framework used to identify the causal impact of peers’ background characteristics on
women’s labor market outcomes.

5 Identification Strategy and Empirical Model

Challenges and intuition. The identification of peer effects is notoriously challenging.

The main threat to their identification relates to selection, or endogeneous peer formation.
In this setting, peer groups are not formed at random, since individuals choose their
majors and universities. As a result, the characteristics of the peers they are exposed to are
likely correlated with their unobserved characteristics that drive their choice of a specific
peer group in the first place and that plausibly affect their success in the labor market,
leading to correlated effects in the Manski terminology (Manski 1993). In the absence of
randomization of students into peer groups, which is difficult to implement on a large
scale, my identification strategy overcomes the selection issue by leveraging within-
degree variation in the geographical origins of the peers to which consecutive cohorts of
students are exposed. This approach has been first proposed by Hoxby 2000 to assess the

27
impact of classmates gender and race on students’ educational outcomes, and has become
widely used for studying peer effects in education15. A key assumption in this strategy is
that changes in the geographical origins of students enrolling in a given program over
consecutive years are as good as random, i.e. no variables simultaneously influence the
characteristics of the students and the outcomes of interest. Several tests, detailed in the
next section, are conducted to verify whether this crucial identifying assumption holds in
practice. Implementing this approach requires access to longitudinal administrative data,
which is essential for the analysis. My dataset is particularly valuable because it includes
a comprehensive panel of Master’s programs - covering nearly all students in the country
- across multiple enrollment cohorts from 2012 to 2016. This level of coverage is unique in
the context of university education, as previous datasets were limited to small, selected
samples, often from a single university.

5.1 The empirical model

The associated empirical model is:

K
(2) Y = FP FP MP MP k

idc θd + αc + γFLFPidc + δ FLFP−i,dc + δ FLFP−i,dc + ( ∑βkxidc) + εidc

k=1

where i refers to a student who attended degree d in cohort c. The main outcomes of
interest are monthly earnings, weekly hours of work, an indicator variable for full-time
employment and hourly wages. Additionally, the empirical model will also be used to
assess the effects of peers’ characteristics on a number of other job characteristics, such
as the type of occupation and industry, location, or contract type. The empirical model
is estimated separately in the two subsamples of female and male students, allowing
for gender-specific peer effects. The main parameters of interest are δFP and δMP, i.e. the
treatment effects of exposing a student to a set of peers that are coming from places where
the FLFP is, on average, one percentage point higher. These estimates are causal under
the assumption that within degree-across cohort variation in the geographical origins
of students is as good as random. I provide evidence in favor of this assumption. The
15. A non-exhaustive list of papers includes: Cattan, Salvanes, and Tominey 2022, Cools, Fernandez, and
Patacchini 2022, Brenøe and Zölitz 2020, Black, Devereux, and Salvanes 2013, Angrist and Lang 2004, Olivetti,
Patacchini, and Zenou 2020, Carrell, Hoekstra, and Kuka 2018, Lavy, Paserman, and Schlosser 2012, Lavy and
Schlosser 2011, Bifulco, Fletcher, and Ross 2011

28
associated variables are the sample moments of the leave-one-out distribution of the FLFP
in the province of origin of students who belong to a specific gender, degree and cohort:

FP ∑j≠i FLFPjdc MP ∑j FLFPjdc

FLFP = if female=1; FLFP = if female=1;
−i,dc −i,dc
nFdc −1 nM
dc
FP ∑j FLFP MP ∑j≠i FLFPjdc
FLFP = jdc
if female=0; FLFP = if female=0;
−i,dc −i,dc
dc−1
n Fdc nM

Since the leave-one-out strategy introduces a mechanical negative correlation between the
FLFP in the own province of origin of a student and the average FLFP in the provinces
of her same-sex peers (Angrist 2014), I also condition for the FLFP in a student’s own
province of origin, FLFPidc. Alternatively, in a sensitivity analysis, I condition for province
of origin FEs instead of the FLFP in the province of origin. Note that, the linear-in-means
model assumes that what matters in a peer group is the average peers’ characteristic,
regardless of which peers’ allocation is leading to it. However, as pointed out by Boucher
et al. 2024, in many real-life situations peer effects are not linear in means. Additionally,
the most relevant policy implications of peer effects rely on non-linear models in which
not all individuals affect and are affected by their peers in the same way. In Section, I will
provide an analysis of non-linearities and heterogeneity.
Taking advantage of the panel nature of the data source (repeated observations on Master’s
programs), the inclusion of degree fixed effects θd allows to account for degree characteris-
tics that are constant across cohorts, for example whether the program tends to be attended
by students with specific set of background characteristics. More specifically, it accounts
for time-invariant unobserved determinants of earnings of students of a given gender
who graduate from a given program. The baseline specification also include enrollment
cohort fixed effects αc, to account for confounding factors affecting all individuals within
the same cohort. Finally, εidc is the error term, which is composed of a degree-specific
random element and an individual random element. Standard errors are clustered at the
degree level to account for potential correlation in students’ outcomes within degrees.

MP
Discussion. OLS estimates of δFP and δMP are unbiased if FLFP −i,mc and FLFP−i,mc are un-
FP

correlated with time-varying unobserved determinants of students’ earnings, conditional

on degree and cohort FEs. Therefore, for equations (1) to yield valid causal estimates of

29
these paramaters, the key identifying assumption is that cross-cohort changes in students’
geographical origins are random within degrees. Conceptually, this assumption is likely
to hold given the rules governing university admission in Italy. In fact, a majority of
Master’s degrees are selective (more than 55% according to estimates of the Ministry of
Education), meaning that admission is limited to a fixed number of students. Typically,
admission rules are decided by universities and involve an entrance exam, a standardized
test, or consideration of average grades from the Bachelor’s degree. Hence, in selective
FP MP
programs, variation in FLFP−i,mc and FLFP−i,mc reflect year-to-year variations in the geo-
graphic origins of students whose admission scores are high enough to be admitted into a
program. This design assumes that such variations are idiosyncratic, conditional on the
student’s bachelor’s GPA. In contrast, in non-selective programs, year-to-year changes in
the geographical origins of students stem from shifts in the applicant pool’s composition.
Other sources of cross-cohort variations in students’ composition can include changes
in admission policies, such as adjustments in the size of programs or in the admission
requirements. In the absence of granular data on admission policies, in some sensitivity
analyses I will use a data-driven approach to detect possibly non-random changes in size
and/or composition and exclude this subset of degrees from the analysis. The objective
of the next section is to describe the sources of the identifying variation and to provide
credible evidence that this variation is idiosyncratic.

5.2 Validity of the empirical strategy

Threat to identification. One critical concern in identifying peer effects is the presence
of correlated effects, in the Manski terminology (Manski 1993). In essence, similarities in
economic outcomes among individuals within a peer group are likely to stem from shared
individual characteristics or common shocks, rather than from social influence alone. In
this context, this translates into the possibility that cross-cohort changes in students’ geo-
graphical origins within master’s programs can correlate with time-varying unobserved
determinants of students’ labor market outcomes. For example, this could occur if labor
market trends in a specific region influence the applicant pool for programs in that region,
or if shifts in the student composition within a program impact the selection of new
FP MP
students. If this happens, then FLFP−i,mc and FLFP−i,mc are correlated with time-varying

30
determinants of outcomes εidc, leading to biases in δ FP and δMP. For this identification
strategy to effectively capture social influence, it is crucial that these cross-cohort fluctua-
tions are effectively random. The objective of this section is to examine the validity of this
identifying assumption through a series of checks.

Balancing tests for cohort composition. One empirical test of this assumption is to
verify that there is no selection, based on observables, into peer groups. Precisely, while
sorting of students into programs based on time-invariant characteristics - such as the
average peers’ composition - is admitted, I need to rule out that students systematically
sort into programs based on the specific composition of their cohort. To assess the plausi-
bility of the key identifying assumption that time-variant and unobservable factors are not
driving the results, I test whether there is systematic selection based on a wide range of
observable student characteristics. Specifically, I perform a battery of balancing checks in
which I test whether the peer composition in a Master’s cohort is systematically related to
a large vector of high-quality measures of student background characteristics observable
in the institutional data. For these placebo tests, I pick as characteristics pre-determined
covariates, that cannot be causally affected by peers but that might be correlated with
unobserved characteristics of other students enrolling in the same programs. These char-
acteristics include academic performance in previous education (e.g., Bachelor’s degree or
high school grades), and indicators of family socio-economic status, derived from detailed
information about the occupations and educational backgrounds of both parents. Tables B
and B present the results of these placebo checks on the subsamples of female and male
students, respectively. They report OLS estimates of δFP and δMP from equation 2. Each
column corresponds to a different regression, where the dependent variable is a different
pre-determined covariate. Results indicate that none of the estimated correlations appear
to be significantly different from zero in the model, indicating that the exposure to peers
from egalitarian provinces in the Master is unrelated to outcomes measured before entry
in the Master.
I also use a second approach to test whether students in a cohort with a different peers’
geographical composition have characteristics that are associated with different labor
market outcomes. I proceed in two steps. I first run separate regressions for female and

31
male students, where I predict their labor market outcomes - e.g. their labor supply at
the extensive and intensive margin as well as their earnings - based on all observable
pre-determined covariates (age, type of high-school (10 classes), high-school grade, BSc
grade, dummy for mother’s/father’s citizenship (Italian vs. not), FEs for education titles
of mother and father (10 classes), FEs for occupations of mother and father (12 classes)).
Second, I use the predicted outcomes as dependent variables in 2 to check balancedness
with respect peers’ composition. The advantage of this approach is that it quantifies any
non-balancedness of student characteristics. Table 4 summarizes these results, separately
for female and male students. The predicted labor market outcomes for the two sub-
samples appear balanced, providing strong evidence that cross-cohort changes in peers’
geographical composition are not related to covariates that explain changes in students’
labor market outcomes.
Finally, I perform a third test. I use all observable pre-determined covariates to predict
FP MP
the two treatment variables - FLFP −i,mc and FLFP−i,mc - in two regressions that include
cohort and degree fixed effects (FEs). I then perform an F-test for the joint significance
of all regressors. The two F-test results are 0.381 and 0.221, respectively, indicating that
we cannot reject the hypothesis that the regressors jointly do not explain the treatment
variables. Overall, I take the results from these two sets of tests as encouraging indi-
cation that the treatment variable is unlikely to be correlated with other time-varying
unobservable individual determinants of labor market outcomes. In fact, drawing from
Altonji and Blank 1999, we can reasonably infer that the degree of selection on observable
characteristics serves as a reliable indicator of the degree of selection on unobservables.
Identifying variation. Implementing this empirical strategy requires that there is suffi-
cient variation in the geographical composition of peers across cohorts within a master’s
program. This variation is necessary to obtain precise estimates of the peer effects I aim to
identify. Table 5 reports descriptive statistics for the average FLFP in the province of origin
of peers within a degree, before and after removing degree and cohort fixed effects. In
the cross-section, the standard deviation of this measure is 8.50 percentage points among
female peers and 8.59 percentage points among male peers, and is reduced to 1.86 and
2.13 after netting out degree and cohort fixed effects. The same descriptive statistics for
the other measures of gender culture are shown in A.15. Most of the variation in students’

32
 TABLE 4. Balancedness of Predicted Labor Market Outcomes

Panel A. Female sample

(1) (2) (3) (4)
Pr(employed) Log(monthly earnings) Log(weekly hours) Pr(fulltime)

δˆ F P -0.000 -0.000 -0.001 -0.000

(0.001) (0.002) (0.001) (0.001)
δ̂ M P -0.000 -0.000 0.002 0.001
(0.001) (0.002) (0.001) (0.001)

Degree FE ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 146,476 146,476 146,476 146,476
R-squared 0.119 0.246 0.239 0.273
Panel B. Male sample
(1) (2) (3) (4)
Pr(employed) Log(monthly earnings) Log(weekly hours) Pr(fulltime)

δˆ F P -0.000 -0.001 -0.001 -0.001

(0.001) (0.001) (0.001) (0.001)
δ̂ M P 0.001 0.001 0.002 0.001
(0.001) (0.002) (0.001) (0.001)

Degree FE ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 106,448 106,448 106,448 106,448
R-squared 0.204 0.259 0.316 0.324
Notes: OLS estimates of a regression of predicted labor market outcomes on: the average FLFP in the provinces
of origin of female and male peers and the FLFP in the own province of origin. Regressions include cohort and
degree fixed effects. The estimating equation is 2 in the main text. All regressors are standardised. Standard
errors are clustered at degree level. Labor market outcomes are predicted separately for female and male
students based on regressions of labor market outcomes on a set of pre-determined invididual covariates:
age, type of high-school (10 classes), high-school grade, BSc grade, dummy for mother’s/father’s citizenship
(Italian vs. not), FEs for education titles of mother and father (10 classes), FEs for occupations of mother and
father (12 classes).

characteristics occurs across different degrees, indicating that there is sorting of students,
as expected. However, there is some variation within a degree over time. Between one-
fourth and one-fifth of the total variation in peers’ gender culture remains unexplained
after accounting for fixed effects: I rely on this variation to estimate peer effects, and all

33
estimates are precisely estimated. Interestingly, the magnitude of this residual variation
aligns with what we would expect from random fluctuations. To explore this further, I
conducted 500 simulations in which students were randomly assigned to degrees and
cohorts - think about it, using a distribution of FLFP that matches the mean and standard
deviation of the actual sample distribution. Across these simulations, the residual standard
deviation of the average FLFP in the province of origin of peers is centered around 1.57
percentage points, with a range of 1.53 to 1.62 percentage points, consistent with the
values observed in the actual sample. As a further randomization check, I inspect whether

TABLE 5. Raw and Residual Variation of Peers’ Gender Culture

Mean SD Min Max

A: Avg FLFP in province of origin of female peers

Raw cohort variable 49.65 8.50 29.87 66.66
Residuals: net of degree and cohort FEs 0.00 1.86 -13.80 10.79

B: Avg FLFP in province of origin of male peers

Raw cohort variable 49.72 8.59 27.33 66.66
Residuals: net of degree and cohort FEs 0.00 2.13 -16.80 14.51

Notes: The table reports descriptive statistics for the average FLFP in the province of origin of female (Panel
A) and male (Panel B) students within degrees, before and after removing degree and cohort fixed effects.
The unit of observation is a degree-cohort pair, leading to a total of 7,160 observations.

the variation in students’ geographical composition is consistent with variation that we

would expect with natural random fluctuations. Figure 5 plots the average FLFP in peers’
provinces, with separate panels for female peers (Panel a) and male peers (Panel b), after
residualizing on degree and cohort fixed effects. Figure A.3 provide the same graphical
evidence using all other measures of gender culture used in the paper. Deviations from
the average students’ composition closely follow the normal distribution, which is plotted
for comparison. The shape of the distribution further supports the idea that student’s
geographical composition is as good as random, conditional on the included controls.
Another graphical illustration of the identifying variation is shown in Figure A.4, which
plots the time series of the average FLFP of female peers across the years, for a randomly
picked program within each decile of program’s size.

34
 FIGURE 5. Year-to-Year Variation in Students’ Geographical Origins

(a) Female peers (b) Male peers

Notes: The figure plots the distribution of residuals from a OLS regression of the average FLFP in the province
of origin of female (Panel a) or male students (Panel b) on cohort and degree fixed effects. One observation
corresponds to a degree-cohort pair. Histograms are presented by bins of 0.75. The normal distribution is
plotted for comparison.

Identifying variation and program’s size. If the variations in students’ characteristics

across cohorts within a degree program were random, the law of large numbers would
predict that the average characteristics of peers in a cohort would converge to the true
value of the program as the program size increases. This leads to a testable prediction: the
magnitude of cross-cohort changes in students’ geographical origins within a Master’s
program should become smaller as the program size grows. This is shown in Figure 6,
which plots year-to-year changes in students’ origins for degrees in the lowest and highest
quintile of size. The lowest quintile corresponds to degrees with an average size below 22
students, while the highest quintile includes degrees with an average size between 70 and
413 students. It is also further tested in Table A.16, which provides descriptive statistics
for the main variable defining peers’ gender culture, in different groups of degrees catego-
rized by the quintiles of their size. It is clear that while the raw standard deviation of the
variable shows little change across quintiles, the residual variation decreases significantly
as program size increases, consistent with random fluctuations. These findings underscore
a key advantage of this dataset compared to those used in previous studies on peer effects,
that examined entire schools as peer groups. Specifically, the granular data on program
composition allows for the analysis of peer groups with small cohort sizes (median=34,
mean=47), where cross-cohort changes in peers’ characteristics are likely driven by ran-

35
dom fluctuations.
FIGURE 6. Year-to-Year Variation in Students’ Geographical Origins by Program Size

(a) Female peers (b) Male peers

Notes: The figure plots the distribution of residuals from a OLS regression of the average FLFP in the province
of origin of female (Panel a) or male students (Panel b) on cohort and degree fixed effects. The distributions
are shown separately for degree programs in the first and highest quintiles of size. Degree programs are
divided into quintiles based on their average size across five cohorts: the first quintile includes degrees with
fewer than 21 students, while the fifth quintile includes degrees with 70 to 413 students. Each observation
represents a degree-cohort pair. Histograms are presented by bins of 0.75.

Other checks. The evidence provided in this section supports the hypothesis that year-
to-year changes in the geographical composition of students are random within degrees.
However, it does not entirely rule out the influence of simultaneous institutional factors,
such as fluctuations in the local labor market, that might affect outcomes across different
student cohorts.To address these potential concerns, I undertake several robustness checks.
First, I extend the baseline model by incorporating either (i) degree-specific linear time
trends or (ii) region-specific linear time trends. Second, I perform sensitivity analyses
excluding degree programs that might exhibit non-random changes in size and/or compo-
sition. This includes programs with observable trends in size or geographical composition,
as well as those with changes in the average or standard deviation of students’ academic
performance. These results are included as part of the robustness checks.

36
6 Baseline Estimates of Peer Effects

6.1 Estimation Results

This section presents the baseline estimates of gender-specific peer effects δFP and δMP,
derived from the empirical model of equation 2. These estimates come from specifications
that condition on degree and cohort fixed effects, as well as the FLFP in the own province
of origin of a student, and correspond most closely to the hypothetical experiment de-
scribed in Section . The analysis is first conducted on the sample of female students, and
is then extended to the sample of male students as a placebo.

Estimates of peer effects on female earnings and labor supply. Estimates of the empirical
model on the sample of female students are presented in Table 6. The outcome variables
are monthly earnings, weekly hours worked and hourly wages, all in logarithmic forms,
and an indicator variable of fulltime employment. Regressors are standardised. Results
indicate that women who study in cohorts where female classmates are born in places
with higher FLFP increase their labor supply along the intensive margin, both through
higher take-up of full-time jobs and increases in weekly hours worked (Columns 2 and
FP
3). The magnitude of this effect is large: a one standard deviation increase in FLFP−i,mc
(8.50 percentage points) is associated with a 3.3% increase in weekly hours and in a 1.9
percentage points increase in the likelihood of fulltime employment one year after grad-
uation, a 2.5% increase relative to the mean. This translates into a 3.7% increase in their
monthly earnings. These estimated peer effects are economically significant, comparing
to 33% − 40% of the size of the gender differences in the same outcomes and accounting
for 45% − 76% of the gap between women born in low-FLFP and high-FLFP provinces.
These findings also highlight the presence of gender-specific peer effects for women.
Specifically, the positive effects on earnings and hours worked are entirely attributable to
variations in the geographical origins of female peers. In contrast, the estimates for δMP
are zero across all outcomes, suggesting that exposure to male classmates from areas with
higher FLFP does not affect women’s earnings or labor supply decisions. There are several
reasons why peers’ gender might matter in this context. One key reason involves the
mechanisms generating peer effects. Earlier, I documented a much stronger relationship

37
between female outcomes in the labor market and gender norms in their province of origin,
whereas men’s outcomes showed little to no relationship with these cultural measures. A
natural implication is that increasing the share of women from high-FLFP areas within a
cohort systematically raises the representation of women who hold beliefs and preferences
associated with acceptances of full-time jobs. This association is much weaker among men.
Consequently, if peer effects operate through mechanisms such as (i) conformism or (ii) so-
cial learning, it is reasonable to expect little or no influence from male peers in this setting.
However, opposite-sex peers might influence women’s labor market decisions through
other channels, especially in contexts where couples are likely to form (Bursztyn, Fujiwara,
and Pallais 2017). For instance, men raised in more egalitarian environments might have
different expectations regarding their partners’ work behavior, as shown in Fernández,
Fogli, and Olivetti 2004. If students pair up with classmates, this could be a significant
channel through which the gender culture of men might impact women’s choices. Yet, this
potential explanation is not supported by these findings or by evidence from the original
survey. As I will discuss in Section 10, the share of couples formed among students in the
same program is low. Another possible explanation for why peer gender matters relates
to the structure of social networks. Previous studies in sociology have documented that
the quality and quantity of interactions differ between same-sex and opposite-sex peers
(see, for instance, cook). Homophily—the tendency for peer groups to be segregated by
gender—can be critical in determining how information spreads (Currarini, Jackson, and
Pin 2009). To explore the relevance of this channel, I collected data on network structure
through the original survey. The results will be presented in Section 10.
Estimates of peer effects on female occupational choices. Table 7 presents estimates
of peer effects on the types of occupations and industries women enter one year after
graduation (Columns 1-4). Occupations and industries are categorized based on their
average monthly earnings and the share of full-time employment. The dataset includes 20
occupations and 21 industries. An occupation or industry is classified as "high-earnings"
or "high-fulltime" if it ranks above the median in these distributions. These four indicators
are used as outcome variables. The results indicate that exposure to female classmates from
high-FLFP areas also affects women’s occupational choices (Columns 1-2). Specifically,
FP
a one standard deviation increase in FLFP −i,mc is associated with a 1.7 percentage point

38
 TABLE 6. Estimates of peer effects on earnings and labor supply - Female sample

(1) (2) (3) (4)

Log(monthly Log(weekly Pr(fulltime) Log(hourly
earnings) hours) wage)

δˆ F P 0.037*** 0.033*** 0.019** 0.003

(0.013) (0.012) (0.009) (0.012)
δ̂ M P -0.000 0.001 -0.002 -0.002
(0.010) (0.009) (0.007) (0.010)

Degree FE ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 69,645 69,645 69,645 69,645
R-squared 0.287 0.246 0.280 0.100
Notes: OLS estimates of a regression of women’s earnings and labor supply one year after graduation on:
the average FLFP in the provinces of origin of female and male peers and the FLFP in the own province
of origin. Regressions include cohort and degree fixed effects. All the estimates are done on the sample of
women who are employed one year after graduation and with non-missing information on these variables.
Standard errors clustered at degree level. All regressors are standardised.

increase in the likelihood of choosing a high-earnings occupation and a 1.5 percentage

point increase in the likelihood of selecting an occupation with a high share of full-time
jobs, representing a 4.6% and 2.9% increase relative to the mean, respectively. As high-
lighted in the previous paragraph, these peer effects are gender-specific. To assess the role
of occupational changes in the rise of women’s labor supply, I re-estimated the model
for weekly hours worked, including occupation and industry fixed effects. Although
the estimated coefficient for δFP is reduced by about one-third, the coefficient on weekly
hours remains large and statistically significant. This suggests that changes in occupations
account for only part of the increase in women’s labor supply. This finding aligns with
earlier evidence showing that occupational differences between men and women explain
less than a third of the total gap in labor supply.

Estimates of peer effects on other job characteristics. While exposure to female peers
from high-FLFP provinces affects women’s labor supply, earnings, and occupational
choices, it does not impact sorting along other observable dimensions, such as employers’
characteristics. For example, hourly wages are not affected (Table 6, Column 4), and there

39
 TABLE 7. Estimates of peer effects on occupations and industries - Female sample

(1) (2) (3) (4) (5)

Occupation Industry
High-earn High-fulltime High-earn High-fulltime Log(weekly
hours)

δ̂ F P 0.018** 0.016* 0.014 0.008 0.023**

(0.009) (0.009) (0.010) (0.009) (0.011)
δ̂ M P -0.004 -0.005 -0.003 -0.009 -0.000
(0.006) (0.007) (0.007) (0.006) (0.009)

Occ. & ind. FE ✓

Degree FE ✓ ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓ ✓
Observations 68,216 68,216 68,419 68,419 69,645
R-squared 0.361 0.466 0.272 0.398 0.349
Notes: OLS estimates of regressions of types of occupations and industries one year after graduation on: the
average FLFP in the provinces of origin of female and male peers and the FLFP in the own province of origin.
The dependent variables in Columns (1) and (3) are constructed from the distribution of earnings across
occupations and industries, respectively. Specifically, indicators of high-earning occupations (industries) are
based on whether an occupation (industry) pays above-median earnings. The dependent variables in Columns
(2) and (4) are constructed from the distribution of fulltime jobs across occupations and industries, respectively.
Specifically, indicators of high-fulltime occupations (industries) are based on whether an occupation (industry)
has above-median shares of fulltime jobs. Regressions include cohort and degree fixed effects. In Column 5, I
add occupation FEs (20 classes) and industry FEs (21 classes). All the estimates are done on the sample of
women who are employed one year after graduation and with non-missing information on these variables.
Standard errors clustered at degree level. All regressors are standardised.

is no significant effect on the industry in which women are employed. Table A.17 presents
estimates of the empirical model for other job characteristics observed in the data, such
as whether the employer is in the public or private sector or the type of employment
contract (permanent, no contract and self-employment). None of these variables show any
influence from peer exposure.

Effect of peers on male outcomes. In Table 8, I repeat the analysis on the male sam-
ple as a placebo test. The rationale here is that if the FLFP (and the other measures used
in this paper) in a student’s province of origin accurately reflects a set of beliefs and
preferences that are gender-specific, we should not expect the gender culture of peers to
have a direct impact on men’s labor supply and earnings—just as men’s decisions are not

40
influenced by these measures in their own province of origin. However, indirect effects,
e.g. in the form of spillovers, could still occurr, for instance, if some men feel pressure due
to women’s rising aspirations. Results on male students indicate that exposure to female
or male peers from high-FLFP provinces has no impact on men’s weekly hours and the
likelihood of fulltime employment. Peers of both genders have only small positive effects
on men’s hourly wages.

TABLE 8. Estimates of peer effects on earnings and labor supply - Male sample

(1) (2) (3) (4)

Log(monthly Log(weekly Pr(fulltime) Log(hourly
earnings) hours) wage)

δˆ F P 0.013 -0.000 -0.001 0.014*

(0.008) (0.008) (0.006) (0.008)
δ̂ M P 0.013 -0.005 0.004 0.018*
(0.011) (0.010) (0.008) (0.010)

Degree FE ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 57,476 57,476 57,476 57,476
R-squared 0.246 0.233 0.270 0.107
Notes: OLS estimates of a regression of men’s earnings and labor supply one year after graduation on: the
average FLFP in the provinces of origin of female and male peers and the FLFP in the own province of origin.
Regressions include cohort and degree fixed effects. All the estimates are done on the sample of men who are
employed one year after graduation and with non-missing information on these variables. Standard errors
clustered at degree level. All regressors are standardised.

7 Robustness and Validation of the Design

This section has two primary goals. The first is to assess the validity of the key identifying
assumption - that cross-cohort changes in students’ geographical origins are random
within degrees - using a series of tests that address different forms of selection and the
presence of correlated effects. Once these concerns about the identification of peer effects
are mitigated, the second objective is to investigate whether the estimated effects can be
attributed to peers’ gender culture or other characteristics. Additionally, I re-estimate the
model using alternative specifications to ensure that the results are robust and not overly

41
dependent on specific assumptions regarding the empirical model and the clustering of
standard errors.

7.1 Validation of the design

To organize the analysis, consider that the unobserved determinant of students’ earnings,
represented by εidc in equation 2, is composed of two elements: a degree-specific random
element, vdc, which reflects time-varying inputs at the degree level, and an individual
FP
random element, uidc . For the identification strategy to be valid, changes in FLFP−i,mc
MP
and FLFP−i,mc must be uncorrelated with both uidc and vdc. In Section, results from a

series of balancing tests demonstrated that changes in FLFP−i, mcFP and FLFP−i, mcMP
are uncorrelated with several observed individual characteristics, such as students’ de-
mographics, average ability, education history, and socio-economic status (all measured
prior to entry into the program). A key finding is that improvements in the labor market
outcomes of different cohorts of women are not driven by differences in their observed
characteristics, as shown by the balancedeness of their predicted outcomes. Given the high
number of observed characteristics, these tests provide credible evidence that changes in
FLFP−i, mcFP and FLFP−i, mcMP are likely uncorrelated with changes in uidc. Therefore,
the main objective of this section is to test for potential biases due to correlated effects.
There are two primary sources of time-varying changes in the degree-specific random
component: (i) changes in program characteristics, such as those driven by admission
policies affecting program size and the dispersion of students’ abilities, and (ii) regional
shocks that impact the labor market outcomes of all students working within a given local
labor market.

Trends. In the baseline specification, the inclusion of degree fixed effects accounts for
time-invariant factors that affect the outcomes of students graduating from a given pro-
gram. However, concerns might arise that the estimates of the main parameters could be
confounded by degree-specific trends. This could happen if, for example, trends in stu-
dents’ outcomes of a specific program, e.g. related to changes in its quality - systematically
FP
change the selection of students into the program. In this case, changes in FLFP −i,dc and
MP
FLFP −i,dc would be correlated with time-varying unobserved factors driving the trends

42
in outcomes, which would bias the estimates. To address these concerns, I extend the
baseline specification by incorporating degree-specific linear time trends to account for
potential changes in degree effects over time. The empirical model now becomes:

−i,dc + δ FLFP −i,dc + εidc

Yidc = θd + αc + ϕd טt + γFLFP idc + δFPFLFPFP MP
(3) MP

where ϕd × ˜t is a degree-specific linear time trend, and ˜t indicates the distance between
a students’ cohort and a reference cohort (i.e. 2012). Table A.18 reportes the estimates of
δˆ F P and δˆ M P from the augmented model.The results are similar to those of the baseline
model, suggesting that it is unlikely that students select into degrees based on trends in
outcomes.
As an additional robustness check, I re-estimate the main specification augmented with
region-specific linear time trends, following the specification below:

(4) Yidc = θd + αc + ψr × ˜t + γFLFPidc + δFPFLFPFP

−i,dc + δ FLFP −i,dc + εidc
MP MP

where ψr × ˜t is a region-specific linear time trend, and ˜t indicates the distance between
a students’ cohort and a reference cohort (i.e. 2012). Including time trends at the region
of study level helps control for potential trends in student outcomes within a particular
region that could influence students’ selection. The results of this specification are reported
in Table A.19. The estimates remain unchanged, which reduces concerns about potential
confounding factors related to regional trends.

Non-random changes in peers’ composition. This empirical strategy assumes that that
cross-cohort variations in the geographical origins of students within degrees are due to
random fluctuations. In selective programs, this assumption is plausible because changes
FP MP
in FLFP−i,dc and FLFP−i,dc result from shifts in the geographical origins of students with
sufficiently high admission scores. In non-selective programs, however, these changes
are driven by shifts in the applicant pool—are these random, or influenced by selection?
Previous checks suggested that it is unlikely that students systematically select based on
trends in outcomes within specific degrees or regions. Ideally, I would like to conduct
an additional test, by restricting the analysis to selective degrees that don’t experience

43
changes in admission requirements. However, due to data limitations, I cannot perform
this test, instead I use a data-driven approach to identify degrees likely to experience
non-random changes in student composition. First, I re-estimate the baseline model on
a subset of degrees that show no trends in size. To conduct this exercise, I run separate
regressions for each degree in the dataset, treating the size of a program in a given cohort
as the dependent variable and regressing it on a constant and a linear time trend. A degree
is considered to have a trend in size if the p-value from the statistical test on the time
variable does not exceed 0.10, which applies to about one-fourth of the degrees. I then
re-estimate the baseline model on the degrees that were not flagged. The results, shown
in Table A.20, are consistent with those from the main sample, with coefficients being
generally higher in this cleaner subset. I conduct an additional set of sensitivity analyses
to examine whether the estimates vary across different samples of degrees by excluding
those with significant changes in student composition over time. Specifically, I investigate
whether degrees experience too large shocks in the following student characteristics: (i)
average student ability, measured by the average final grade in the Bachelor’s degree
(prior to entering the Master’s program), (ii) the dispersion of student ability, assessed
through the standard deviation of Bachelor’s grades within a cohort, and (iii) cohort size.
Such changes might indicate shifts in admission policies or in students’ selection. To gauge
the severity of these shocks within programs over time, I analyze the residual variation in
characteristics (i)-(iii) after accounting for degree and cohort fixed effects16. Specifically,
for each degree d in the dataset and each characteristic Y in (i)-(iii), I construct a measure
summarizing the cross-cohort variation in peers’ characteristics within a degree:
1 Tmax Y
(5) ZY = ∑ ∣r ∣
d Tmax t=1 dt

where rYdt represents the residual obtained from regressing the average characteristic Y
in a cohort of a degree on degree and cohort fixed effects. T max denotes the maximum
number of cohorts a degree is observed in the data (for 92% of degrees, T max = 5). I
transform ZYd into a relative measure by dividing it by the mean characteristic of the
1
program across years, T max ∑Tt=1
max
Ydt . Next, I categorize degrees based on their position
in the distribution of this relative measure, separately for each characteristic Y. I then
16. Note that the unit of observation is a degree-cohort pair.

44
conduct a set of sensitivity analyses, as shown in Table A.21, with the dependent variable
being log(monthly earnings). The estimates across different samples are very similar, and
in some cases higher, to the benchmark specification, which is presented in Column 1.

7.2 Sensitivity to Sample Restrictions

This sub-section assesses the sensitivity of peer effect estimates to alternative sample
restrictions. Specifically, I investigate whether the results vary among samples defined
by degree size and the proportion of students who completed their Bachelor’s degree at
the same university. The results on log(monthly earnings) are presented in Table A.23. I
then discuss how these restrictions impact the baseline specification, which is provided in
Column (1) for comparison.
Heterogeneity by degree size. One concern is that my identification strategy might not
be valid in very small programs, where students have more chances to anticipate the
peers’ composition they will face. However, Column (2) shows that the benchmark es-
timates remain almost unchanged when excluding degrees in the bottom decile of the
size distribution (where the average number of students per degree is fewer than 13). This
suggests that the benchmark estimates are not influenced by noise or possible endogenous
peer formation stemming from very small programs. Second, estimated effects are similar
to those of the benchmark specification when degrees in the highest decile of the size
distribution are dropped (where, on average, there are more than 87 students), as indicated
in Column (3). In this sample of large programs, the effects vanish and standard errors
become very large, as shown in Column (4). The latter finding is consistent with evidence
provided in Section that the magnitude of cross-cohort variations in peers’ geographical
origins become significantly smaller as the degree size increases. Lastly, estimated peer
effects are significantly higher in degrees with sizes below the mean (43 students), while
they become smaller and less precisely estimated in degrees above the mean. This aligns
with the idea that, in large programs, the network structure is likely to adjust to changes
in the initial cohort composition. For example, students can segregate into distinct social
networks, possibly related to their background characteristics, resulting in decreased
beneficial social interactions among out-group members, as in Carrell, Sacerdote, and
West 2013.

45
Heterogeneity by proportion of students with Bachelor at the same institution. Another
concern arises in degrees where a significant portion of students completed their Bach-
elor’s degree at the same institution. In such cases, students may have moved together
from a shared Bachelor’s program to a shared Master’s program. This could result in cross-
cohort variation in peers’ origins being driven by selection rather than being idiosyncratic.
To address this concern, I exclude degrees where the vast majority of students completed
their Bachelor’s at the same institution (Column 6). The estimates remain robust after this
exclusion. Additionally, I provide results from the empirical model on a sample of degrees
where the proportion of students who completed their Bachelor’s at the same institution
falls within the bottom 25% (Column 7). Despite the significant reduction in sample size,
the estimated effects are larger in magnitude than the benchmark estimates and precisely
estimated.
Heterogeneity by student attendance to classes. As a placebo, I conduct the analysis
focusing on the subset of students who do not attend classes in the Master’s program.
This subset is identified through the pre-graduation survey, where some students report
working full-time for the entire duration of the Master’s program, constituting 8.7% of
the sample. Since these students likely have limited interactions with their peers, peer
effects within this subgroup should be minimal. The results of the analysis regarding
heterogeneity by students’ attendance are detailed in Table ??. As expected, the findings
reveal significant heterogeneity in the effects of peers: while the peer effects are substantial
for students with high attendance to classes, there is no evidence of an effect for students
with low attendance.

7.3 Sensitivity to Measures of Gender Culture

This sub-section assesses the sensitivity of peer effect estimates to different measures
of local gender culture. Specifically, I explore whether the results change when using
six additional proxies for local gender culture, as described in Section. The results for
log(monthly earnings) are presented in Table A.24. Each column corresponds to a separate
regression using a different proxy for peers’ gender culture, as indicated by the column
labels. Column (1) replicates the baseline specification from Column (1) of Table 6 for
reference. Across the different measures of students’ gender culture, the estimates remain

46
consistent, with slightly larger effects observed when gender culture is defined using
the labor market behavior of younger women and female graduates, suggesting that
younger women might be a more relevant reference group. When gender culture is based
on women’s labor market behavior, the peer effect estimates range from 0.035 to 0.041.
Using the proxy based on firms’ gender culture, the estimates are somewhat smaller (0.016)
but remain statistically significant.

7.4 Which peers’ characteristics matter?

The main finding of this paper is that exposure to female classmates from provinces with
more egalitarian gender cultures positively impacts women’s labor market outcomes.
But which peer characteristics drive this effect? This section investigates whether the
observed effects are specifically due to peers’ gender culture or other factors. In Section,
I documented that movers from high- and low-FLFP areas do not differ significantly in
key aspects. To further ensure the robustness of these findings, I incorporate controls for
alternative peer characteristics into the analysis. The results of these tests are presented
in Table A.25. Each column represents a separate regression on log(monthly earnings),
following the specification in 2 and adding controls for different peer characteristics.
For instance, Column 1 includes controls for the share of female and male peers with
working mothers, while Column 6 accounts for the share of peers with above-median
grades before entering the Master’s program. Other regressions include controls for peers’
socio-economic background and educational history. The peer effect estimates remain
stable or even slightly increase with these additional controls. This suggests that the
estimated effects are unlikely to be confounded by other peer characteristics, such as
ability or socio-economic status. In Columns 8 and 9, I further control for characteristics
of degree programs, such as program size and the share of female peers. Adding these
controls does not change the estimates, reinforcing the conclusion that the peer effects are
not driven by shocks occurring within specific programs.
Other checks: fixed effects. Finally, I modify specification 2 to include for province of
origin FEs instead of controlling for the FLFP in the own province of origin of the student.

47
The adjusted empirical model now becomes:

MP
−i,dcp + δ FLFP −i,dcp + εidcp
Yidcp = θd + α c + ω p + δFPFLFPFP MP
(6)

where ωp are province of origin fixed effects. The results of this specification are equivalent
to those of the main specification, as shown in Table A.22.

8 Beyond the Linear-in-Means Model

The Linear-in-Means (LiM) model, as presented in equation 2, is the most frequently esti-
mated framework in the peer effects literature. This model posits that a student’s outcome
is a linear function of the average background characteristics of their peers. However,
this approach imposes strict assumptions on the nature of peer effects. First, it constrains
the size of peer effects (δFP and δMP) to be the same regardless of where the student falls
within the distribution of student background. Second, it assumes that peer effects operate
exclusively through the mean, disregarding any influences from other aspects of the peers’
background distribution. For instance, mean-preserving increases in the variance of peers’
background characteristics are assumed to have no effect. The linear-in-means model has
the virtue of simplicity but does suffer from the disadvantage of not being a convincing
description of the world (Sacerdote 2014, 2011). Peer effects are often non-linear in various
contexts (Boucher et al. 2024)17. The accuracy of the LiM as a model carries substantial
implications for social welfare. Nonlinearities in peer effects open up the possibility that
the outcomes of some students could be improved by a change in peers without detriment
to others. Conversely, if peer effects are strictly linear in means, then regardless of how
peers are arranged, society would have the same average level of outcomes. This section
aims to explore the potential for such nonlinearities in peer effects.

First, I turn to specifications in which I allow the effects of peers to vary with a stu-
dent’s own gender culture. Specifically, I associate each student with the quartile of FLFP
17. Beginning with Hoxby and Weingarth 2005, a number of empirical studies have investigated these
non-linearities, including Tincani 2024, Booij and Leuven 2017, Feld and Zölitz 2017, Lavy, Paserman, and
Schlosser 2012, Imberman, Kugler, and Sacerdote 2012, Burke and Sass 2013, Carrell, Fullerton, and West
2009, Hanushek and Rivkin 2009.

48
in her province origin, relative to the distribution of FLFP in this sample.

9 Mechanisms

The previous section presents evidence that peers’ gender culture significantly influences
women’s early career labor market decisions, with these effects being notably large. To gain
further insight into the driving forces of peer effects, this section examines the relevance
of possible mechanisms. In the first subsection, I examine whether endogenous variations
in effort or networks can explain the results. In the following subsection, I draw on data
from both the institutional survey and the original survey I conducted to explore the roles
of cultural transmission and social learning.

9.1 What peers don’t do

1. Human capital. As previously noted, the FLFP in a student’s province of origin does
not serve as a proxy for her ability (Table A.11). Women from provinces in the highest and
lowest quartiles of FLFP show no significant differences in observed measures of ability,
such as their grades in prior education (e.g., Bachelor’s degree) or their GPA and gradua-
tion grades in the Master’s program. However, there is a concern that local characteristics
in a student’s province of origin could influence unobserved factors, such as motivation.
If women from high-FLFP areas possess higher levels of these unobserved traits, which
could explain their better labor market outcomes, then exposure to a higher proportion
of such individuals might affect other students’ efforts. To investigate this possibility, I
replicate the empirical analysis using indicators of women’s academic performance as the
outcomes, as shown in Table A.26. The outcome variables include GPA in the Master’s pro-
gram (on a 0–30 scale), graduation grade (on a 66–110 scale), and an indicator of delayed
graduation (fuoricorso). The table provides the sample averages of these variables. The
results indicate that peers’ gender culture has no significant effect on women’s academic
performance. None of the estimated coefficients is statistically different from zero, and the
magnitudes of the estimates are small. For example, a one standard deviation increase
in the gender culture of female peers is associated with only a 0.071-point increase in
graduation grade, representing a minimal 0.06% increase relative to the mean. Based

49
on this evidence, I conclude that human capital is not a mediating factor in enhancing
women’s labor market outcomes.

2. Geographic mobility. Another plausible explanation is that exposure to peers from

high-FLFP provinces directly influences women’s decisions on where to search for a job.
For example, women from high-FLFP areas might share information about their local
labor markets, or the friendships formed during their studies could encourage women
to follow their peers into these markets. If either of these channels operates, we would
expect the geographic origins of a woman’s peers to impact her mobility decisions. To test
this hypothesis, I use characteristics of the local labor market where a woman works as
outcome variables in the empirical model described by equation 2. The results, presented
in Table A.27), indicate that the geographic origins of peers, as measured by the FLFP in
their province of origin, do not affect women’s mobility decisions. The outcome variables
examined include the FLFP in the province of work, whether a woman is employed in the
same region where she studied or elsewhere, and whether she is working in a province
different from her birth province. Sample averages of these variables are provided in the
table. // A few descriptive facts regarding women’s mobility are worth noting: 68.4% of
students find their first job in the same region where they studied, 5% work abroad, and
the remainder work in a different region than where they studied, possibly their province
of origin. Among those who migrated to another province for their studies, 39% return to
their home province, with significant variation across quartiles of gender culture. Women
from the highest quartile of FLFP are nearly twice as likely to return to their province of
origin compared to those from the lowest quartiles. . Results from Table A.27 indicate
that exposure to peers from high-FLFP provinces does not significantly impact women’s
mobility decisions, such as choosing to work in the region where they studied or opting to
work outside their birth province. None of the estimated coefficients is statistically differ-
ent from zero, and the magnitudes of the estimates are small. For instance, a one standard
deviation increase in peers’ gender culture corresponds to a mere 0.155 increase in the
FLFP in the woman’s province of employment, amounting to a marginal 0.28% increase
relative to the mean. Furthermore, the main findings remain stable when accounting for
fixed effects for the province of employment, as shown in Table. Overall, these results

50
suggest that the observed peer effects on female earnings and labor supply are not driven
by changes in geographic mobility.

Networks. Economically beneficial labor market connections, such as networks to better

firms, offer an alternative explanation for the observed earnings gains among female
students. Unfortunately, the absence of firm identifiers in the data precludes a formal
test of the extent to which networks contribute to the estimated peer effects. Nonetheless,
I provide suggestive evidence that these networks are unlikely to be a major driver of
increases in women’s labor supply. The underlying logic of this test is straightforward:
cross-cohort changes in the average female labor force participation (FLFP) in peers’
provinces within a degree program could correlate with shifts in the relative proportions
of students who are "local" versus "movers." In this context, "local" refers to students who
study in their province of birth. If locals have better connections to local firms, e.g. through
their parents, they could share new information about job opportunities or provide job
referrals to peers with different information sets. Given the documented importance of
networks at labor market entry (Zimmerman 2019, Kramarz and Skans 2014, Hampole
and Wong 2024, Fischer, Sandoy, and Walldorf 2023, Einiö 2023), this could be a significant
channel influencing earnings. In practice, if locals indeed have better connections to local
firms, two conditions would lead to an upward bias in the estimated effects of peers’
culture: (i) cross-cohort variation in peers’ gender culture correlates with increases in the
share of locals within a program, and (ii) peers serve as networks facilitating access to
better firms. To test this hypothesis, I include controls for the shares of female and male
peers who are locals in the main specification. The results, presented in Table A.28, show
that the estimates of the effects of peers’ gender culture remain robust. Additionally, the
magnitude of these estimates slightly increases, likely due to the small but significant
negative effect of a higher share of local students on women’s earnings and labor supply.

9.2 Cultural transmission and learning

Given the lack of empirical support for channels related to students’ networks or human
capital, I explore two alternative explanations that could rationalize my findings and align
with existing theories of cultural change. One view is that peers could affect the beliefs

51
and preferences of women in a way that influence their labor market behavior—a process
known as cultural transmission, as outlined by Bisin and Verdier 2001. The second, not
mutually exclusive, mechanism involves social learning, where students share new infor-
mation with peers with different information sets, following the framework of Fernandez
2013. In this section, I present evidence supporting both mechanisms. First, I show that stu-
dents’ social environment affect their self-reported preferences for various job attributes,
particularly those that have been found to be relevant to gender wage disparities. The
following section will then explore the social learning mechanism, drawing on novel data
on students’ beliefs collected through the original survey.

Endogeneous preferences. This analysis leverages data from the institutional survey
administered to all graduating students, with a near 100% response rate due to the sur-
vey’s compulsory nature. This comprehensive dataset offers a significant advantage over
previous studies, which relied on smaller, less representative samples from select fields
and universities. In the survey, students’ preferences for job attributes are directly elicited
by asking them to rank various attributes on a scale from 1 to 5. Building on prior re-
search into gender differences in job-search preferences (Wiswall and Zafar 2018, Mas and
Pallais 2017, Eriksson and Kristensen 2014), I construct indexes to measure preferences
for pecuniary aspects (i.e. salary and career progression) and flexibility (i.e. leisure time
and hours flexibility). These indices are calculated as unweighted averages of the scores
assigned to each job characteristic and are standardized for ease of interpretation. Addi-
tionally, I create a binary indicator to identify whether a student assigns the maximum
value (5/5) to the job’s social utility—a relevant attribute that has been overlooked in
previous studies but where I observe significant gender differences. While prior research
has typically treated these preferences as exogenous, I examine whether, and to what
extent, these preferences are endogenous to the social environment students encounter in
college. Accordingly, I use these measures as outcome variables in the empirical model,
with the results presented in Table A.29. The findings suggest that women socialized in
cohorts with more female peers from high-FLFP provinces tend to place less importance
on non-pecuniary job factors, particularly those related to flexibility and social utility.
Specifically, a one standard deviation increase in peers’ gender culture decreases women’s

52
preferences for hours flexibility by 2.7% of a standard deviation. In contrast, there is
no statistically significant impact on women’s preferences for pecuniary aspects. These
estimated effects on students’ self-reported preferences align with cultural transmission
explanations, indicating potential shifts in women’s job preferences.

10 Original survey on students’ beliefs

The primary objective of this survey is to complement existing data sources and gain a
deeper understanding of the mechanisms of peer influence, with a focus on social learning.
Specifically, the survey has been designed in a way to (1) shed light on possible asymme-
tries in women’s beliefs depending on the gender culture in their province of birth, (2)
test whether beliefs are relevant in their job acceptance decisions, and (3) explore beliefs’
updating.

Gender culture and women’s beliefs. Since the seminal contribution of Fernandez 2007,
numerous papers have argued that the gender culture in a woman’s country/region
of origin or ancenstry shapes her beliefs, thereby influencing her labor market choices
(Boelmann, Raute, and Schönberg 2023, Kleven 2024, Ichino et al. 2024, Fernandez 2013,
Fogli and Veldkamp 2011, Fernandez and Fogli 2009). While previous studies propose
hypotheses about which beliefs are influenced by the environment—such as beliefs regard-
ing gender identity or the perceived costs of working - all of these studies are essentially
agnostic regarding the precise sources of gender norms. Conceptually, in this context, we
can think of the gender culture in a woman’s province of birth as affecting a wide array
of beliefs: for example, beliefs about the role of women in society, perceptions regarding
employers’ discrimination, beliefs on the job offer distribution, as well as expectations
about long-run outcomes, such as fertility and expectations of future labor supply at the
time of motherhood. All of them, in turn, could influence the labour supply decisions of
young women and, particularly, the acceptance of part-time job offers. I have designed
this survey to elicit women’s beliefs regarding these various aspects. In this section, I
will first present evidence of disparities in beliefs stemming from local gender culture
and show how these differences translate into acceptances of part-time jobs. I will offer a
theoretical illustration of the mechanisms by which these beliefs affect women’s decisions

53
to accept part-time jobs. Finally, leveraging data from two survey waves, I will present
evidence of beliefs’ updating.

10.1 Survey Design and Administration

I have conducted the survey among graduate students currently enrolled at a large public
university in Italy (the University of Bologna). This represents the largest university in
Italy, contributing to approximately 7% of all graduates. Importantly, it offers a multitude
of cultural backgrounds, as it attracts a significant number of students from various
provinces and regions across the country (88.8% and 69.6%, respectively, in the sample, as
shown in Table A.30).
To construct a sample of analysis, I have randomly selected a sample of Master’s degree
programs and, within each program, I have randomly chosen one course from the first
semester in the first year and one from the first semester of the second year. Students
attending these courses have been invited to take part in the survey. The administration
was done in person. Specifically, upon agreement with lecturers, I went in person to one
class - usually in the first/last 15 minutes - and I encouraged students to voluntarily
complete a 10-minute questionnaire on their mobile phones through the SurveyMonkey
platform. Before, I took some minutes to provide general information on the study18. To
incentivize participation, students had the chance to enter three lotteries with gift cards
worth €100 19. The response rate reached 97% among attending students. To have a sense of
the representativeness of this sample, consider that around 77% of students attend classes
regularly, based on self-reported attendance in the AlmaLaurea questionnaire. Students
were not informed in advance about my intervention to ensure that their attendance in
class would be orthogonal to the survey administration. These two features attenuate
concerns related to selection. The survey was conducted between November 2023 and
February 2024. I chose to run the survey 3-4 months after the start of the academic
year to strike a balance between students being able to give informed responses to the
questions—especially about the program’s social environment— and learning of students

18. Specifically, I informed students that the questionnaire was about their beliefs and labor market expec-
tations and was needed for a study on students’ career decisions after college. To avoid priming, I did not
disclose that the study focused on peer influence or its connection to gender inequalities.
19. The gift cards were generic and could be used across multiple brands or providers, to avoid any potential
selection bias from the choice of a specific provider.

54
in the first year being not yet complete. A total of 899 students from 34 Master’s programs
participated in the survey. Among them, 535 identified as women, 348 as men, and 13 as
non-binary. The sample included 571 students in their first year and 322 in their second
year. This disparity is attributed to the curriculum structure, with mandatory courses
mainly offered in the first year. Consequently, the second-year cohort tends to be smaller
due to the greater flexibility in choosing optional courses.

10.2 Sample Selection and Description

I exclude from the sample students who are on Erasmus or attending a bachelor’s program
(less than 1%), as well as international students or students with missing information on
the country/province of origin (5.8%). The number of female students in the resulting
sample is 490, 319 of whom are in the first year of the Master’s program and 171 in the
second year. The characteristics of this sample are summarized in Table A.30, which
provides summary statistics for the entire sample and for sub-samples based on whether
students come from provinces with below-median or above-median FLFP. This sample
differs from the main one in a few dimensions: it over-represents students from high-SES
backgrounds, indicated by a higher proportion of parents with university degrees, and
includes more students from provinces with higher FLFP. There is also a higher percentage
of students who migrated from other provinces and regions. While the distribution of
fields of study is generally similar to the main sample—particularly in the representation
of students in economics and the social sciences—this sample features a slightly higher
representation of students in the humanities and sciences, with fields such as architecture
and healthcare not represented. When comparing the characteristics of students by FLFP
in their province of origin, it is immediate to observe that these groups differ in the role
models they are exposed to, both in the family and in society. Students from low-FLFP
provinces are more likely to come from families where fathers have higher education levels
than mothers, while the opposite is true for students from high-FLFP areas. Additionally,
mothers from low-FLFP areas are more likely to have experienced significant child penalties,
such as pausing their careers during early childhood, compared to mothers from high-
FLFP regions. Apart from these background characteristics, the two groups are similar in
two key areas: women’s fertility expectations and their intentions to search for a job after

55
graduation.

10.3 Learning about The Job Offer Distribution

In this sub-section, I investigate the role of asymmetries in beliefs regarding the job offer
distribution and the process of learning. The underlying idea is that women from areas
with less favorable labor market conditions for women may hold different expectations
about job offer arrival rates compared to those from more egalitarian areas. To investigate
this, I gather students’ expectations regarding key parameters of a job-search model,
including the overall job offer arrival rate and the relative arrival rates of part-time versus
full-time job offers.
Elicitation. Students’ beliefs about the job offer distribution have been elicited through
hypothetical scenarios that aim at reproducing a realistic setting of job search:

1. Consider the following scenario: you have graduated from the Master’s program in which you are
currently enrolled and you start searching for a job. You submit 10 applications to positions aligned
with your field of study. When applying, you don’t know the specific working conditions—such as
the monthly salary or whether the contract is part-time or full-timea.
Out of these 10 applications, how many job offers do you expect to receive? (α) Provide your answer
on a scale from 0 to 10.

You receive your first job offer. What do you believe is the probability that the employer will propose
a part-time contract (less than 28 hours/week)? (γ) - Provide your answer on a scale from 0 to 100.

On a scale from 1 (my answer is very close to the true probability) to 5 (I answered at random), how
certain are you about your previous answer? (σ)

2. While waiting for responses to your applications, an employer contacts you and offers a part-time
position (28 hours/week) with a net monthly salary in line with your expectations. You must decide
whether to accept the offer or turn it down and wait for responses from the other applications.
What is the probability that you will accept this part-time job offer? Provide your answer on a scale
from 0 to 100.
Note that in Italy, 91% of online job postings do not include salary or salary ranges, and precise information
about working hours is often limited (Burning glass data)

Note that, similar to Wiswall and Zafar 2021, students’ intentions to accept a part-time
job offer are elicited using a stated probabilities approach rather than a discrete choice

56
approach. This method accounts for students’ uncertainty when reporting their choices
in the survey. A discrete choice model is, in fact, a special case of the stated probabilities
approach, representing a scenario where there is no resolvable uncertainty. In this case,
individuals would assign a probability of exactly 1 or 0 to the decision of accepting a
part-time job. However, the data clearly reject this: only 4.83% of reported probabilities
are 1, and less than 1% of probabilities are 0.
Place of birth and asymmetries in beliefs. The analysis of these beliefs is presented
in Table 9. To explore potential asymmetries based on local gender culture, Panel (a)
focuses on baseline beliefs collected during the students’ first year, aiming to capture
the student’s initial perceptions before any influence from peers. Specifically, the table
shows predictions from a linear regression model where each dependent variable—listed
in the rows—is regressed on an indicator denoting whether a woman originates from a
province with below-median or above-median FLFP, controlling for field fixed effects. The
table provides the predicted values with their standard errors, as well as the p-values
testing the significance of the differences between the two groups. The results indicate
that women tend to have different expectations of these parameters depending on the
FLFP in their province of origin. First, women from low-FLFP areas expect a slightly
lower arrival rate of job offers (α), though this difference is negligible and not statistically
significant—out of ten applications, they expect to receive, on average, 3.21 offers com-
pared to 3.52 for women from high-FLFP provinces. More notably, there is a substantial
gap in their expectations regarding the proportion of part-time job offers. Women from
low-FLFP areas expect a 6.45 percentage point higher likelihood of receiving a part-time
versus full-time job offer, which represents a significant 12.6% increase compared to their
high-FLFP counterparts. Additionally, these women are 7 percentage points more likely to
indicate they would accept a part-time offer, marking a 12% increase relative to peers from
high-FLFP provinces. Importantly, these differences cannot be attributed to differences in
observed characteristics between the two groups, as they are unchanged after controlling
for students’ background characteristics (age, family background), job search intentions,
and expected job location, as shown in Table A.31.

Beliefs’ updating. Panel (b) of Table 9 investigates how these believes evolve over time,

57
 TABLE 9. Baseline and Updated Beliefs on the Job Offer Distribution

Below-med FLFP Above-med FLFP

Pred SE Pred SE P-value
a. Baseline Beliefs (T=0)
α: Expected arrival rate of job offers (%) 32.06 1.77 35.24 1.24 0.15
γ: Expected % of part-time job offers 57.64 2.29 51.19 1.61 0.02
Perceived uncertainty (1-5) 2.70 0.12 3.01 0.08 0.04
Prob. to accept part-time job offer 67.43 2.04 60.39 1.44 0.01

b. Updated Beliefs (T=1)

α: Expected arrival rate of job offers (%) 32.20 2.17 32.19 1.73 1.00
γ: Expected % of part-time job offers 52.47 2.90 50.70 2.31 0.64
Perceived uncertainty (1-5) 2.73 0.14 2.75 0.11 0.90
Prob. to accept part-time job offer 62.48 2.81 62.37 2.23 0.98
Notes: This table presents predictions from a linear regression model, where the dependent
variable is regressed on an indicator for whether the FLFP in the birth province is above
or below the median, along with fixed effects for the field of study. Each row represents
a different regression, with the dependent variable specified in Column 1. For each
regression, the table reports the predicted dependent variable for women from provinces
with low versus high FLFP, along with the standard errors. The last column provides the
p-value for the difference between these two groups. In Panel (a), the sample consists of
all first-year female Master’s students (319), and in Panel (b), it includes all second-year
female Master’s students (164). Between 60% and 65% of the students are from provinces
with above-median FLFP.

by focusing on answers from students in the second year. Overall, we observe conver-
gence in these beliefs between the two groups, consistent with learning. More precisely,
the results show that students significantly update their beliefs about the likelihood of
receiving a part-time job offer, in a way that the gap originally observed between the two
groups has narrowed considerably (by more than 70%). An analysis of the variance in
beliefs within fields further supports evidence of learning. On average across degrees,
the standard deviation of students’ baseline beliefs γ in the first year is 24.28, which
decreases by more than a third in the second year—a statistically significant reduction at
the 1% level. What is particularly interesting is the asymmetry in this learning process.
Women born in low-FLFP provinces experience strong beliefs’ updating and revise their
beliefs downwards regarding the probability of receiving a part-time offer relative to a

58
full-time one by more than 5 percentage points (a 9% decrease), converging to the values
expressed by their peers from high-FLFP areas who experience only little updating. Why
do the two groups update their beliefs differently? One possible explanation is the initial
asymmetry in the information available to them. Women from low-FLFP areas might
have started with more biased beliefs about job offer arrival rates in their destination
labor markets, leading to a more significant adjustment over time. This is a plausible
channel, as women from low-FLFP areas are typically exposed to labor markets that differ
substantially, in terms of women’s outcomes, than those in their home regions, as reflected
in their intentions to work predominantly in the North of Italy (Table A.30). These results
provide evidence of asymmetric beliefs’ updating, which is consistent with the asymmetry
in the estimated peer effects. While I cannot precisely quantify the contribution of peers
versus other social influences in the process of beliefs’ updating, the results suggest social
learning as a plausible mechanism driving peer influence in this context.

10.4 A Model of Job Search

To study the relation between beliefs and job search, I propose a McCall type model
(McCall 1970) where risk-neutral female graduates search for their first post-graduation
job. For the time being, I abstract from students’ gender norms when I lay out the model,
and I later introduce parameter heterogeneity when I discuss the model’s prediction for
differences in part-time acceptances between women from high- and low-FLFP areas.

10.5 Model Setup

My modeling framework is based on a standard model of labor market search à la McCall,

augmented to allow for heterogeneous worker beliefs.
Consider an economy where three states exists: an individual can be unemployed, em-
ployed in a part-time job or employed in a full-time job. For simplicity, consider that
part-time (P) and full-time jobs (F) are characterized by a fixed number of weekly hours.
I express the per-period number of hours in a part-time job as hP = θhF , with θ < 1. The
model makes a number of key assumptions. Time t is discrete. All individuals discount the
future at rate β ∈ (0, 1). Students are risk-neutral: they have preferences over consumption
represented by the instantaneous utility function u(c) = y, i.e. they maximize expected

59
lifetime labor income. As is typical in job-search models, this model abstracts from other
sources of income, so that instantaneous income is given by the following specification:

⎧
⎪ yF if employed in full-time job
⎪
⎪
y = ⎨y P if employed in part-time job
⎪
⎪b if unemployed
⎩

b represents any income associated with not working, such as the pecuniary value of
leisure and public unemployment insurance (UI) benefits. yP and y F are total per period
income associated with either a part-time or a full-time job. For simplicity, I fix the number
of hours in a full-time job to unity, so that y F = w and yP = θw.
Unemployed jobseekers search for jobs and, in each period, job offers arrive with prob-
ability α∗. A share γ ∗ of job offers are part-time. A job offer is a random draw from a
wage distribution F(w), which has support and non-zero density on [wmin, w̄ ]. Note that,
for simplicity, I assume that the wage distribution is the same for part-time and full-time
jobs. In each period, if an unemployed worker receives a job offer, she decides whether to
accept the offer and leave unemployment or remain unemployed and enjoy the value of
leisure b. I do not allow for on-the-job search or job destruction, meaning that employment
- both part-time and full-time - is an absorbing state. I further assume that the environment
is stable, i.e. that the arrival rates of full-time and part-time job offers do not change over
the course of the search spell. Individuals are infinitely lived and, therefore, the model is
stationary. Throughout, I use ∗ to indicate “true” or “actual” probabilities of receiving job
offers, to distinguish these from the workers’ beliefs.

10.6 Workers’ Beliefs

I start with the notion that workers make decisions with possibly limited knowledge
about job offer arrival rates. Specifically, I assume that workers do not necessarily know
the per-period probability of receiving a job offer α∗ and the relative share of part-time
job offers γ ∗. Define αt and γt as the worker’s current beliefs about α∗ and γ∗. I refer
to biased beliefs if αt ≠ α ∗ or γt ≠ γ ∗. While beliefs potentially evolve over time due to
learning, in this version of the model I abstract from this possibility and assume that α

60
and γ are not time-varying. Workers take their decisions on whether to accept job offers
based on their subjective beliefs α and γ. I abstract away from other potential biases in
beliefs, for example on the wage offer distribution F(w), that I assume to be commonly
known to all individuals.

10.7 Perceived Values of Employment and Unemployment

I characterize the perceived flow values of unemployment and employment 20. For a
worker with beliefs α and γ, the perceived value of unemployment equals

(7)

U(α, γ) = b + βα[γ ∫ max{V (yP), U(α, γ)} dG(yP) + (1 − γ) ∫ max{V (yF ), U(α, γ)} dG(yF )]
yP yF

+β(1 − α)U(α, γ)

where b is the flow value of unemployment, and α and γ are the worker’s beliefs regarding
the per-period probability of receiving a job offer and the relative share of part-time job
offers. The values of part-time and full-time employment at wage w are respectively

(8) V (y P ) = y P + βV (y P ) → (1 − β)V (y P ) = θw

(9) V (yF ) = yF + βV (yF ) → (1 − β)V (yF ) = w

Note that here, absent job destruction and on-the-job search, it is assumed that, once a
worker has accepted a job offer, she remains at her current job at all future periods.

10.8 Reservation Wages

A job-seeker’s decision to accept a job offer is determined by the reservation wage prop-
erty: each job offering wages above the reservation value are accepted. The job seeker
determines their reservation wage in order to maximize their perceived continuation
value at any point during the search spell. I define the reservation earnings, R(α, γ), as
the total per-period income at which a job seeker is indifferent between accepting a job
20. I refer to them as perceived as they are based on workers’ beliefs about the arrival rates of part-time and
full-time job offers instead of the actual ones.

61
and remaining unemployed. The resulting expression for the reservation earnings equals

(10) V (R(α, γ)) − U(α, γ) = 0 → R(α, γ) = (1 − β)U(α, γ)

Note that, because part-time and full-time jobs differ in the number of working hours, the
reservation earnings condition expressed above is verified for two different reservation
wages, separately for the two job types. I define the reservation wages for part-time and

full-time jobs as wR,P(α, γ) and wR,F (α, γ). The expressions are

R(α, γ)
(11) wR,P(α, γ) = and wR,F (α, γ) = R(α, γ)
θ

Any job offering wages above these values is accepted. Note that reservation wages are
determined based on workers’ beliefs. In the following propositions, I outline how biases
in beliefs regarding α∗ and γ∗ theoretically impact reservation wages.

Proposition 1. Ceteris paribus, reservation wages are increasing in beliefs α.

Proposition 2. Ceteris paribus, reservation wages are decreasing in beliefs γ.

The proofs are contained in Appendix Section A.

10.9 Heterogeneous Beliefs and Model’s Predictions

I now introduce heterogeneity in workers’ beliefs into the model, considering that workers
hold beliefs αi and γi about the arrival rates of job offers, where i ∈ (L, H). However,
throughout the model, I assume that all workers actually face the same true arrival
rates, denoted as (α ∗, γ ∗). I define the beliefs of women from low-FLFP and high-FLFP
provinces as (αL, γL) and (α H , γ H ), respectively. Drawing on the empirical analysis of
students’ beliefs from the previous section, I assume:

(12) αL < αH and γL > γH

i.e. women from low-FLFP areas expect a lower probability of receiving any job offer and
a higher likelihood of receiving a part-time offer relative to a full-time one.

62
A direct corollary of Propositions 1 and 2 is that, all else equal, if women from low-
FLFP provinces hold more pessimistic beliefs about the arrival rates of job offers and
the likelihood of receiving part-time versus full-time offers compared to women from
high-FLFP areas, they will have lower reservation earnings. Consequently, they will have
a higher likelihood of accepting a part-time job offer. I test this prediction using data
on students’ expectations about job offer arrival rates and their intentions to accept a
part-time offer, elicited on a probabilistic scale through a realistic hypothetical scenario.
These intentions are indicative of how students might behave when faced with part-time
job offers during their actual job search. Figure 7 illustrates the relationship between these
beliefs and students’ job search behavior, showing a strong correlation between their
expectations of part-time versus full-time job offers and the likelihood of accepting a part-
time offer. Estimates from a simple linear regression indicate that a one standard deviation
increase (23 percentage points) in the expected probability of receiving a part-time offer
results in an approximately 8 percentage point increase in the acceptance rate of part-time
jobs, representing more than a third of the standard deviation in the sample. These beliefs
alone explain 13% of the variation in students’ acceptances of part-time jobs in the sample.
Estimates are robust to the inclusion of field (or degree) fixed effects (β = 0.27). Table A.32
shows the estimated coefficients from a regression of students’ part-time job acceptances
on their beliefs about γ (Columns 1-2) and α (Columns 3-4), both with and without
controls for field fixed effects. These results additionally show a negative relationship
between expected job offer arrival rates and the likelihood of accepting a part-time offer,
consistent with the model’s predictions.

Job finding probabilities. The individual job-finding probability is defined as:

(13) λi = α∗[γ∗P(yP ≥ Ri(αi, γi)) + (1 − γ∗)P(yF ≥ Ri(αi, γi))]

that I rewrite as

∗ ∗ Ri (αi, γi) ∗
(14) λi = α [γ (1 − F( )) + (1 − γ )(1 − F(Ri(αi, γi)))]
θ

where λi represents the per-period probability of exiting unemployment. This probability

63
 FIGURE 7. Acceptance of part-time jobs and expected share of part-time job offers

Notes. This figure presents binned scatter plots of the probability of accepting a part-time job against the
expected arrival rate of part-time relative to full-time job offers (γ). One observation represents a student in
the sample. The figure also displays the estimated coefficient β and its standard error from a simple linear
regression of the intended probability of accepting a part-time offer on the expected percentage of part-time
job offers.

depends on the true arrival rates of job offers, α ∗ and γ ∗, as well as on women’s beliefs
about these parameters through their reservation earnings, Ri(αi, γi), which are indexed
by i to reflect heterogeneity in workers’ beliefs. A second implication of equation (12) in the
model is that λL > λH at any point in time, implying that women from low-FLFP areas have
higher job-finding rates due to behavioral differences driven by their beliefs. Specifically,
since women with more pessimistic beliefs are less selective and have lower reservation
earnings, they are less likely to reject job offers and more likely to exit unemployment
earlier. It is important to note that this result relies on a simplifying assumption of the
model—that job search effort is exogenously determined. An extended version of the
model, which includes endogenous job search effort (available in the appendix), can
rationalize the dynamics of job-search behavior observed in the data.

10.10 Expectations of Fertility and Future Child Penalties

This sub-section explores how local gender culture shapes expectations about fertility and
anticipated child penalties. Previous studies, such as Boelmann, Raute, and Schönberg
2023 and Kleven 2024, have shown that the gender norms women are exposed to during

64
childhood have long-lasting effects on their labor supply decisions after becoming mothers.
Even in the Italian context, the magnitude of child penalties varies significantly across
geographical areas (Casarico and Lattanzio 2023). If women accurately anticipate these
differences in future behavior, their expectations about fertility and future labor supply
may vary based on the gender norms they were exposed to when growing up. For
instance, women from less egalitarian backgrounds might expect higher employment
costs associated with motherhood. As implied by a dynamic labor supply model in the vein
of Adda, Dustmann, and Stevens 2017, these differences in expectations likely influence
women’s career choices even before they have children, particularly in selecting jobs or
occupations with different opportunity costs of child-rearing. Whether women anticipate
these differences in behavior is an empirical question, that I examine in this sub-section. To
achieve this goal, I gathered data on women’s expectations regarding fertility and future
labor supply.

1. Would you like to have children in the future? Yes/No/Don’t know/Already

have
2. At what age do you expect to have your first child?
3. Expected labor supply at motherhood:
• Scenario 1. Suppose that your partner is earning enough to support your family. What do
you think you will do when your child is young (0-2 years)? Answer: No work/Work
part-time/Work full-time
• Scenario 2. Suppose that your partner is earning enough to support your family and that
in the area you live a full-day place in childcare is available to you. What do you think you
will do when your child is young (0-2 years)? Answer: No work/Work part-time/Work
full-time

The analysis of their baseline and updated expectations is presented in Table A.33.
At the start of the first year, there are no differences in fertility expectations that can
explain the observed labor supply disparities between women from high and low-FLFP
provinces. In fact, women from low-FLFP provinces are less likely to expect to have
children and anticipate having their first child at a later age compared to their high-FLFP

65
peers. Crucially, when considering future labor supply, those from low-FLFP areas are
less likely to foresee reductions in working hours due to motherhood as they are more
likely to expect to continue working full-time during the early years of parenthood. This
pattern holds across both their unconditional expectations (elicited under scenario 1) and
their expectations when full-day childcare is available near their residence (elicited under
scenario 2). Additionally, both groups of women anticipate that childcare availability
will impact their future labor supply, as the share expecting to work full-time increases
by 43%-58%, depending on the group, when access to full-day childcare is available.
One year later, both groups have revised their expectations upwards, particularly when
full-day childcare is available. Around 80% of women in the two groups expect to work
full-time in the early years of motherhood. Contrary to expectations, these results show
that women from less egalitarian backgrounds do not anticipate higher employment costs
of motherhood when entering the labor market, compared to women from high-FLFP
areas. One plausible explanation is that young women may underestimate the career costs
of motherhood— a phenomenon particularly pronounced among the college-educated
(Kuziemko et al. 2018). Other sets of explanations relate to intergenerational shifts in
the magnitude of child penalties or positive selection bias in the sample of women from
low-FLFP areas. Importantly, unlike the asymmetric updating of beliefs regarding job
offers, these results indicate that preferences and beliefs about maternal employment
evolve symmetrically across both groups.

11 Conclusions

66
References
Acciari, P., A. Polo, and G. Violante. 2021. “And Yet It Moves: Intergenerational Mobility in Italy.”
American Economic Journal: Applied Economics 14(3):118–163.

Adda, J., C. Dustmann, and K. Stevens. 2017. “The Career Costs of Children.” Journal of Political
Economy 125(2):293–337.

Alesina, A., P. Giuliano, and N. Nunn. 2013. “ON THE ORIGINS OF GENDER ROLES: WOMEN
AND THE PLOUGH.” The Quarterly Journal of Economics 128(2):469–530.

Alfonsi, L., M. Namubiru, and S. Spaziani. 2024. “Meet Your Future: Experimental Evidence on
the Labor Market Eects of Mentors.” WP.

Altonji, J. G., and R. M. Blank. 1999. “Race and Gender in the Labor Market.” Handbook of Labor
Economics 3(C):3143–3259.

Angelov, N., P. Johansson, and E. Lindahl. 2016. “Parenthood and the Gender Gap in Pay.” Journal
of Labor Economics 34(3).

Angrist, J. 2014. “The perils of peer effects.” Labour Economics 30:98–108.

Angrist, J., and K. Lang. 2004. “Does School Integration Generate Peer Effects? Evidence from
Boston’s Metco Program.” American Economic Review 94(5):1613–1634.

ANVUR. 2023. “Rapporto 2023 sul Sistema della formazione superiore e della ricerca.” Tech. report.

Azmat, G., and R. Ferrer. 2017. “Gender Gaps in Performance: Evidence from Young Lawyers.”
Journal of Political Economy 125(5):1306–1355.

Bandiera, O., V. Bassi, R. Burgess, I. Rasul, M. Sulaiman, and A. Vitali. 2023. “The search for
good jobs: Evidence from a six-year field experiment in uganda.” Journal of Labor Economics.

Barrios Fernandez, A. 2023. “Peer Effects in Education.” Oxford Research Encyclopedia of Economics
and Finance.

Bertrand, M. 2020. “Richard T. Ely Lecture – Gender in the 21st Century.” American Economic
Association Papers and Proceedings 110(5):1–24.

Bertrand, M., P. Cortes, C. Olivetti, and J. Pan. 2021. “Social Norms, Labor Market Opportunities,
and the Marriage Gap Between Skilled and Unskilled Women.” Review of Economic Studies
88(4):1936–1978.

67
Bertrand, M., C. Goldin, and L. F. Katz. 2010. “Dynamics of the Gender Gap for Young Profes-
sionals in the Financial and Corporate Sectors.” American Economic Journal: Applied Economics
2(3):228–255.

Bertrand, M., E. Kamenica, and J. Pan. 2015. “Gender Identity and Relative Income Within
Households.” Quarterly Journal of Economics 130(2):571–614.

Bifulco, R., J. M. Fletcher, and S. L. Ross. 2011. “The Effect of Classmate Characteristics on Post-
secondary Outcomes: Evidence from the Add Health.” AMERICAN ECONOMIC JOURNAL:
ECONOMIC POLICY 3(1):25–53.

Bisin, A., and T. Verdier. 2000. “BEYOND THE MELTING POT: CULTURAL TRANSMISSION,
MARRIAGE, AND THE EVOLUTION OF ETHNIC AND RELIGIOUS TRAITS.” Quarterly
Journal of Economics CXV(3):955–988.

. 2001. “The Economics of Cultural Transmission and the Dynamics of Preferences.” Journal
of Economic Theory 97(2):298–319.

Black, S. E., P. J. Devereux, and K. G. Salvanes. 2013. “Under Pressure? The Effect of Peers on
Outcomes of Young Adults.” Journal of Labor Economics 31(1).

Blau, F. D., and L. M. Kahn. 2017. “The Gender Wage Gap: Extent, Trends, and Explanations.”
Journal of Economic Literature 55(3):789–865.

Blau, F. D., L. M. Kahn, and K. L. Papps. 2011. “Gender, Source Country Characteristics and
Labor Market Assimilation among Immigrants: 1980-2000.” Review of Economics and Statistics
93(1):43–58.

Boelman, B., A. Raute, and U. Schönberg. Forthcoming. “Wind of Change? Cultural Determinants
of Maternal Labor Supply.” American Economic Journal: Applied Economics.

Boelmann, B., A. Raute, and U. Schönberg. 2023. “Wind of Change? Cultural Determinants of
Maternal Labor Supply.” CEPR Discussion Paper No. 16149 137.

Booij, A.S., and E. Leuven. 2017. “Ability Peer Effects in University: Evidence from a Randomized
Experiment.” Review of Economic Studies 84:547–578.

Boucher, V., M. Rendall, P. Ushchev, and Y. Zenou. 2024. “Toward a General Theory of Peer
Effects.” Econometrica 92(2):543–565.

Bovini, G., M. De Philippis, and L. Rizzica. 2023. “The Origins of the Gender Pay Gap: The Role
of School-to-Work Transition.” Temi di discussione, Rome: Bank of Italy.

68
Braccioli, F., P. Ghinetti, S. Moriconi, C. Naguib, and M. Pellizzari. 2023. “Education Expansion,
College Choice and Labour Market Success.” CESifo Working Paper No. 10842.

Brandimarti, E. 2023. “Self-Selection, University Courses and Returns to Advanced Degrees.”

Unpublished manuscript.

Brenøe, A. A., and U. Zölitz. 2020. “Exposure to More Female Peers Widens the Gender Gap in
STEM Participation.” Journal of Labor Economics (38)4.

Burke, M., and T. Sass. 2013. “Classroom Peer Effects and Student Achievement.” Journal of Labor
Economics 31(1):51–82.

Bursztyn, L., A.W. Cappelen, B. Tungodden, A. Voena, and D. Yanagizawa-Drott. 2023. “How
Are Gender Norms Perceived?” NBER WORKING PAPER 31049.

Bursztyn, L., T. Fujiwara, and A. Pallais. 2017. “’Acting Wife:’ Marriage Market Incentives and
Labor Market Investments.” American Economic Review 107(11):3288–3319.

Caliendo, M., D. A. Cobb-Clark, and A. Uhlendorff. 2015. “Locus of Control and Job Search
Strategies.” Review of Economics and Statistics 97(1):88–103.

Campa, P., A. Casarico, and P. Profeta. 2011. “Gender Culture and Gender Gap in Employment.”
CESifo Economic Studies 57(1):156–182.

Carlana, M. 2019. “Implicit Stereotypes: Evidence from Teachers’ Gender Bias.” The Quarterly
Journal of Economics 134(3):1163–1224.

Carrell, S. E., M. Hoekstra, and E. Kuka. 2018. “The long-run effects of disruptive peers.” American
Economic Review 108(11):3377–3415.

Carrell, S. E., B. I. Sacerdote, and J. E. West. 2013. “FROM NATURAL VARIATION TO OP-
TIMAL POLICY? THE IMPORTANCE OF ENDOGENOUS PEER GROUP FORMATION.”
Econometrica 81(3):855–882.

Carrell, S.E., R.L. Fullerton, and J.E. West. 2009. “Does Your Cohort Matter? Measuring Peer
Effects in College Achievement.” Journal of Labor Economics 27(3):439–464.

Carrer, L., and L. de Masi. 2024. “Municipal-level Gender Norms: Measurement and Effects on
Women in Politics.”

Casarico, A., and S. Lattanzio. 2023. “Behind the child penalty: understanding what contributes to
the labour market costs of motherhood.” Journal of Population Economics 36:1489–1511.

69
Cattan, S., K. G. Salvanes, and E. Tominey. 2022. “First Generation Elite: The Role of School
Networks.” IZA Discussion Paper No. 15560.

Cavalli-Sforza, L., and M. Feldman. 1981. “Cultural Transmission and Evolution: A Quantitative
Approach.” (Princeton, NJ: Princeton University Press).

Commission/EACEA/Eurydice, European. 2016. “National Student Fee and Support Systems in

European Higher Education – 2016/17. Eurydice Facts and Figures.” Luxembourg: Publications
Office of the European Union.

Conlon, J. J., L. Pilossoph, M. Wiswall, and B. Zafar. 2018. “Labor Market Search With Imperfect
Information and Learning.” NBER Working Paper.

Cools, A., R. Fernandez, and E. Patacchini. 2022. “The Asymmetric Gender Effects of High Flyers.”
Labour Economics 79.

Cortes, P., and J. Pan. 2019. “When Time Binds: Substitutes to Household Production, Returns to
Working Long Hours and the Gender Wage Gap among the Highly Skilled.” Journal of Labor
Economics 37(2):351–398.

. 2023. “Children and the Remaining Gender Gaps in the Labor Market.” Journal of Economic
Literature 61(4):1359–1409.

Cortes, P., L. Pilossoph, E. Reuben, and B. Zafar. 2023. “Gender Differences in Job Search and
the Earnings Gap: Evidence from the Field and the Lab.” Quarterly Journal of Economics 138
(4):2069–2126.

Cortés, P., G. Kos¸ar, J. Pan, and Z Basit. 2022. “Should Mothers Work? How Perceptions of the
Social Norm Affect Individual Attitudes Toward Work in the U.S.” WORKING PAPER 30606.

Cortés, P., G. Kos¸ar, J. Pan, and B. Zafar. 2022. “Should Mothers Work? How Perceptions of the
Social Norm Affect Individual Attitudes Toward Work in the U.S.” NBER Working Paper No.
w30606.

Currarini, S., M. O. Jackson, and P. Pin. 2009. “An economic model of friendship: Homophily,
minorities, and segregation.” Econometrica 77(4):1003–1045.

Einiö, E. 2023. “The Long-Term Impacts of Mixing the Rich and Poor: Evidence from Conscript
Dorms.” Working Paper.

Eriksson, T., and N. Kristensen. 2014. “Wages or Fringes? Some Evidence on Trade-offs and
Sorting.” Journal of Labor Economics 32(4):899–928.

70
EVS. 1990-2008. “European Values Study Longitudinal Data File 1981-2008.” GESIS Data Archive,
Cologne.

Feld, J., and U. Zölitz. 2017. “Understanding Peer Effects: On the Nature, Estimation, and Channels
of Peer Effects.” Journal of Labor Economics 35(2).

Fernandez, R. 2007. “Women, Work, and Culture.” Journal of the European Economic Association
5(2-3):305–332.

. 2013. “Cultural Change as Learning: The Evolution of Female Labor Force Participation
over a Century.” American Economic Review 103 (1):472–50.

Fernandez, R., and A. Fogli. 2006. “Fertility: The Role of Culture and Family Experience.” Journal
of the European Economic Association 4(2-3):552–561.

. 2009. “Culture: An Empirical Investigation of Beliefs, Work, and Fertility.” American

Economic Journal: Macroeconomics 1 (1):146–177.

Fernández, R., A. Fogli, and C. Olivetti. 2004. “Mothers and Sons: Preference Formation and
Female Labor Force Dynamics.” The Quarterly Journal of Economics 119(4):1249–1299.

Fischer, A., T. M. Sandoy, and J. Walldorf. 2023. “Peers, Heirs, and Careers: Labor Market Effects
of Alumni Networks.” Working Paper.

Flabbi, L., and A. Moro. 2012. “The effect of job flexibility on female labor market outcomes:
Estimates from a search and bargaining model.” Journal of Econometrics 168(1):81–95.

Fogli, A., and L. Veldkamp. 2011. “NATURE OR NURTURE? LEARNING AND THE GEOGRA-
PHY OF FEMALE LABOR FORCE PARTICIPATION.” Econometrica 79 (4):1103–1138.

Fortin, N. M. 2015. “Gender Role Attitudes and Women’s Labor Market Participation: Opting-
Out, AIDS, and the Persistent Appeal of Housewifery.” Annals of Economics and Statistics
117/118:379–401.

Hampole, M., and A. Wong. 2024. “Peer Effects and the Gender Gap in Corporate Leadership:
Evidence from MBA Students.” Working Paper.

Hanushek, E. A., and S. G. Rivkin. 2009. “Harming the Best: How Schools Affect the Black-White
Achievement Gap.” Journal of Policy Analysis and Management 29(3):366–393.

Hoxby, C. M. 2000. “The Effects of Class Size on Student Achievement: New Evidence from
Population Variation.” The Quarterly Journal of Economics 115(4):1239–85.

71
Hoxby, C.M., and G. Weingarth. 2005. “Taking race out of the equation: School reassignment and
the structure of peer effects.” Working paper.

Ichino, A., M. Olsson, B. Petrongolo, and P. Skogman-Thoursie. 2024. “Economic incentives,

childcare and gender identity norms.” Journal of Labor Economics.

Imberman, S. A., A. D. Kugler, and B. I. Sacerdote. 2012. “Katrina’s Children: Evidence on the
Structure of Peer Effects from Hurricane Evacuees.” American Economic Review 102(5):2048–82.

ISTAT. 2016. “Iscritti all’università.” http://dati.istat.it/Index.aspx?DataSetCode=DCISISCRITTI.

Jäger, S., C. Roth, N. Roussille, and B. Schoefer. 2024. “Worker Beliefs About Outside Options.”
Quarterly Journal of Economics 139(3):1505–1556.

Jessen, J., S. Schmitz, and F. Weinhardt. 2024. “Immigration, Female Labour Supply and Local
Cultural Norms.” The Economic Journal 134(659):1146–1172.

Kerwin, K. C., J. Guryan, and J. Pan. 2024. “The Effects of Sexism on American Women: The Role
of Norms vs. Discrimination.” The Journal of Human Resources 59 (2).

Kleven, H. 2024. “The Geography of Child Penalties and Gender Norms: A Pseudo-Event Study
Approach.” NBER Working Paper 30176.

Kleven, H., C. Landais, and G. Leite-Mariante. Forthcoming. “The Child Penalty Atlas.” The
Review of Economic Studies.

Kleven, H., C. Landais, and J. E. Søgaard. 2019. “Children and Gender Inequality: Evidence from
Denmark.” American Economic Journal: Applied Economics 11 (4):181–209.

Kramarz, F., and O. N. Skans. 2014. “When Strong Ties are Strong: Networks and Youth Labor
Market Entry.” The Review of Economic Studies 81(3):1164–1200.

Kuziemko, I., J. Pan, J. Shen, and E. Washington. 2018. “The Mommy Effect: Do Women Anticipate
the Employment Effects of Motherhood?” NBER Working Papers 24740, National Bureau of
Economic Research, Inc.

Lavy, V., D. Paserman, and A. Schlosser. 2012. “Inside the Black Box of Ability Peer Effects:
Evidence from Variation in the Proportion of Low Achievers in the Classroom.” The Economic
Journal 122(559):208–237.

Lavy, V., and A. Schlosser. 2011. “Mechanisms and Impacts of Gender Peer Effects at School.”
AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS 3(2):1–33.

72
Manski, C.F. 1993. “Identification of Endogenous Social Effects: The Reflection Problem.” The
Review of Economic Studies 60(3):531–542.

Mas, A., and A. Pallais. 2017. “Valuing Alternative Work Arrangements.” American Economic
Review 107(12):3722–3759.

Maurin, E., and J. Moschion. 2009. “The Social Multiplier and Labour Market Participation of
Mothers.” American Economic Journal : Applied Economics 1(1):251–272.

McCall, J. J. 1970. “Economics of Information and Job Search.” Quarterly Journal of Economics
84(1):113–126.

Mertz, M. B., M Ronchi, and V. Salvestrini. 2024. “Female Representation and Talent Allocation
in Entrepreneurship: The Role of Early Exposure to Entrepreneurs.” Working Paper.

Mueller, A. I., and J. Spinnewijn. 2023. “Expectations Data, Labor Market and Job Search.”
Handbook of Economic Expectations, 677–713.

Mueller, A. I., J. Spinnewijn, and G. Topa. 2021. “Job Seekers’ Perceptions and Employ- ment
Prospects: Heterogeneity, Duration Dependence, and Bias.” American Economic Review 111(1):324–
363.

Olivetti, C., E. Patacchini, and Y. Zenou. 2020. “Mothers, Peers and Gender-Role Identity.” Journal
of the European Economic Association 18(1):266–301.

Rattini, V. 2022. “The Effects of Financial Aid on Graduation and Labor Market Outcomes: New
evidence from Matched Education-Labor Data.” CESifo Working Paper No. 10010.

Sacerdote, B. 2011. “Peer Effects in Education: How Might They Work, How Big Are They and
How Much Do We Know Thus Far?” Handbook of the Economics of Education 3:250–77.

. 2014. “Experimental and Quasi-Experimental Analysis of Peer Effects: Two Steps Forward?”
Annual Review of Economics 6:253–72.

Tincani, M. M. 2024. “Heterogeneous Peer Effects and Rank Concerns in the Classroom.” CEPR
Discussion Paper No. 18921.

Wiswall, M., and B. Zafar. 2018. “Preference for the Workplace, Investment in Human Capital,
and Gender.” The Quarterly Journal of Economics 133 (1):457–507.

. 2021. “Human Capital and Expectations about Career and Family.” Journal of Political
Economy 129(5):1361–1424.

73
Zimmerman, S. D. 2019. “Elite Colleges and Upward Mobility to Top Jobs and Top Incomes.”
American Economic Review 109(1):1–47.

74
Appendix
A Appendix: Proofs

Proof of Proposition 1
PROOF. The perceived value of unemployment can be rewritten as

y¯P y¯F
(1 − β)U(α, γ) = b + βα[γ ∫ (V (y P) − U(α, γ)) dG(y ) P+ (1 − γ) ∫ (V (y ) −F U(α, γ)) dG(y )] F
R R

Using the reservation earnings rule and plugging the values of employment, this becomes

α [γ y¯P
(y P – R(α, γ)) dG(y P ) + (1 − γ) ∫
y¯F
(y F – R(α, γ)) dG(y F )]
R(α, γ) = b + β ∫
1−β R R

Rearranging yields

α [γ y¯P
(y P – R(α, γ)) dG(y P ) + (1 − γ) ∫
y¯F
(y F – R(α, γ)) dG(y F )]
R(α, γ) = b + β ∫
1−β R R

reservation earnings are set to equal the flow value of unemployment and the expected
surplus associated with job offers. Note that for all values that R can take, the expected
surplus from a full-time job exceeds the expected surplus from a part-time job.
Rearranging yields

α [ y¯F
(y F – R(α, γ)) dG(y F ) − γ( ∫
y¯F
(y F – R(α, γ)) dG(y F )−
R(α, γ) = b + β ∫
1−β R R
y¯P
∫ (yP – R(α, γ)) dG(y P ))]
R

Substituting yF = w and yP = θw, and using the fact that G(yF ) = F(w) and wR,P = Rθ, I
rewrite the expression that implicitly defines reservation earnings as:

α [ w¯
(w − R(α, γ)) dF(w) − γ( ∫
w¯
(w − R(α, γ)) dF(w)−
R(α, γ) = b + β ∫
(A.1) 1−β R R
w¯
∫ R
(θw − R(α, γ)) dF(w))]
θ

Differentiating both sides of equation (A.1) with respect to α and applying the Leibniz

75
rule yields
w¯ w¯ w¯
∂R(α, γ) β [∫ (w − R(α, γ)) dF(w) − γ( ∫ (w − R(α, γ)) dF(w) − ∫ (θw − R(α, γ)) dF(w))]
=
∂α 1−β R R
R

βα ∂R(α, γ) ∂R(α, γ) θ ∂R(α, γ) R

+ [− (1 − F(R)) + γ (1 − F(R)) − γ (1 − F( ))]
1−β ∂α ∂α ∂α θ

Rearranging, I get to the following expression

β
[ w¯
1−β ∫ R (w − R(α, γ)) dF(w) − γ( ∫ R (w − R(α, γ)) dF(w) − ∫ R (θw − R(α, γ)) dF(w))]
w¯ w¯
∂R(α, γ)
∂γ =
θ
βα
1+ [(1 − F(R))(1 − γ) + γ(1 − F( R ))]
1−β θ

Both the numerator and the denominator of the right-hand side are positive. Hence,
∂R(α,γ)
∂γ
> 0, i.e. reservation earnings are increasing the perceived probability of receiving
a job offer.

Proof of Proposition 2

PROOF. Differentiating both sides of equation (A.1) with respect to γ and applying the
Leibniz rule yields
∂R(α, γ) βα ∂R(α, γ) w¯ w¯
= [ − (1 − F(R)) − (∫ (w − R(α, γ)) dF(w) − ∫ (θw − R(α, γ)) dF(w))
∂γ 1−β ∂γ
R
R θ
∂R(α, γ) ∂R(α, γ) R
−γ( − (1 − F(R)) + (1 − F( )))]
∂γ ∂γ θ

which yields the following expression

( (w − R(α, γ)) dF(w) − (θw − R(α, γ)) dF(w))
w¯ w¯

∂R(α, γ) ∫R ∫R
∂γ = − θ
[1 + βα ((1 − F(R))(1 − γ) + γ(1 − F( R )))]
1−β θ

Because the expected surplus from a full-time always exceeds that of a part-time job, the
numerator is positive. The denominator is also positive. It follows that ∂R(α,γ)
∂γ
< 0, i.e.
reservation earnings are decreasing in a workers’ beliefs of receiving a part-time relative
to a full-time job offer.

B Additional Figures and Tables

76
TABLE A.1. Summary Statistics of Demographics, Performance and Family Background

Female Male
Mean SD Obs Mean SD Obs P-value
Individual characteristics
Age at enrollment 24.3 4.0 182792 24.5 4.1 133678 0.00
GPA during Master 27.8 1.5 182792 27.4 1.7 133678 0.00
Final grade during Master 108.6 5.6 182792 107.4 6.3 133678 0.00
Time to completion of Master (years) 2.5 0.6 182792 2.6 0.6 133678 0.00
Bachelor grade 101.3 7.4 162091 99.1 8.2 116258 0.00
High school: general track (%) 84.1 36.6 182792 71.3 45.2 133678 0.00
science (%) 40.2 49.0 182792 56.6 49.6 133678 0.00
humanities (%) 21.2 40.9 182792 10.4 30.6 133678 0.00
foreign language (%) 10.5 30.6 182792 1.9 13.8 133678 0.00
social sciences (%) 10.3 30.4 182792 1.4 11.6 133678 0.00
arts (%) 1.9 13.6 182792 1.0 9.8 133678 0.00
High school: technical track (%) 12.8 33.5 182792 25.0 43.3 133678 0.00
High school: vocational track (%) 1.2 10.9 182792 1.7 12.9 133678 0.00
High school grade 83.6 11.6 178593 80.8 12.1 130134 0.00
Field of study
Science, chemistry, biology (%) 13.3 34.0 182792 13.1 33.7 133678 0.06
Engineering (%) 8.2 27.5 182792 27.0 44.4 133678 0.00
Humanities (%) 24.7 43.1 182792 10.4 30.5 133678 0.00
Political and social sciences (%) 11.6 32.0 182792 7.9 26.9 133678 0.00
Economics and statistics (%) 18.5 38.9 182792 24.3 42.9 133678 0.00
Psychology (%) 11.7 32.1 182792 3.1 17.4 133678 0.00
Healthcare (%) 4.0 19.7 182792 2.1 14.4 133678 0.00
Architecture (%) 3.9 19.5 182792 4.9 21.6 133678 0.00
Agriculture (%) 1.9 13.8 182792 2.9 16.9 133678 0.00
Family background
Mother: university education (%) 18.9 39.2 167637 22.0 41.4 119745 0.00
Father: university education (%) 20.0 40.0 167637 24.0 42.7 119745 0.00
Mother: high-school education (%) 50.3 50.0 167637 50.9 50.0 119745 0.00
Father: high-school education (%) 46.0 49.8 167637 47.2 49.9 119745 0.00
Mother is employed (%) 71.0 45.4 163753 73.2 44.3 116921 0.00
Father is employed (%) 99.3 8.0 162735 99.4 7.5 117051 0.00
Mother: low SES (%) 59.5 49.1 163753 55.8 49.7 116921 0.00
Mother: medium SES (%) 30.2 45.9 163753 32.6 46.9 116921 0.00
Mother: high SES (%) 10.4 30.5 163753 11.6 32.0 116921 0.00
Father: low SES (%) 45.6 49.8 162735 40.5 49.1 117051 0.00
Father: medium SES (%) 23.1 42.1 162735 24.2 42.8 117051 0.00
Father: high SES (%) 31.3 46.4 162735 35.3 47.8 117051 0.00
Notes. The table compares mean characteristics between female and male students in the sample. Variables in
this panel were collected in the administrative data and in the institutional survey.

77
TABLE A.2. Summary Statistics of Initial Job Characteristics, Search Behavior and Civil
Status

Female Male
Mean SD Obs Mean SD Obs P-value

Respond to follow-up survey (%) 73.7 44.0 182792 73.2 44.3 133678 0.00
Married/cohabiting with partner (%) 16.1 36.8 134506 9.5 29.3 97709 0.00
Has children (%) 3.7 18.8 134514 2.2 14.6 97724 0.00
LM Participation at time of survey
Employed (%) 53.9 49.8 134681 61.8 48.6 97823 0.00
Internship (%) 13.4 34.1 134411 10.0 30.0 97615 0.00
Searching for a job (%) 20.5 40.3 134558 15.1 35.9 97707 0.00
Graduate education (%) 5.9 23.5 134414 4.0 19.6 97613 0.00
In a PhD program (%) 5.7 23.2 134476 8.1 27.3 97663 0.00
Job characteristics (if employed)
Net monthly earnings (€) 1077.8 499.3 69659 1324.5 509.6 57494 0.00
Weekly hours worked 32.9 13.2 72378 38.6 10.9 60297 0.00
Full-time job (%) 69.3 46.1 69659 86.2 34.5 57494 0.00
Hourly wage 8.9 6.4 69659 8.9 5.7 57494 0.67
High earnings occupation (%) 36.6 48.2 71062 61.3 48.7 59558 0.00
High fulltime occupation (%) 51.1 50.0 71062 74.3 43.7 59558 0.00
High earnings industry (%) 34.1 47.4 71264 47.8 50.0 59715 0.00
High fulltime industry (%) 38.0 48.5 71264 61.7 48.6 59715 0.00
Permanent contract (%) 22.8 42.0 72280 29.3 45.5 60237 0.00
Fixed-term contract (%) 50.9 50.0 72280 49.1 50.0 60237 0.00
Self-employment 16.5 37.2 72280 16.1 36.8 60237 0.04
No contract (%) 6.7 25.0 72280 3.1 17.2 60237 0.00
Private sector (%) 76.7 42.3 72420 86.2 34.5 60348 0.00
Public sector (%) 16.5 37.1 72420 11.0 31.3 60348 0.00
No profit (%) 6.8 25.2 72420 2.8 16.4 60348 0.00
Mismatch
Use skills acquired during Master (%) 42.1 49.4 72446 47.6 49.9 60352 0.00
Master necessary for current job (%) 47.6 49.9 72437 54.8 49.8 60319 0.00
Training in the Master good for job (%) 51.3 50.0 72413 56.6 49.6 60312 0.00
Job search process
Keep job found during studies (%) 25.8 43.8 72539 21.6 41.2 60415 0.00
Current job: first job after grad. (%) 80.2 39.8 72568 81.1 39.2 60435 0.00
Numbers of jobs from grad. 1.3 0.6 72788 1.2 0.6 60685 0.00
Start of job search:months from grad. 0.4 1.9 71524 0.4 1.7 59255 0.00
Accepted job offer :months from grad. 2.8 3.8 72284 2.5 3.5 60291 0.00
On the job search (%) 20.9 40.7 134674 19.9 39.9 97818 0.00
Notes. The table compares mean characteristics between female and male students in the sample. Variables in
this panel were collected in the follow-up survey conducted one year after graduation.

78
 FIGURE A.1. Heatmaps of other indicators of gender culture

(a) Female LFP (%) - Age: 25-34 (b) Ratio of female to male LFP (%) - Age: 25-34

(c) Percentage of firms without preference for (d) Standardised index of individual gender
male workers attitudes
Notes. Panel (a) and (b) present the FLFP and the ratio FLFP/MLFP of young women (25-34) in Italy. These are
constructed as averages from 2004-2007. Panel (c) presents the percentage of firms without hiring preferences
for male workers in 2003. Panel (d) presents a standardised index of gender culture, based on individual
answers to questions related to gender attitudes in the World Value Survey (2000). Provinces white represent
those with missing values. All of these measured are defined at the province level.

79
 TABLE A.3. Summary statistics of measures of gender culture in the sample, by gender

Female Male
Mean SD Mean SD P-value
Female labor force participation (age: 15-64) 49.3 11.2 50.2 11.1 0.00
Female/Male labor force participation (age: 15-64) 66.3 11.9 67.2 11.8 0.00
Female labor force participation (age: 25-34) 64.6 15.3 65.8 15.1 0.00
Female/Male labor force participation (age: 25-34) 73.9 13.1 74.9 12.9 0.00
Male labor force participation (age: 15-64) 73.7 4.6 74.0 4.5 0.00
Male labor force participation (age: 25-34) 86.5 6.5 87.0 6.4 0.00
% of female graduates in full-time job 56.0 9.6 56.8 9.5 0.00
% of female/male graduates in full-time job 71.6 6.7 72.1 6.6 0.00
% of firms without hiring pref. for male workers 34.5 7.8 35.1 7.8 0.00
Notes: The Table presents summary statistics f the measures of gender culture 1-5 presented in Section 3, by
gender. The sample of female and male students include, respectively, 182,792 and 133,678 students. Students
are assigned to provinces based on their residence province prior to enrollment in the Master.

TABLE A.4. Mobility patterns by gender in the sample

Female Male
Mean SD Mean SD P-value

Moved to another province for Master (%) 58.9 49.2 55.4 49.7 0.00
Moved to another region for Master (%) 31.3 46.4 29.1 45.4 0.00
Bachelor and Master in same univ. (%) 71.5 45.1 75.7 42.9 0.00
Gender culture in province of university
Female labor force participation (age: 15-64) 52.9 10.5 53.7 10.2 0.00
Female/Male labor force participation (age: 15-64) 70.2 11.3 71.0 11.0 0.00
Female labor force participation (age: 25-34) 69.1 14.2 70.2 13.7 0.00
Female/Male labor force participation (age: 25-34) 78.2 11.9 79.0 11.5 0.00
Male labor force participation (age: 15-64) 74.9 4.0 75.1 3.8 0.00
Male labor force participation (age: 25-34) 87.7 6.1 88.2 5.9 0.00
% of female graduates in full-time 58.8 8.8 59.7 8.6 0.00
% of female/male graduates in full-time 73.3 6.0 73.9 5.8 0.00
% of firms without hiring pref. for male workers 58.5 8.2 59.1 7.8 0.00
Notes. The table provides summary statistics regarding students’ mobility for their studies. Besides mobility
rates by gender, it provides information on the local gender culture in the province of studies of students. The
sample of female and male students include, respectively, 182,792 and 133,678 students.

80
 TABLE A.5. Mobility to local labor markets by gender in the sample

Female Male
Mean SD Mean SD P-value

Work in province of studies (%) 45.2 49.8 43.7 49.6 0.0

Work in region of studies (%) 68.4 46.5 65.2 47.6 0.0
Work abroad (%) 5.0 21.8 5.3 22.4 0.0
Work outside province of origin (%) 44.0 49.6 51.4 50.0 0.0
Gender culture in province of work
Female labor force participationn (age:15-64) 54.6 9.8 55.6 9.2 0.0
Female/Male labor force participationn (age:15-64) 71.7 10.4 72.8 9.8 0.0
Female labor force participationn (age:25-34) 71.7 13.2 73.0 12.3 0.0
Female/Male labor force participationn (age:25-34) 79.7 11.0 80.8 10.3 0.0
Male labor force participationn (age:15-64) 75.7 3.8 76.0 3.6 0.0
Male labor force participationn (age:25-34) 89.2 5.7 89.8 5.4 0.0
% of female graduates in full-time 60.6 8.5 61.6 8.1 0.0
% of female/male graduates in full-time 74.6 5.9 75.2 5.6 0.0
% of firms without hiring pref. for male workers 58.5 8.3 59.3 8.1 0.0
Notes. The table provides summary statistics regarding students’ mobility to local labor markets in their first
job after graduation. Besides mobility rates by gender, it provides information on the local gender culture in
the province of work of students. The sample includes female and male students who are employed at the
moment of the survey, corresponding to 69,659 and 57,494 students.

FIGURE A.2. Students from below-median FLFP provinces within degrees

(A) Distribution of degrees (B) Students’ represented

Notes. Panel (A) represents degrees by the % of students from above-median FLFP areas in 2016, categorize
in brackets 0-10, ..., 90-100. One unit corresponds to a degree (N=1,572). Panel (B) shows the percentage of
students represented by degrees that fall within each of these brackets.

81
 TABLE A.6. Mobility patterns by gender culture in province of origin (Female sample)

Q1 FLFP Q4 FLFP
(N=48,896) (N=44,103)
Mean SD Mean SD P-value

Moved to another province for Master (%) 57.8 49.4 63.0 48.3 0.00
Moved to another region for Master (%) 37.2 48.3 27.1 44.5 0.00
Work in different province than birth (%) 54.8 49.8 37.8 48.5 0.00
Types of mobility (only for movers)
FLFP (age: 15-64) in prov. of university 49.7 11.7 60.0 4.4 0.00
Size of university 33797.4 16373.0 36606.5 18288.8 0.00
Nb. of students in the degree 80.3 61.8 80.9 59.4 0.21
% of female students in the degree 69.0 18.6 65.6 18.5 0.00
% of movers in the degree 60.4 20.6 72.0 14.8 0.00
% of movers (region) in the degree 32.6 25.7 41.9 19.4 0.00
% of peers from above-median FLFP prov 27.0 31.3 67.1 17.4 0.00
Field of study
Science, chemistry, biology (%) 13.8 34.5 12.6 33.2 0.00
Engineering (%) 8.3 27.5 6.2 24.2 0.00
Humanities (%) 25.7 43.7 27.1 44.5 0.00
Political and social sciences (%) 11.7 32.2 11.9 32.4 0.45
Economics and statistics (%) 14.2 34.9 17.0 37.6 0.00
Psychology (%) 15.6 36.3 12.3 32.9 0.00
Healthcare (%) 4.3 20.2 3.8 19.1 0.00
Architecture (%) 2.7 16.1 4.8 21.5 0.00
Agriculture (%) 1.4 11.8 2.2 14.6 0.00
Notes. The table provides summary statistics regarding students’ mobility in the sample of female students,
contrasting students from provinces in the first vs. fourth quartiles of FLFP. Besides mobility rates by gender,
it provides information on the characteristics of mobility in the sample of movers (57.8% and 63% of the two
samples).

82
 TABLE A.7. Average characteristics of degrees

Mean SD p50 Min Max

Size of degree 47.0 43.0 34.0 4.0 410.0

% of female students 55.6 21.3 56.3 3.8 97.1
% of movers 55.3 23.2 56.3 0.0 100.0
% of movers (region) 28.5 23.1 25.0 0.0 91.7
% of students from above-median FLFP provs. 40.8 36.3 50.0 0.0 100.0
% of students with BSc at same univ. 72.9 22.6 78.4 0.0 100.0
The table presents average characteristics of degrees related to students’ composition. The
unit of observation is a degree. There are 1,572 degrees in the sample.

TABLE A.8. Gender differences in the extensive margin of labor supply

(1) (2) (3) (4)

Has contract in LM Employed today Internship today Has worked

Female 0.000 -0.014*** 0.015*** -0.002

(0.000) (0.003) (0.002) (0.003)

Degree FEs ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓
Observations 232,026 232,504 232,026 232,504
R-squared 0.114 0.128 0.078 0.104
Notes: The table reports coefficients from regressions of graduates’ labor market participation on a female
dummy, after including degree and cohort fixed effects. The dependent variables are as follows. Column
1: indicator for whether a student is participating in the labor market with either a standard contract or an
internship. Column 2: indicator for whether a student is employed with a standard contract one year after
graduation. Column 3: indicator for whether a student is employed with an internship contract. Column
4: indicator for whether a student has been employed at any point during the first year after graduation.
Standard errors are clustered at degree level.

83
 TABLE A.9. The gender earnings gap at labor market entry, with controls for job types

(1) (2) (3) (4)

Log(monthly earnings) Log(weekly hours) Pr(fulltime) Log(wage)

Female -0.087*** -0.057*** -0.032*** -0.030***

(0.003) (0.003) (0.002) (0.003)

GPA ✓ ✓ ✓ ✓
Job characteristics ✓ ✓ ✓ ✓
Degree FEs ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓
Observations 127,153 127,153 127,153 127,153
R-squared 0.407 0.382 0.381 0.181
Notes: The table reports coefficients from regressions of graduates’ labor market outcomes on a female
dummy, after including degree and cohort fixed effects and controlling for covariates (GPA, prov. of work
FEs, occupation FEs (20 classes), industry FEs (21 classes)). The sample consists of female and male students
who are employed one year post graduation. Standard errors are clustered at the degree level.

TABLE A.10. The gender earnings gap excluding individuals with children or married

(1) (2) (3) (4)

Log(monthly earnings) Log(weekly hours) Pr(fulltime) Log(wage)

Female -0.104∗∗∗ -0.079∗∗∗ -0.045∗∗∗ -0.030∗∗∗

(0.004) (0.004) (0.003) (0.003)

Degree FEs ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓
Observations 106,360 106,360 106,360 106,360
Nb. of degrees 1,570 1,570 1,570 1,570
R-squared 0.313 0.269 0.309 0.093
Notes: The table reports coefficients from regressions of graduates’ labor market outcomes on a female
dummy, after including degree and cohort fixed effects. The sample consists of female and male students
who are employed one year after graduation, excluding individuals with children or those who are married
or cohabiting with a partner. Standard errors are clustered at the degree (master x university) level.

84
 TABLE A.11. Selection of female movers by FLFP in place of birth

(1) (2) (3) (4)

Low FLFP High FLFP P-value Observations

Individual characteristics
Age at enrollment 23.84 23.83 0.84 16,496
Bachelor grade 101.01 101.84 0.00 14,736
High school type: general track (%) 88.53 81.31 0.00 16,496
science (%) 47.50 41.50 0.00 16,496
humanities (%) 23.63 15.16 0.00 16,496
foreign language (%) 9.10 13.13 0.00 16,496
social sciences (%) 7.17 9.64 0.00 16,496
arts (%) 1.12 1.88 0.00 16,496
High school type: technical track (%) 9.87 15.81 0.00 16,496
High school type: vocational track (%) 0.90 0.80 0.54 16,496

Family background
Years of educucation mother 12.22 12.48 0.00 14,958
Years of educucation father 12.13 12.09 0.63 14,958
Mother: university degree (%) 19.03 19.67 0.51 15,185
Father: university degree (%) 19.82 19.58 0.82 15,185
Mother: high-school degree (%) 48.74 50.27 0.14 14,958
Father: high-school degree (%) 45.60 44.07 0.18 14,958
Mother is in the LF (%) 62.24 81.84 0.00 14,866
Father is in the LF (%) 99.36 99.38 0.93 14,766
Mother: low SES (%) 37.61 45.79 0.00 10,503
Mother: medium SES (%) 48.83 39.84 0.00 10,503
Mother: high SES (%) 13.55 14.37 0.38 10,503
Father: low SES (%) 45.28 45.21 0.95 14,671
Father: medium SES (%) 26.39 21.11 0.00 14,671
Father: high SES (%) 28.33 33.68 0.00 14,671
Notes: This table examines the selection of female movers into degree programs based on their province
of birth, categorized by the FLFP (top versus bottom quartile). The table compares the ability, educational
histories and socio-economic background of female movers by place of birth. For each pre-determined
characteristic, a separate regression is estimated following the empirical model 1. Predicted values for each
group are presented in Columns 1 and 2, while Column 3 reports the p-value from a significance test on α.

85
 TABLE A.12. Estimates of gender culture on men’s labor supply at labor market entry

(1) (2) (3) (4) (5) (6)

Log(weekly hours) Pr(fulltime)

Q4 vs. Q1 FLFP 0.039∗∗∗ 0.039∗∗∗ 0.038∗∗∗ 0.011∗ 0.008 0.08

(0.008) (0.008) (0.009) (0.006) (0.006) (0.006)

Province of job FEs ✓ ✓ ✓ ✓

GPA ✓ ✓
Mother’s occupation ✓ ✓
Father’s occupation ✓ ✓
Degree FEs ✓ ✓ ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓ ✓ ✓
N 15,597 15,595 14,014 15,597 15,595 14,014
Notes: The table reports coefficients from separate regressions of men’s labor market outcomes on a dummy
variable indicating whether the student originates from a province with FLFP in the highest vs. lowest
quartile. All regressions include controls for degree and cohort fixed effects. The sample consists of male
movers, defined as men working in a different province from their birth province, who are employed one
year post-graduation. Variations in sample sizes across columns arise from missing parental background data
for some students. Standard errors are clustered at the degree level.

86
 TABLE A.13. Balancing tests for cohort composition - Female students

Panel A. Educational history and ability

(1) (2) (3) (4) (5) (6) (7) (8)
Age HS science HS HS foregin HS social HS BSc grade Bsc grade >
humanities languages sciences vocational p50
(Mean) (24.3) (0.40) (0.21) (0.11) (0.10) (0.01) (101.3) (0.50)

δˆ F P -0.030 -0.003 0.002 0.001 0.003 0.002 0.078 -0.001

(0.092) (0.006) (0.005) (0.004) (0.004) (0.001) (0.124) (0.008)
δˆ M P -0.110 -0.009* -0.000 0.002 -0.004 0.001 -0.009 -0.001
(0.095) (0.005) (0.004) (0.003) (0.004) (0.001) (0.090) (0.006)

Degree FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Observations 182,792 182,792 182,792 182,792 182,792 182,792 162,091 162,091
87

R-squared 0.168 0.139 0.107 0.125 0.128 0.020 0.204 0.149

Panel B. Parental background

(1) (2) (3) (4) (5) (6) (7) (8)
Mother: univ. Father: univ. Mother: HS Father: HS Mother: low Mother: high Father: low Father: high
SES SES SES SES
(Mean) (0.19) (0.20) (0.48) (0.44) (0.43) (0.15) (0.45) (0.32)

δˆ F P -0.002 -0.009 0.007 -0.005 -0.009 -0.004 0.007 -0.007

(0.006) (0.006) (0.007) (0.007) (0.008) (0.006) (0.007) (0.007)
δˆ M P -0.004 -0.006 -0.001 0.006 0.007 -0.002 0.008 0.001
(0.004) (0.004) (0.005) (0.005) (0.006) (0.005) (0.006) (0.005)

Degree FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Observations 167,637 167,637 167,637 167,637 116,206 116,206 161,674 161,674
R-squared 0.043 0.042 0.019 0.015 0.039 0.026 0.036 0.038
Notes: The table reports OLS estimates of ˆδFP and ˆδMP from the baseline model (equation 2). Each column corresponds to a different regression, where the dependent
variables are pre-determined covariate of a student, related to educational history and ability (Panel A), and parental background (Panel B). Regressions include
cohort and degree fixed effects. The sample consists of all female students in the sample (N=182,792). Variations in sample sizes across columns arise from missing
information on some of the covariates. All regressors are standardised. Standard errors are clustered at degree level.
 TABLE A.14. Balancing tests for cohort composition - Male students

Panel A. Educational history and ability

(1) (2) (3) (4) (5) (6) (7) (8)
Age HS science HS HS foregin HS social HS BSc grade Bsc grade >
humanities languages sciences vocational p50
(Mean) (24.5) (0.56) (0.10) (0.02) (0.01) (0.02) (99.1) (0.50)

δˆ F P 0.045 0.010 -0.006 0.000 -0.001 0.000 0.239* 0.017**

(0.059) (0.006) (0.004) (0.002) (0.001) (0.002) (0.129) (0.008)
δˆ M P -0.070 -0.003 0.004 -0.002 0.003 -0.000 -0.136 -0.005
(0.074) (0.007) (0.005) (0.002) (0.002) (0.002) (0.124) (0.008)

Degree FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Observations 133,678 133,678 133,678 133,678 133,678 133,678 116,256 116,256
88

R-squared 0.218 0.106 0.121 0.070 0.055 0.044 0.234 0.173

Panel B. Parental background

(1) (2) (3) (4) (5) (6) (7) (8)
Mother: univ. Father: univ. Mother: HS Father: HS Mother: med Mother: high Father: med Father: high
SES SES SES SES
(Mean) (0.22) (0.24) (0.49) (0.45) (0.40) (0.16) (0.40) (0.36)

δˆ F P -0.000 -0.012** -0.012* 0.005 -0.001 -0.000 0.009 -0.007

(0.006) (0.006) (0.007) (0.007) (0.008) (0.006) (0.007) (0.007)
δˆ M P -0.005 -0.001 0.000 0.004 -0.004 -0.004 -0.015** 0.014*
(0.006) (0.006) (0.007) (0.007) (0.008) (0.006) (0.007) (0.007)

Degree FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cohort FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Observations 119,743 119,743 119,743 119,743 85,590 85,590 116,389 116,389
R-squared 0.037 0.036 0.020 0.019 0.038 0.030 0.033 0.041
Notes: The table reports OLS estimates of ˆδFP and ˆδMP from the baseline model (equation 2). Each column corresponds to a different regression, where the dependent
variables are pre-determined covariate of a student, related to educational history and ability (Panel A), and parental background (Panel B). Regressions include
cohort and degree fixed effects. The sample consists of all male students in the sample (N=133,678). Variations in sample sizes across columns arise from missing
information on some of the covariates. All regressors are standardised. Standard errors are clustered at degree level.
 FIGURE A.3. Year-to-Year Variation in Students’ Geographical Origins

Notes: The figure plots the distribution of residuals from a OLS regression of the average characteristic in the
province of origin of female (red) or male students (blue) on cohort and degree fixed effects. One observation
corresponds to a degree-cohort pair. Histograms are presented by bins of 0.75. The normal distribution is
plotted for comparison. 89
TABLE A.15. Raw and Residual Variation of Additional Measures of Peers’ Gender Culture

Mean SD Min Max

A: Avg FLFP (25-34) in province of origin of female peers

Raw cohort variable 65.00 11.93 39.85 85.00
Residuals: net of degree and cohort fixed effects 0.00 2.43 -19.43 16.92

B: Avg FLFP (25-34) in province of origin of male peers

Raw cohort variable 65.09 12.02 39.33 85
Residuals: net of degree and cohort fixed effects 0.00 2.81 -24.04 18.37

C: Avg FLFP/MLFP in province of origin of female peers

Raw cohort variable 66.76 8.88 43.62 85.36
Residuals: net of degree and cohort fixed effects 0.00 2.01 -14.64 10.59

D: Avg FLFP/MLFP in province of origin of male peers

Raw cohort variable 66.81 8.99 43.02 85.29
Residuals: net of degree and cohort fixed effects 0.00 2.31 -18.80 15.40

E: % of fulltime female graduates in prov. of female peers

Raw cohort variable 55.93 7.53 40.11 68.93
Residuals: net of degree and cohort fixed effects 0.00 1.56 -14.37 9.04

F: % of fulltime female graduates in prov. of male peers

Raw cohort variable 55.95 7.59 40.11 68.93
Residuals: net of degree and cohort fixed effects 0.00 1.74 -13.79 10.48

G: Firms gender culture in province of origin of female peers

Raw cohort variable 53.88 5.88 37.00 71.00
Residuals: net of degree and cohort fixed effects 0.00 1.83 -15.09 9.31

H: Firms gender culture in province of origin of male peers

Raw cohort variable 54.09 6.04 36.00 68.00
Residuals: net of degree and cohort fixed effects 0.00 2.09 -12.18 11.08
Notes: The table reports descriptive statistics for the main measures of gender culture in the province of
origin of female (Panel A) and male (Panel B) peers, before and after removing degree and cohort fixed effects.
The unit of observation is a degree-cohort pair, leading to a total of 7,160 observations.

90
 FIGURE A.4. Time series of exposure to peers by deciles of program’s size

Notes: This figure plots the evolution in time series of the average FLFP in the province of origin of female
peers within 10 randomly picked degrees. All programs were divided into deciles based on the average size
across all years. One program was randomly chosen within each decile.

91
TABLE A.16. Raw and Residual Variation of Peers’ Gender Culture by Quintiles of Degree
Size

Mean SD Min Max

A: Avg FLFP in province of origin of female peers

Quintile 1 (<21 students)
Raw cohort variable 48.68 8.87 32.09 66.18
Residuals: net of degree and cohort FEs -0.00 2.37 -13.83 9.05
Quintile 2 (21-31 students)
Raw cohort variable 50.79 7.80 31.14 66.18
Residuals: net of degree and cohort FEs 0.00 2.01 -9.68 8.21
Quintile 3 (32-42 students)
Raw cohort variable 50.00 8.45 29.89 66.18
Residuals: net of degree and cohort FEs 0.00 1.91 -12.82 8.40
Quintile 4 (43-70 students)
Raw cohort variable 49.73 8.38 31.02 65.99
Residuals: net of degree and cohort FEs 0.00 1.54 -9.77 10.78
Quintile 5 (70-413 students)
Raw cohort variable 49.72 8.35 30.06 63.19
Residuals: net of degree and cohort FEs 0.00 1.23 -10.72 4.77

B: Avg FLFP in province of origin of male peers

Quintile 1 (<21 students)
Raw cohort variable 48.76 9.09 29.87 66.18
Residuals: net of degree and cohort FEs -0.00 2.37 -11.36 11.33
Quintile 2 (21-31 students)
Raw cohort variable 50.76 7.86 32.09 65.10
Residuals: net of degree and cohort FEs 0.00 2.24 -13.17 9.65
Quintile 3 (32-42 students)
Raw cohort variable 50.18 8.54 29.49 65.33
Residuals: net of degree and cohort FEs 0.00 2.11 -11.47 9.11
Quintile 4 (43-70 students)
Raw cohort variable 49.78 8.40 29.77 63.71
Residuals: net of degree and cohort FEs 0.00 1.91 -9.50 14.47
Quintile 5 (71-413 students)
Raw cohort variable 49.85 8.36 29.48 66.37
Residuals: net of degree and cohort FEs 0.00 1.55 -9.61 11.76
Notes: The table reports descriptive statistics for the average FLFP in the province of origin for of female (Panel
A) and male (Panel B) peers. These statistics are provided for groups of degrees categorized by quintiles
based on the size of the degree programs. Quintiles are determined using the average size of programs
calculated over five cohorts. The number of students in each program is indicated in parentheses next to
the corresponding quintile. For instance, degrees in the first quintile include programs with fewer than 21
students.

92
 TABLE A.17. Estimates of Peer Effects on Job Characteristics - Female sample

(1) (2) (3) (4) (5)

Permanent No contract Self-employment Public No-profit

δFP -0.002 -0.007 -0.003 0.001 0.007

(0.010) (0.005) (0.008) (0.010) (0.005)
δMP -0.008 -0.004 -0.012* 0.001 0.003
(0.007) (0.004) (0.007) (0.008) (0.005)

Degree FE ✓ ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓ ✓
Observations 69,417 69,417 69,417 69,556 69,556
R-squared 0.137 0.098 0.144 0.203 0.153
Notes: OLS estimates of regressions of types of contract and sector one year after graduation on: the average
FLFP in the provinces of origin of female and male peers and the FLFP in the own province of origin. All
the dependent variables are indicator variables. All the estimates are done on the sample of women who are
employed one year after graduation and with non-missing information on the dependent variables. Standard
errors clustered at degree level. All regressors are standardised.

TABLE A.18. Estimates of Peer Effects on Female Earnings - Controls for Degree Trends

(1) (2) (3) (4)

Log(monthly Log(weekly Pr(fulltime) Log(hourly
earnings) hours) wage)

δˆ F P 0.045*** 0.039*** 0.029*** 0.007

(0.014) (0.014) (0.010) (0.014)
δ̂ M P -0.002 -0.009 -0.005 0.006
(0.011) (0.011) (0.008) (0.011)

Degree FE ✓ ✓ ✓ ✓
Degree trends ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 69,645 69,645 69,645 69,645
R-squared 0.308 0.266 0.299 0.124
Notes: OLS estimates of a regression of women’s earnings and labor supply one year after graduation on:
the average FLFP in the provinces of origin of female and male peers and the FLFP in the own province
of origin. Regressions include cohort and degree fixed effects, as well as degree-specific linear time trends.
All the estimates are done on the sample of women who are employed one year after graduation and with
non-missing information on the dependent variables. Standard errors clustered at degree level. All regressors
are standardised.

93
TABLE A.19. Estimates of Peer Effects on Female Earnings - Controls for Region Trends

(1) (2) (3) (4)

Log(monthly Log(weekly Pr(fulltime) Log(hourly
earnings) hours) wage)

δˆ F P 0.036*** 0.030** 0.019** 0.005

(0.013) (0.012) (0.009) (0.012)
δ̂ M P -0.001 -0.001 -0.002 -0.001
(0.010) (0.009) (0.007) (0.009)

Degree FE ✓ ✓ ✓ ✓
Region trends ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 69,645 69,645 69,645 69,645
R-squared 0.288 0.246 0.280 0.100
Notes: OLS estimates of a regression of women’s earnings and labor supply one year after graduation on: the
average FLFP in the provinces of origin of female and male peers and the FLFP in the own province of origin.
Regressions include cohort and degree fixed effects, as well as region (of studies)-specific linear time trends.
All the estimates are done on the sample of women who are employed one year after graduation and with
non-missing information on the dependent variables. Standard errors clustered at degree level. All regressors
are standardised.

94
TABLE A.20. Estimates of Peer Effects on Female Earnings Excl. Degrees with Trends in
Size

(1) (2) (3) (4)

Log(monthly Log(weekly Pr(fulltime) Log(hourly
earnings) hours) wage)

δˆ F P 0.052*** 0.029* 0.030*** 0.021

(0.015) (0.015) (0.011) (0.014)
δ̂ M P 0.003 0.000 -0.002 0.002
(0.012) (0.011) (0.008) (0.010)

Degree FE ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 47,246 47,246 47,246 47,246
R-squared 0.286 0.250 0.278 0.095
Notes: OLS estimates of a regression of women’s earnings and labor supply one year after graduation on: the
average FLFP in the provinces of origin of female and male peers and the FLFP in the own province of origin.
Regressions include cohort and degree fixed effects. The sample excludes degrees that experience trends in
size over time.The estimates are done on the sample of women, studying in these degrees, who are employed
one year after graduation and with non-missing information on the dependent variables. Standard errors
clustered at degree level. All regressors are standardised.

95
 TABLE A.21. Validity of the Empirical Strategy - Estimates of Peer Effects on Female Earnings

(1) (2) (3) (4) (5) (6) (7)

Benchmark ∆ size<=p75 ∆ size<=p50 ∆ avg ∆ avg ∆ sd ∆ sd
grades<=p75 grades<=p25 grades<=p75 grades<=p25

δˆ F P 0.037*** 0.049*** 0.053*** 0.037** 0.041* 0.036** 0.054**

(0.013) (0.013) (0.015) (0.015) (0.025) (0.014) (0.022)
δ̂ M P -0.000 -0.002 -0.001 -0.003 -0.002 0.001 -0.004
(0.010) (0.011) (0.013) (0.011) (0.014) (0.012) (0.014)

Degree FE ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nb. of degrees 1,572 1,163 770 1,170 390 1,171 389
96

Observations 69,645 58,363 42,518 59,040 25,564 60,613 28,741

R-squared 0.287 0.294 0.309 0.280 0.251 0.278 0.284
Notes: The table reports estimates of the baseline specification 2. The dependent variable is log(monthly earnings) in all columns. Regressions include cohort and
degree fixed effects. In each column, a different sample of degrees is used. Column (1) and (2) refer to the sample of degrees that experience cross-cohort changes in
size below the 75th and 50th percentiles (see definitions in Section 6.1). Column (3) and (4) refer to the sample of degrees that experience cross-cohort changes in the
average ability of students below the 75th or 25th percentile. Column (5) and (6) refer to the sample of degrees that experience cross-cohort changes in the standard
deviation of ability of students below the 75th or 25th percentile. The estimates are done on the sample of women, studying in these degrees, who are employed
one year after graduation and with non-missing information on the dependent variables. Standard errors clustered at degree level. All regressors are standardised.
TABLE A.22. Other Checks - Estimates of peer effects on female earnings and labor supply

(1) (2) (3) (4)

Log(monthly Log(weekly Pr(fulltime) Log(hourly
earnings) hours) wage)

δˆ F P 0.037*** 0.033*** 0.018* 0.003

(0.013) (0.012) (0.009) (0.012)
δ̂ M P -0.001 -0.000 -0.003 -0.002
(0.010) (0.009) (0.007) (0.010)

Province of origin FEs ✓ ✓ ✓ ✓

Degree FE ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 69,645 69,645 69,645 69,645
R-squared 0.290 0.248 0.282 0.102
Notes: this table reports OLS estimates from specification 2. Relative to the baseline specification, this includes
controls for province of origin fixed effects, instead of the FLFP in the province of origin of the student.
Regressions include cohort and degree fixed effects. All the estimates are done on the sample of women who
are employed one year after graduation and with non-missing information on these variables. Standard errors
clustered at degree level. All regressors are standardised.

97
 TABLE A.23. Sensitivity to Sample Restrictions - Estimates of Peer Effects on Female Earnings

(1) (2) (3) (4) (5) (6) (7) (8)

Degree size Students with Bsc in same uni
Benchmark >p10 <=p90 <=mean >mean >p90 <=p90 <=p25

δˆ F P 0.037*** 0.032** 0.039*** 0.050*** 0.017 0.017 0.037*** 0.048**

(0.013) (0.013) (0.014) (0.016) (0.020) (0.029) (0.013) (0.019)
δ̂ M P -0.000 -0.000 -0.004 -0.014 0.015 0.011 0.000 0.008
(0.010) (0.010) (0.012) (0.012) (0.016) (0.017) (0.010) (0.013)

Degree FE ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nb. of degrees 1,572 1,403 1,399 1,037 519 157 1,404 391
98

Observations 69,645 68,409 46,721 22,804 46,841 22,924 65,453 21,886

R-squared 0.287 0.287 0.264 0.254 0.300 0.332 0.284 0.254
Notes: The table reports estimates of the baseline specification 2. The dependent variable is log(monthly earnings) in all columns. Regressions include cohort and
degree fixed effects. In each column, a different sample of degrees is used. Column (1) and (2) refer to the sample of degrees that experience cross-cohort changes in
size below the 75th and 50th percentiles (see definitions in Section 6.1). Column (3) and (4) refer to the sample of degrees that experience cross-cohort changes in the
average ability of students below the 75th or 25th percentile. Column (5) and (6) refer to the sample of degrees that experience cross-cohort changes in the standard
deviation of ability of students below the 75th or 25th percentile. The estimates are done on the sample of women, studying in these degrees, who are employed
one year after graduation and with non-missing information on the dependent variables. Standard errors clustered at degree level. All regressors are standardised.
 TABLE A.24. Sensitivity to Measures of Gender Culture - Estimates of Peer Effects on Female Earnings

(1) (2) (3) (4) (5) (6) (7)

Measures of gender culture in province of origin of peers
FLFP FLFP (young) FLFP/MLFP FLFP/MLFP % of female female/male Firms’
(young) grad. fulltime grad. fulltime culture

δˆ F P 0.037*** 0.041*** 0.035*** 0.036*** 0.040*** 0.041*** 0.016**

(0.013) (0.014) (0.012) (0.013) (0.012) (0.011) (0.008)
δ̂ M P -0.000 0.003 0.000 0.002 0.010 0.009 0.004
(0.010) (0.011) (0.010) (0.010) (0.010) (0.009) (0.006)

Degree FE ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓ ✓ ✓ ✓
99

Nb. of degrees 1,572 1,572 1,572 1,572 1,572 1,572 1,572

Observations 69,645 69,645 69,645 69,645 69,645 69,645 69,645
R-squared 0.287 0.287 0.287 0.287 0.288 0.288 0.287
The table presents estimates from the baseline specification 2, using alternative measures of peers’ gender culture. The dependent variable in all columns is
log(monthly earnings). Each column uses a different proxy for peers’ gender culture, as specified in the column labels. Regressions include cohort and degree
fixed effects. The estimates are done on the sample of women, who are employed one year after graduation and with non-missing information on the dependent
variables. Standard errors clustered at degree level. All regressors are standardised.
 TABLE A.25. Robustness checks - Estimates of Peer Effects on Female Earnings Controlling for Other Peers’ Characteristics

Dependent variable: log(monthly earnings)

(1) (2) (3) (4) (5) (6) (7) (8) (9)

δ̂ F P 0.041*** 0.041*** 0.040*** 0.039*** 0.039*** 0.036*** 0.040*** 0.037*** 0.037***

(0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.013) (0.013)
δ̂ M P 0.002 0.003 0.003 0.003 0.003 0.002 -0.000 -0.000 -0.000
(0.010) (0.010) (0.010) (0.010) (0.010) (0.010) (0.010) (0.010) (0.010)

Share of peers with work. mother ✓

Share of peers with high-SES mother ✓
Share of peers with high-SES father ✓
Share of peers with college educ mother ✓
100

Share of peers with college educ father ✓

Share of high-ability peers ✓
Share of peers from general track ✓
Degree size ✓
Share of female peers ✓

Degree FE ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nb. of degrees 1,548 1,548 1,548 1,549 1,549 1,546 1,556 1,556 1,556
Observations 62,857 62,857 62,451 64,242 64,242 62,098 69,553 69,645 69,645
R-squared 0.293 0.293 0.293 0.291 0.291 0.292 0.288 0.287 0.288
Notes: The table presents estimates from the baseline specification 2 on log(monthly earnings). Each column represents a different regression, with an added
control for an alternative peer characteristic, disaggregated by gender. For instance, in Column 1, I include controls for the proportion of female and male peers
with working mothers. All regressions account for cohort and degree fixed effects and are conducted on the sample of women employed one year post-graduation,
with non-missing data on the relevant variables. Variation in sample size across columns results from missing values in certain covariates. Standard errors are
clustered at the degree level, and all regressors are standardized.
 TABLE A.26. Estimates of peer effects on human capital - Female sample

(1) (2) (3) (4)

GPA Final grade Time to Pr(delayed
completion grad.)
(Mean) (27.8) (108.6) (2.5) (0.35)

δFP 0.047 0.071 -0.004 -0.007

(0.029) (0.102) (0.010) (0.008)
δMP 0.039 0.066 -0.006 -0.005
(0.024) (0.085) (0.008) (0.007)

Degree FE ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 182,792 182,792 182,792 182,792
R-squared 0.244 0.174 0.161 0.148
Notes: OLS estimates of a regression of indicators of academic performance on: the average FLFP in the
provinces of origin of female and male peers and the FLFP in the own province of origin. Regressions include
cohort and degree fixed effects. All the estimates are done on the full sample of women. All regressors are
standardised. All regressors are standardised, while the dependent variables are not. The mean values of the
dependent variables are provided in the table. Standard errors clustered at degree level.

101
 TABLE A.27. Estimates of peer effects on geographic mobility - Female sample

(1) (2) (3) (4)

FLFP in prov. of Prov work = Reg work = Prov of work ≠
work univ. univ. birth
(Mean) (54.6) (0.45) (0.68) (0.44)

δFP 0.155 0.007 0.013 0.007

(0.151) (0.011) (0.011) (0.010)
δMP 0.126 -0.010 -0.005 0.009
(0.123) (0.008) (0.008) (0.007)

Degree FE ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 68,751 72,367 72,367 72,367
R-squared 0.586 0.156 0.152 0.181
Notes: OLS estimates of a regression of indicators of women’s geographic mobility on: the average FLFP in
the provinces of origin of female and male peers and the FLFP in the own province of origin. Regressions
include cohort and degree fixed effects. All the estimates are done on the sample of women employed one
year post-graduation, with non-missing data on the relevant variables. The dependent variables are as follows:
FLFP in the province of employment (Column 1), an indicator of whether the province of employment is
the same as that of the university attended (Column 2), an indicator of whether the region of employment
matches the university’s region (Column 3), and an indicator of whether the province of employment differs
from the province of birth (Column 4). The means of these variables are reported in parentheses. Standard
errors clustered at degree level. All regressors are standardised.

102
TABLEA.28. Robustness Checks - Estimates of Peer Effects on Female Earnings and Labor
Supply Controlling for Share of Local Students

(1) (2) (3) (4)

Log(monthly Log(weekly Pr(fulltime) Log(hourly
earnings) hours) wage)

δFP 0.045*** 0.041*** 0.022** 0.003

(0.013) (0.012) (0.010) (0.013)
δMP -0.002 -0.002 -0.001 -0.000
(0.010) (0.010) (0.007) (0.010)
Share of female stayers -0.011* -0.010* -0.004 -0.000
(0.006) (0.006) (0.004) (0.005)
Share of male stayers 0.003 0.006 -0.001 -0.003
(0.005) (0.005) (0.004) (0.004)

Degree FE ✓ ✓ ✓ ✓
Cohort FE ✓ ✓ ✓ ✓
Observations 69,645 69,645 69,645 69,645
R-squared 0.288 0.246 0.280 0.100
Notes: OLS estimates of a regression of women’s earnings and labor supply one year after graduation on: the
average FLFP in the provinces of origin of female and male peers and the FLFP in the own province of origin,
as well as the share of local female and male peers. A student is defined as local if she studies at university
in her province of birth. Regressions include cohort and degree fixed effects. All the estimates are done on
the sample of women employed one year post-graduation, with non-missing data on the relevant variables.
Standard errors clustered at degree level. All regressors are standardised.

103
 TABLE A.29. Estimates of Peer Effects on Job-Search Preferences

(1) (2) (3)

Index Pecuniary Index Flexibility Job’s social utility

δFP 0.003 -0.027* -0.012*

(0.009) (0.015) (0.007)
δMP 0.001 0.006 0.001
(0.007) (0.011) (0.005)

Degree FE ✓ ✓ ✓
Cohort FE ✓ ✓ ✓
Observations 165,116 163,855 164,214
R-squared 0.089 0.043 0.093
Notes: OLS estimates of regressions of valuation of job attributes on: the average FLFP in the provinces
of origin of female and male peers and the FLFP in the own province of origin. The dependent variables
in Columns (1)-(3) measure the importance students place on different job characteristics. Answers come
the question: "How much do you value attribute X in the job you are searching?" (scale 1-5). Specifically,
Column (1) reflects preferences for pecuniary job attributes, i.e. as salary and career progression, based
on a standardized index constructed from students’ rankings on a 1-5 scale. The index in Column (2) is
constructed by averaging students’ rankings of job attributes related to flexibility (i.e. leisure time and hours
flexibility). Both indexes in (1) and (2) have been standardised. The dependent variable in Column (3) is an
indicator variable for whether a student gives maximum value to the social utility of a job. Regressions include
cohort and degree fixed effects. The estimates are done on the sample of women who fill in the institutional
pre-graduation survey (91.7%). Standard errors clustered at degree level. All regressors are standardised.

104
TABLE A.30. Main Characteristics of the Sample of Prospective Students - Original Survey

All Low FLFP High FLFP

Mean SD N Mean SD Mean SD
Background Characteristics
Age 23.4 1.8 487 23.6 2.3 23.3 1.5
Changed province for Master (%) 88.8 31.6 490 100.0 0.0 82.6 37.9
Changed region for Master (%) 69.6 46.0 490 100.0 0.0 53.0 50.0
FLFP in province of origin 54.6 11.2 489 41.9 8.9 61.5 4.0
Mother: university level (%) 31.4 46.5 468 27.3 44.7 33.7 47.3
Father: university level (%) 28.0 44.9 465 32.7 47.1 25.3 43.6
Mother: full-time at childbirth (%) 49.9 50.1 465 45.7 50.0 52.2 50.0
Mother: part-time at childbirth (%) 30.1 45.9 465 23.2 42.3 33.9 47.4
Mother: no work at childbirth (%) 20.0 40.0 465 31.1 46.4 14.0 34.7
Field of study
Major: Economics (%) 19.2 39.4 480 20.0 40.1 18.7 39.1
Major: Humanities (%) 45.2 49.8 480 40.0 49.1 48.1 50.0
Major: Science (%) 20.4 40.4 480 23.5 42.5 18.7 39.1
Major: Social Sciences (%) 15.2 35.9 480 16.5 37.2 14.5 35.3
First year (%) 65.1 47.7 490 61.8 48.7 66.9 47.1
Second year (%) 33.5 47.2 490 38.2 48.7 30.9 46.3
Above second year (%) 1.4 11.9 490 0.0 0.0 2.2 14.7
Civil Status and Fertility Expectations
Single (%) 48.1 50.0 468 46.1 50.0 49.2 50.1
Has a partner (%) 46.4 49.9 468 48.5 50.1 45.2 49.9
Cohabits with partner (%) 5.6 22.9 468 5.5 22.8 5.6 23.1
Partner in same program (%) 3.4 18.2 468 2.4 15.4 4.0 19.5
Intend to have children (%) 54.0 49.9 470 52.7 50.1 54.8 49.9
Maybe children (%) 33.2 47.1 470 35.8 48.1 31.8 46.6
Does not intend to have children (%) 12.6 33.2 470 11.5 32.0 13.1 33.8
Has children already (%) 0.2 4.6 470 0.0 0.0 0.3 5.7
Expected age at first child 31.3 2.8 351 31.7 3.2 31.1 2.6
Intended Job Search
Intend to search for a job (%) 79.7 40.3 488 80.8 39.5 79.1 40.7
Intend to pursue further education (%) 19.1 39.3 488 18.0 38.6 19.6 39.8
Intend to keep job (%) 1.2 11.0 488 1.2 10.8 1.3 11.2
Job location: North (%) 61.2 48.8 485 62.2 48.6 60.7 48.9
Job location: Centre (%) 15.7 36.4 485 14.0 34.8 16.6 37.3
Job location: South (%) 2.7 16.2 485 7.0 25.5 0.3 5.7
Job location: Abroad (%) 20.4 40.3 485 16.9 37.5 22.4 41.7
Notes: this table summarizes the main characteristics of the sample of prospective students that participated
in the original survey. It reports the mean and standard deviation of variables related to students’ background,
fields of study, civil status, partner information, fertility expectations, and labor market intentions. These
statistics are reported for the overall sample (490 students), as well as for the two subsamples of students
from above-median (317 students) and below-median (173 students) FLFP provinces.

105
TABLEA.31. Baseline and Updated Beliefs on the Job Offer Distribution - Robustness
Checks

Below-med FLFP Above-med FLFP

Pred SE Pred SE P-value
a. Baseline Beliefs (T=0)
α: Expected arrival rate of job offers (%) 32.30 1.80 35.05 1.26 0.23
γP: Expected % of part-time job offers 57.48 2.43 50.33 1.70 0.02
Perceived uncertainty (1-5) 2.80 0.13 2.93 0.09 0.44
Prob. to accept part-time job offer 67.26 2.17 59.66 1.50 0.01

b. Updated Beliefs (T=1)

α: Expected arrival rate of job offers (%) 32.82 2.47 32.43 1.93 0.91
γP: Expected % of part-time job offers 52.41 3.11 51.89 2.43 0.90
Perceived uncertainty (1-5) 2.63 0.15 2.74 0.12 0.57
Prob. to accept part-time job offer 62.03 2.94 63.94 2.28 0.63
Notes: This table presents predictions from a linear regression model, where the dependent
variable is regressed on an indicator for whether the FLFP in the birth province is above or
below the median, along with fixed effects for the field of study and controls for students’
background characteristics (age, parents’ education), job search intentions, and expected
job location. Each row represents a different regression, with the dependent variable
specified in Column 1. For each regression, the table reports the predicted dependent
variable for women from provinces with low versus high FLFP, along with the standard
errors. The last column provides the p-value for the difference between these two groups.
In Panel (a), the sample consists of all first-year female Master’s students without missing
information on the covariates (291), and in Panel (b), it includes all second-year female
Master’s students without missing information on the covariates (148). Between 60% and
65% of the students are from provinces with above-median FLFP.

106
 TABLE A.32. Beliefs on arrival rates of job offers and acceptance of part-time jobs

(1) (2) (3) (4)

Probability to accept part-time job offer

Expected percentage of part-time offers (γ) 0.327** 0.272**

(0.066) (0.056)
Expected arrival rate of job offers (α) -0.148** -0.078
(0.028) (0.035)

Field FEs ✓ ✓
Observations 463 463 464 464
R-squared 0.125 0.171 0.014 0.101
The table presents estimated coefficients from regressions of the elicited probability of accepting a part-time
job offer on workers’ expected probability of receiving a job offer (Columns 1-2) or the expected percentage
of part-time offers (Columns 3-4). In Columns 2 and 4, I include controls for the field of study. The sample
consists of all female students with non-missing values for these variables, drawn from both the first and
second year of the program.

107
 TABLE A.33. Baseline and Updated Expectations of Fertility and Future Labor Supply

Below-med FLFP Above-med FLFP

Pred SE Pred SE P-value
a. Baseline Expectations (T=0)
Fertility: yes 0.50 0.05 0.54 0.04 0.50
Fertility: don’t know 0.38 0.05 0.35 0.03 0.61
Fertility: no 0.12 0.03 0.11 0.02 0.76
Age of expected fertility 31.58 0.31 30.88 0.23 0.07
Labor supply at motherhood (Scenario 1)
Work full-time 0.49 0.05 0.43 0.04 0.40
Work part-time 0.49 0.05 0.55 0.04 0.38
No work 0.03 0.02 0.03 0.02 0.90
Labor supply at motherhood (Scenario 2)
Work full-time 0.70 0.05 0.68 0.04 0.81
Work part-time 0.28 0.05 0.31 0.04 0.58
No work 0.03 0.02 0.01 0.01 0.21

b. Updated Expectations (T=1)

Fertility: yes 0.57 0.07 0.61 0.05 0.63
Fertility: don’t know 0.32 0.06 0.23 0.04 0.27
Fertility: no 0.10 0.04 0.15 0.04 0.41
Age of expected fertility 32.65 0.44 31.13 0.35 0.01
Labor supply at motherhood (Scenario 1)
Work full-time 0.67 0.06 0.41 0.06 0.00
Work part-time 0.33 0.06 0.55 0.06 0.02
No work 0.04 0.03
Labor supply at motherhood (Scenario 2)
Work full-time 0.79 0.05 0.80 0.05 0.85
Work part-time 0.21 0.05 0.20 0.05 0.85
No work 0.00 0.00
Notes: This table presents predictions from logistic regressions, where the dependent
variable is regressed on an indicator for whether the FLFP in the birth province is above
or below the median, along with fixed effects for the field of study. Each row represents
a different regression, with the dependent variable specified in Column 1. For each
regression, the table reports the predicted dependent variable for women from provinces
with low versus high FLFP, along with the standard errors. The last column provides the
p-value for the difference between these two groups. In Panel (a), the sample consists of
all first-year female Master’s students without missing information on the dependent
variables, and in Panel (b), it includes all second-year female Master’s students without
missing information on the dependent variables. Between 60% and 65% of the students
are from provinces with above-median FLFP.

108