COVID-19 and Gender Equality in Luxembourg

Abstract

Two years after the emergence of the SARS-CoV-2 and the ensuing lockdown measures taken to contain the spread of the disease, the ongoing pandemic has had significant and wide-reaching implications on many aspects of life – health, economic and social – affecting different social groups in highly asymmetric ways.
At the beginning of the crisis, the OECD pointed out the burden that women were carrying: “First and foremost, women are leading the health response: women make up almost 70% of the health care workforce, exposing them to a greater risk of infection. At the same time, women are also shouldering much of the burden at home, given school and childcare facility closures and longstanding gender inequalities in unpaid work. Women also face high risks of job and income loss” (...)

Full text

www.liser.lu
REPORT
March 2022
COVID-19 and
Gender Equality in
Luxembourg

COVID-19 and Gender Equality in
Luxembourg
Commissioned by:
Project Coordinator :
Eugenio PELUSO
Contributors:
Fofo S. AMÉTÉPÉ - STATEC
Francesco ANDREOLI – LISER & Università di Verona
Anne-Sophie GENEVOIS - LISER
Giorgia MENTA - LISER
Eugenio PELUSO – LISER & University of Luxembourg
Ioana C. SALAGEAN - STATEC
Philippe VAN KERM – LISER & University of Luxembourg
Bertrand VERHEYDEN - LISER

- 3 -
Contents
Introduction ........................................................................................................................................... 5!
Chapter 1: A gendered disease? Gender differences in contaminations and severe forms
of COVID-19 ........................................................................................................................................... 8!
1. Introduction .......................................................................................................................................... 8!
2. Data and methods ............................................................................................................................... 9!
2.1 The data .................................................................................................................................... 9!
2.2 Methods ................................................................................................................................... 10!
3. Gender differences in SARS-CoV-2 contaminations ......................................................................... 11!
4. Gender differences in severe forms of COVID-19: hospitalisations and death ................................. 18!
5. Is vaccination the explanation? Gender differences in vaccination rates .......................................... 23!
6. Conclusion ......................................................................................................................................... 25!
References ............................................................................................................................................ 27!
Chapter 2: Gender differences in attitudes towards COVID-19 and health measures ................. 28!
1. Introduction ........................................................................................................................................ 29!
2. Data and descriptive statistic ............................................................................................................. 30!
2.1 Descriptive statistics of respondents’ characteristics .............................................................. 31!
2.2 Descriptive statistics of respondents’ outcomes ...................................................................... 33!
3. Empirical analysis .............................................................................................................................. 34!
4. The impact of Vaxzevria suspensions and of EMA communication on vaccination intentions ......... 39!
5. Conclusion ......................................................................................................................................... 41!
References ............................................................................................................................................ 43!
Appendix ............................................................................................................................................... 45!
Chapter 3: The effects of COVID-19 on gender differences in individual and household
behavior ............................................................................................................................................... 52!
1. Introduction ........................................................................................................................................ 53!
1.1 The effects of COVID-19 on labor market and time-use in developed countries .................... 53!
1.2 Gender differences in social isolation ...................................................................................... 55!
2. Data ................................................................................................................................................... 55!
3. Results .............................................................................................................................................. 56!
3.1 The labor market ..................................................................................................................... 56!
3.2 Subjective financial situation ................................................................................................... 69!

- 4 -
3.3 Social interactions (outside and within the household) ............................................................ 71!
4. Conclusions ....................................................................................................................................... 75!
References ............................................................................................................................................ 77!
Appendix ............................................................................................................................................... 79!
Conclusions and perspectives ......................................................................................................... 83!

- 5 -
Introduction1,2
Two years after the emergence of the SARS-CoV-2 and the ensuing lockdown measures taken to
contain the spread of the disease, the ongoing pandemic has had significant and wide-reaching
implications on many aspects of life – health, economic and social – affecting different social groups in
highly asymmetric ways.
At the beginning of the crisis, the OECD pointed out the burden that women were carrying: “First and
foremost, women are leading the health response: women make up almost 70% of the health care
workforce, exposing them to a greater risk of infection. At the same time, women are also shouldering
much of the burden at home, given school and childcare facility closures and longstanding gender
inequalities in unpaid work. Women also face high risks of job and income loss”.
The predictions set out by the OECD seem to be confirmed worldwide by a growing body of literature
investigating the gender gradient in the COVID-19 crisis. Whether the costs of the pandemic fell more
heavily on one of the genders depends on the outcome of interest as well as on the interaction that
gender has with other important dimensions, such as family formation, the presence of children in the
household and the way the time allocation between childcare, work and leisure differs across spouses
within the same household. This report considers three important domains of the living conditions of
adult Luxembourgish residents that have been affected by the pandemic: their health, their labor
market opportunities and their perceived financial insecurity. All dimensions display gender
inequalities, albeit the direction of the gender gradient is often unclear, per se, and needs to be
analyzed holding other characteristics of the households and of the individuals as constant in our
analysis.
From a health perspective, data show that the incidence of COVID-19 cases has not differed
significantly between men and women. Nonetheless, men have been much more likely to develop
severe forms of COVID-19, namely intensive care admissions and deaths; see, e.g., the “Sex, Gender
and COVID-19 Project” online global data tracker on gender differences in COVID-19 health impacts
(Global Health 50/50, 2021).
Health effects are interlaced with those affecting the socio-economic sphere. Attitudes, behaviors and
habits vary across genders due to history, culture and social norms and are likely to be important
determinants of individual responses to the pandemic. These “mediating factors” have influenced the
different exposure to contamination and the different responses of the population both to the risks of
the pandemic and to the policies implemented to fight this new plague (see, among others, Galasso et
al., 20203). These factors have also exacerbated gender differences in the economic and social
sphere. Prior to the pandemic, women in developed economies were more often than men in low-
paying and insecure jobs. During the pandemic, differences in compliance with social-isolation norms
and in risk-taking behavior have been observed between men and women that can explain some of
the observed gender differences in the socio-economic consequences of COVID-19. Recent evidence
proves that women suffered more than men along the dimensions such as employment, working
hours, earnings and income (see Todorovic et al., 2021a,4 for a review). Traditionally, women are also
1 Author: Eugenio Peluso.
2 We thank Ralph Kass for his inspiring comments, Anne Sophie Génévois for her help with data and Axelle
Depireux and Isabelle Bouvy for their precious assistance.
3 Alon, T., Doepke, M., Olmstead-Rumsey, J., and Tertilt, M. (2020). “The impact of the coronavirus pandemic on
gender equality.” Covid Economics Vetted and Real-Time Papers, 4, 62-85.
4 Todorovic, J., Van Kerm, P., and Peluso, E. (2021a). “Unemployment and working hours of women and men
during the pandemic.” LISER-MEGA series on gender issues in the COVID-19 pandemic.

- 6 -
more likely to shoulder the burden of childcare and elder care, which is particularly relevant during a
period of school closure and shielding of the elderly (Alon et al., 2020,5 or see Todorovic et al.,
2021b,6 for a review).
In order to develop new a new body of quantitative research about the multifaceted effects of COVID-
19, the “Ministère de l’Égalité entre les femmes et les hommes“ (MEGA), which is in charge of
encouraging the development of an egalitarian society and of developing innovative plans for equality
between women and men, prompted the Luxembourg Institute of Socio-Economic Research (LISER)
to integrate the gender dimension in scientific research on the COVID-19 pandemic and its social
consequences. This report presents research focused on gender inequalities in the specific context of
Luxembourg, drawing on refined administrative data on COVID-related cases as well as an innovative
dataset collected during the COVID-19 crisis, which elicits the attitudes of Luxembourgish residents on
health, labor, family, financial and educational dimensions vis-à-vis the pandemic shocks. Based on
such data sources, our empirical analysis allows to account for the institutional and policy context, thus
entailing some of the specificities that might have affected gender differences differently from other
countries.
During the last two years, the LISER has collaborated with the University of Luxembourg, LIH, LIST,
STATEC and other research institutes to analyze the impact of the COVID-19 pandemic in
Luxembourg. More than 10 projects and 70 reports and articles have been produced by LISER’s
researchers to investigate several facets of the impact of the pandemic and understand the effects of
this phenomenon on the Luxembourgish society. Some of these pieces of research, as the report
“Santé pour tous: La COVID-19 au Luxembourg: Le gradient social de l’épidémie”, or the “SEI Socio-
Economic Impacts of COVID-19: Collecting the data” have fuelled the present report, by providing
important data sources based on administrative data and online surveys.
With the analysis of such original data through statistical and econometric tools, this project produces
an assessment of the COVID-19 crisis from the angle of equality between the sexes. We have
analyzed the main effects of the crisis and of the subsequent policy responses on several dimensions
of individual well-being, such as health, income, life conditions, employment, time use and social
activities. We have also explored how gender differences in attitudes have shaped the determinants
and the roots of gender inequalities. Policies carried over to face the COVID-19 emergency have
contributed affecting these dimensions. One the one hand, emergency measures have supported
firms and families from the economic side. On the other hand, these measures have imposed severe
restrictions, such as the limitation of public events, curfews, and the closure of shops and services,
which sacrificed social interaction to minimise COVID-19 infections.
The evidence illustrated in this report brings forward several contributions on the role that individual
attitudes, working conditions and family relationships have played on gender disparities. Our analysis
aims to provide guidance on empirical evidence addressing the following questions: to what extent
does gender segregation by sector and occupation exacerbate differences among men and women in
terms of health, wage and employment? Does the family play a role of safety-net also under these
special circumstances or does it amplify gender differences, due to unbalanced time-use and
increased needs of childcare? Are there differences between men and women in terms of compliance
with policies such as social distancing, testing and vaccination, and if so, what drives these
5 Galasso, V., V. Pons, P. Profeta, M. Becher, S. Brouard and M. Foucault (2020), “Gender differences in COVID-
19 related attitudes and behavior: Evidence from a panel survey in eight OECD countries.” Proceedings of the
National Academy of Sciences of the United States of America 117, 27285–91.
6 Todorovic, J., Van Kerm, P., and Peluso, E. (2021b). “Time use, childcare and home schooling.” LISER-MEGA
series on gender issues in the COVID-19 pandemic.

- 7 -
differences? Do household composition, age and working conditions generate differences in the risk of
loneliness and social isolation or feelings of economic insecurity?
The report develops on three chapters, each aiming at providing separate answers to the questions
highlighted above in each of three distinct domains.
Chapter 1 covers the health effects of the COVID-19 crisis, by analyzing the gender gradient of
infections, severe illness and deaths during the first 18 months of the pandemic in Luxembourg. The
main messages are that (i) if men and women show similar figures at the aggregate level, gender
differences in health are hidden behind age, employment sector and especially family structure and (ii)
men have been much more adversely hit by severe forms of the COVID-19.
Chapter 2 explores how individual traits affect gender differences in complying with the policies
implemented during the COVID-19 crisis in Luxembourg. It shows that attitudes towards social
distancing (mask-wearing, hand-washing, physical distancing), testing and vaccination depend on
socio-demographic factors, on deep personality traits such as risk aversion, as well as beliefs (trust in
science and in the government) and the type of media consumption. These individual characteristics
are globally similar between women and men, with the exception of the perception of COVID-19’s
danger and risk aversion, which are more pronounced among women. Still, even after taking these
characteristics into account, women appear to be more compliant than men with most measures. A
notable exception pertains to vaccination intentions, which are slightly lower among women due to
stronger concerns about side effects.
Chapter 3 completes the analysis by focusing on gender differences of COVID-19 impacts along three
domains: the labor market and time use, economic insecurity, and social interactions. Among several
other results, we observed that women lost their job more often than men during the pandemic.
Additionally, they were more likely than men to benefit of the special leave for family reasons at the
beginning of the pandemic and more likely to be in temporary unemployment in the spring of 2021.
The gender gap also increased in terms of unpaid work among employed men and women in June
2020, which was qualitatively larger than it was before the pandemic. When it comes to perceptions of
economic insecurity, single men and women worry the same or more about their finances and the
economy in 2021 (as compared to April 2020). For those in a couple, instead, we find evidence of an
insurance effect: partnered men and women worry the same or less about their finances and the
economy in 2021 (as compared to April 2020). Last the Chapter shows that women in Luxembourg
had a significantly lower amount of social interactions than men, the gender difference being larger
among the single.

- 8 -
Chapter 1: A gendered disease? Gender
differences in contaminations and severe forms of
COVID-197
1. Introduction
The aim of this first chapter is to examine gender differences in the most immediate impacts of the
outbreak of COVID-19, namely contaminations with the virus and the development of the disease.
Looking back over the first twenty months since the first cases were diagnosed in Luxembourg –on
February 29 2020–, this chapter documents whether women’s health has suffered more than men’s
from the “Severe Acute Respiratory Syndrome Coronavirus” (SARS-CoV-2) infection disease (COVID-
19) in the country.
Why could we expect gender differences in contaminations and in the prevalence of COVID-19? Two
factors need to be distinguished. The first concerns exposure to the virus, the SARS-CoV-2. Exposure
is determined by the extent and nature of social interactions. While these are partly determined by
one’s own actions and behavior (such as limiting contacts, respecting social distancing and isolation
recommendations), they are largely influenced by people’s environment and their living and working
conditions. Family sizes and household structures are key to private interactions. In the professional
sphere, social contacts vary substantially with one’s occupation – notably to the extent that the activity
can be done “remotely” through teleworking arrangements. For the most part, this environment was
set before the onset of the pandemic and individuals had little control over it when the virus struck.
The variation in the (potential) exposure to the virus according to one’s environment means that not
everyone was equally vulnerable. Given gender differences in occupational profiles and in gender
roles in the households, it is easy to think that such vulnerability might have varied along the gender
dimension.
The second factor concerns the risk of developing severe forms of COVID-19 once one is infected.
Infections by the SARS-CoV-2 have had very different health consequences on different people; many
developed mild forms of the disease, some remained asymptomatic, but a fraction of the population
developed severe forms of respiratory complications that necessitated hospitalization and sometimes
had fatal consequences. It quickly emerged that age was a key determinant of the risk of developing
severe forms of the disease. But a series of risk factors were also identified, such as (in no particular
order of importance, nor exhaustivity) having pre-existing chronic kidney, lung or liver diseases,
cancer, diabetes, heart conditions, substance use disorders, mental health conditions, overweight and
obesity, or pregnancy. The extent to which the prevalence of these risk factors may differ across
gender can lead to differences across gender in the risk of suffering from severe forms of COVID-19.
Pregnancy is without a doubt the most gender-biased of such risk factors. Beyond these pre-existing
risk factors, potential gender differences in biological pre-disposition to the specific development of
COVID-19 (related to hormonal, immune and inflammatory responses to the infection) were later
identified (see, e.g., Conti and Younes, 2020). So, not only may have men and women’s exposure to
the virus been different, also the risk of suffering its most severe forms may have differed. As The
Lancet put it “Women and men are affected by COVID-19, but biology and gender norms are shaping
the disease burden” (The Lancet, 2020).
7 Authors: Philippe Van Kerm, Ioana Cristina Salagean, and Fofo Senyo Amétépé.

- 9 -
Various studies across the world have now examined these questions. The picture that emerges
globally is that contaminations have not differed much between men and women, but that men have
been much more likely to develop the most severe forms of COVID-19, those requiring intensive care
admission or that lead to the death of the patient; see, e.g., Lakbar et al. (2020) Peckham et al.
(2020), Ya’qoub et al. (2021) and the “Sex, Gender and COVID-19 Project” online global data tracker
on gender differences in COVID-19 health impacts (Global Health 50/50, 2021). We examine in this
chapter how the situation in Luxembourg compared to the global picture. Did we also not observe
difference in contaminations? Did we also see men suffer from severe COVID-19 complications in
much greater proportions – in spite of the broad availability of care and ability of the health care
system to absorb the inflow of patients throughout the pandemic?
Exploiting rich population data on test results and hospitalizations n Luxembourg between March 2020
and October 2021, this chapter looks for signs of differences in contaminations between men and
women and for gender-related excess mortality or morbidity from COVID-19. The results presented
draw on data compiled in the context of the project “Santé Pour Tous” initiated by the Luxembourg
Ministry of Health, in collaboration with STATEC (Institut national de la statistique et des études
économiques), the Luxembourg Institute of Socio-Economic Research, the Health Directorate (DiSa)
and the Inspection générale de la sécurité sociale du Luxembourg (IGSS). The project had set out to
study social inequalities in health and to draw lessons from the COVID-19 pandemic in Luxembourg
using administrative records provided by the DiSa (on hospitalisations, test results, death records,
vaccination) and the IGSS (on several socio-demographic and economic variables). In a first report,
Van Kerm, Salagean and Amétépé (2022) (henceforth VKSA) present a “social cartography” of
COVID-19 among Luxembourg residents – illustrating how different groups defined by a number of
social, demographic and economic characteristics have been impacted by the SARS-CoV-2 and the
COVID-19. Although gender is one of the characteristics examined in the study, the VKSA report
provides relatively limited detail about how men and women have been affected. Most notably it does
not document how the gender gap in COVID-19 health impacts has varied by age or according to
different private and professional environments. The present chapter summarizes the VKSA results on
gender and then extends them to provide a fuller picture of how men and women’s health has been
directly affected by the pandemic.
The results generally confirm international evidence: Luxembourg has been no clear outlier. On the
whole we do find some marginally higher risk of contamination for women than for men. However, we
also show that one ought to look behind the surface of aggregate gender differences: below the age of
50, contaminations were noticeably more frequent among women than among men. Above the age of
50, the pattern reverses however and old age men were more likely to be contaminated. Examination
of severe forms of COVID-19 confirm the (much) greater vulnerability of men towards developing the
most critical conditions. Men in Luxembourg were approximately two times more likely to die from
COVID-19 or to need admission in intensive care. This relative risk appears particularly big, when
compared to international estimates provided by Global Health 50/50 (2021).
2. Data and methods
2.1 The data
The data put together for the Santé Pour Tous project used in the VKSA report and re-examined here
combine information on a range of socio-demographic and economic characteristics of the population
residing in Luxembourg at the end of 2019 – just before the onset of the pandemic. Variables drawn

- 10 -
from registers of the IGSS contain data on age, sex, household income, household composition,
employment status, country of birth, canton of residence, and receipt of unemployment benefit. The
information is available for the population of residents in Luxembourg in December 2019 aged 6 and
above and covered by the national social security. This represents a population of 48170 residents for
whom all variables are available.
These variables are linked to the registers held by the health directorate on (i) COVID-19 tests
conducted in Luxembourg between March 1 2020 and October 27 2021; (ii) hospital admissions in
Luxembourg with a COVID-19 diagnostic in the same period (and indication whether the stays
required admission in an intensive care unit); (iii) death records with COVID-19 identified as main
cause of death.; and (iv) vaccine injections.
The data on test results allow us to assess the spread of contaminations in the country over the first
twenty months of the epidemic, separately for men and women of different ages and different socio-
demographic background. In total 13.81% of the subjects have been tested positive to a SARS-CoV-2
contamination. With data ending on October 27 2021, the latest wave of infections with the omicron
variant is not taken into account. Also, we only consider “detected cases” confirmed by a positive PCR
test in Luxembourg. In spite of the large number of tests conducted in the country (notably through the
Large Scale Testing infrastructure) it is possible that some infections have remained undetected,
especially when they were asymptomatic and at the early stages of the pandemic. We do not see any
strong reason however why the possible under-coverage would systematically bias estimation of the
gender differences in infections.
Registers of hospital admissions with a COVID-19 diagnostic and death registries allow us to examine
severe forms of the disease. These indicators are also robust measures of the spread of the most
severe forms of the disease in the population – and how it varied across gender -- which do not suffer
from the possible underestimation that may affect contaminations. In total, 0.78% of subjects (3771
people) have been admitted to hospital with a COVID-19 diagnostic, 0.11% were admitted to intensive
care and 0.16% died with COVID-19 identified as primary cause of death.
Finally, listings of vaccine injections allow us to assess the participation to the vaccination campaign
across genders. With data availability ending in October 2021, the campaign for the ‘booster shot’ is
not taken into account but the data cover a period during which residents of all age groups have been
invited to obtain vaccination.
2.2 Methods
Our objective in this chapter is straightforward: we compare contamination rates, hospitalization rates,
ICU admission rates, death rates, and vaccination rates for men and women – on the whole and for
some specific population subgroups (defined by age, household structure, etc.). We refer to gender
differences in those rates as “gender gaps” in COVID-19 impacts.
However, men and women differ in many different respects. Notably, employment rates are higher
among men. Or there are more lone mothers than lone fathers. To the extent that employment status
or household structure are also correlated with exposure to COVID-19, these might contaminate the
comparison of COVID-19 impacts for men and women. To address this concern, we therefore
compare women’s rates to those of a set of men whose observable characteristics are similar to those
of women. The ‘observable characteristics’ are employment status, age, household income,
household structure and country of birth. Note that these characteristics are measured at their value in
December 2019 (for income and household structure) and in February 2020 (for employment and

- 11 -
age). Throughout the text below, estimates for “men” must therefore be interpreted as “men with
observable characteristics similar to those of women” or “adjusted men” for short.
Technically this is done by propensity score reweighting. Each man in our data is attributed an
“adjustment factor” (or weight) which reduces the weight of those men that have a profile (a set of
characteristics) underrepresented among women and that increases the weight of men whose profile
is overrepresented among women. The adjustment factors are computed from logistic regression of
the probability to be a woman conditional on the observed characteristics. When the adjustment
factors are applied, the frequency distributions of characteristics of the male subjects in the dataset
(that is, of employment, of household structure, etc.) are the same as the those of women. Applying
these adjustment factors to the calculation of contamination or hospitalization rates for men provides
our estimates for male rates adjusted for differences in characteristics. Note however, that we can only
adjust for characteristics that are observable in the data. One important missing variable is the level of
education. Housing conditions are also unavailable. These are likely reflected in the income variable,
but probably only partly.
A further elaboration of this adjustment mechanism is applied to examination of severe forms of
COVID-19. A number of factors or co-morbidities have been identified by the medical and
epidemiological literature as risk factors for the development of respiratory complications after infection
by the SARS-CoV-2. These include a range of pre-existing pathologies, such as diabetes, cirrhosis,
cancer, or other conditions such as obesity. Some of these risk factors may be more prevalent among
men than among women (or vice versa). It is useful to try and assess how much gender differences in
these risk factors account for gender differences in the development of severe forms of COVID-19.
The data collection of Santé Pour Tous provides an indicator of pre-existing pathologies. The variable
captures whether subjects have received drug prescriptions for 0, 1, 2 or 3 and more diseases among
a range of diseases identified as COVID-19 risk factors. We have therefore developed a second
adjustment that incorporates this variable in the construction of the weights. Application of this second
set of weights to our male subjects makes them comparable to women in terms of the prevalence of
these risk factors. The remaining gender difference in hospitalization rates after application of the
weights to the male subjects is “net” of the effect of the measured risk factors and therefore reflect the
contribution of other causes. One must however bear in mind that our indicator of risk factors is
somewhat rudimentary as it does not capture the severity of the pre-existing pathologies and ignores
some important risk factors such as obesity, or smoking.
3. Gender differences in SARS-CoV-2 contaminations
We examine first whether men and women have been contaminated by the virus in similar
proportions.
It is generally agreed that the virus has been infecting men and women in similar proportions.
Estimates from the VKSA report indeed show that just under 14% of both men (13.70%) and women
(13.89%) aged six and above residing Luxembourg have been “confirmed cases” between March
2020 and October 2021. After adjusting for a number of potential factors that might affect exposure to
the virus (notably age, employment status, household income, household structure and country of
birth), VKSA find some more noticeable gender differences in expected infections: 14.10% for women
against 13.51% for men. The difference remains small.
The fact that differences in infections appear to increase after adjusting for differences in a number of
characteristics suggests that, although, on the whole, gender differences in contaminations are small,

- 12 -
this might hide variations in some particular sub-populations. We therefore extend here the
estimations presented in VKSA and probe into gender differences for more detailed groups of the
population.
The risk of contamination is expected to vary greatly with age. The environment (the school, the job,
the family) and, correspondingly, the number and nature of human interactions that we are engaged in
differs drastically as people age.
Figure 1.1 shows infection rates for men and women in sixteen different age groups. The rate for
women is marked by a purple dot; the rate for men in the same age group is marked by a green
triangle. The length of the arrows represents the difference between the two rates. It is red and upward
pointing if women have higher infection rates; it is green and downward pointing if women have lower
infection rates. Recall that the rates for men are “adjusted” for a set of observable confounders as
described in Section 1.2. Most importantly here the procedure adjusts for differences between men
and women in household income, employment status, country of birth and family structure. So the
male population reported here has been “adjusted” to be comparable in those dimensions to the
female population.
Figure 1.1 reveals an interesting pattern, hidden from comparison of the aggregate contamination
rates. For all age groups until the age of 50 contamination rates are higher among women than among
men. The differences are most substantial below the age of 30. From the age of 50, the pattern
reverses and contamination rates are lower among women. The differences are most substantial from
the age of 80.
This first set of results is important. It challenges the view that contaminations did not discriminate by
gender. Figure 1.1 shows that contamination rates did indeed vary by gender, but one needs to look at
differences by age groups. The nature of social interactions differs very much by age and it is easy to
think of potential reasons why the gender gap in contaminations varies with age. At younger ages, the
risk of contamination can be driven by one’s employment situation and type of job or the household
structure (notably the presence of children). These factors tend to recede at older ages when
household structures tend to be smaller and people gradually leave employment.

- 13 -
Figure 1.1: Share of the population infected, by gender and age group
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
Following up on the results of Figure 1.1 and to probe further into potential areas where men and
women’s contaminations might differ, Figure 1.2 shows contamination rates for men and women by
employment status. The figure shows contamination rates for inactive individuals (separately for
individuals aged below or above 50), for public and private sector workers, and for the self-employed.
The expectation is that contamination rates may be higher among employed individuals because of
the potential social interactions involved.
Only small gender differences emerge by employment status, however. No difference is observed
among the self-employed. Contaminations are slightly higher among men in the public sector and
among the elderly inactive. Contaminations are higher among women for the younger inactive and
among private sector workers. There is little evidence that infections among women and men vary by
employment status.

- 14 -
Figure 1.2: Share of the population infected, by gender and employment status
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
Examining just “employment” may be too broad to detect differences in infection risks. As has been
widely documented since the onset of the COVID-19 pandemic, exposure to risks differs widely across
types of jobs and occupations. Frontline workers (in the health sector notably) were first exposed to
the virus. More generally occupations differ in the extent of the implied proximity to co-workers and
customers, in the potential exposure to infectious agents, and perhaps most importantly in the context
of the COVID-19 pandemic, in the extent to which the tasks can be conducted remotely by
‘teleworking’ arrangements (see, e.g., Baker et al.,2020; Mongey et al., 2021).
Figure 1.3 shows gender differences in contamination rates by industry of employment. The data are
for salaried workers aged 20 to 65 only. Note that our data only allows a classification of jobs by
industry, not by occupation. Many different occupations are found in all industries. Clerical and
management jobs are likely found in all types of industries (with probably similar degrees of exposure
to the SARS-Cov-2). Industries’ jobs will differ more markedly in industry-specific occupations (such as
nursing in the health sector, sales worker in the trade sector, teachers in the education sector, etc.),
but the share of these industry-specific occupations may differ across industries. Note also that some
industries shown in Figure 1.3 only employ a small number of people in Luxembourg (such as
Agriculture, Interim, Security services, Arts, or Personal care that each employ less than one percent
of resident salaried workers; see VKSA).

- 15 -
Figure 1.3: Share of the population infected among salaried workers, by gender and industry
Note: Salaried employees residing in Luxembourg aged between 20 and 65.
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa)
Figure 1.3 confirms the observation that contaminations on the job did not differ markedly across
gender. In most industries, contamination rates are identical for men and women. The few notable
exceptions are the “HoResCa” (hotels, restaurants, catering services), cleaning services and
agriculture (and to a smaller extent health) where contamination rates have been substantially higher
among women; and the transport and security services where male contamination rates have been
higher. In most of these instances the female contamination rates have been higher in ‘female-
dominated’ sectors, and male contamination rates have been higher in ‘male-dominated’ sectors.
The professional sphere is one big place for social interactions and, therefore, a first candidate for
generating gender differences in contaminations. Results just shown however reveal that, in
Luxembourg, no systematic difference in contamination rates is observed between men and women
along employment characteristics. The second major sphere of social interactions is the family.
Although interactions within the household typically involve less people than on the job, they involve
much closer interactions. Social distancing and other measures to prevent contamination (such as
mask-wearing) do not apply within one’s household. We therefore examine next whether gender
differences in contaminations emerge for different configurations of households.
Figure 1.4 shows contamination rates for men and women living in twelve different household
configurations. First, the figure distinguishes households with children (aged 15 and below) from
childless households. For the latter, households are classified according to the number of adults and
by age of the individual for which contamination is assessed. For the former, the figure shows different
combinations of number of adults and children.

- 16 -
The results shown in Figure 1.4 are striking. For household configurations without children, we
observe very little gender differences in contaminations. Only among singles do we note differences:
single women aged less than 50 appear more likely to be infected than single men aged less than 50;
the pattern is reversed for singles above 50. Once we examine households with children, however,
contamination rates are higher among women, sometimes significantly so. Notably, among single
adults with children, infection rates are much higher among women than men. Also, infection rates are
much higher among women in large households composed of multiple adults and two or more
children.
These results suggest that the presence of children is an important factor driving contamination risks
among adults (note how the infection rates are higher in households with children – for both men and
women), and also that women face a higher risk of contamination in such household configurations.
We conjecture that this pattern echoes gender differences in the time spent in child care. In two-adult
households however the difference in contamination is small; This is not unsurprising and reflects
contamination between partners and between children and adults.
Figure 1.4: Share of the population infected, by gender and household composition
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
The VKSA report documents somewhat surprising and relatively large differences in infections
according to the country of birth of residents in Luxembourg (from less than 10% among residents
born in Germany to more than 25% among residents born in a country of ex-Yougoslavia). While this
partly reflects the impact of differences in the socio-demographic and economic profiles, variables
such as income, employment status or household structure only partly explain differences in infections
across these groups.

- 17 -
Building upon this observation, Figure 1.5 shows gender differences in infection rates for nine groups
based on country of birth. Each group counts at least 10,000 residents in Luxembourg.
Figure 1.5 illustrates the striking differences in contamination rates already outlined in the VKSA
report. Very little variation by gender appears. The largest gender gap observed within any of the
groups concern residents born in Portugal. Infection rates among women born in Portugal are
approximately two percentage points higher than among men born in Portugal. We do not have an
explanation for this difference. We conjecture that the gap could be driven by a higher share of
Portuguese women holding jobs relatively more exposed to the SARS-Cov-2, and/or by the household
structure which, as Figure 1.4 shows, can lead to higher contaminations among women in large
households. Why those factors do not appear to lead to a gender gap in contaminations among
people born in Italy or ex-Yougoslavia remains puzzling.
Figure 1.5: Share of the population infected, by gender and country of birth
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
To summarize, although it is generally considered that the virus SARS-CoV-2 has been “gender-blind”
and contaminated men and women alike, looking behind the surface of the aggregate contamination
rates reveals a more nuanced picture. The overall contamination rates for both men and women are
just under 14% -- at a remarkably similar level. Yet, among young adults in Luxembourg, infection
rates have been higher among women than men, sometimes substantially so (e.g., in the age group
21-24). Among the older age groups the picture is reversed with old age men more likely to be
contaminated than old age women (notably above the age of 80). These differences do not seem to
be related to gender differences in contaminations related to employment status. Relatively little
gender gap is found by industry. Household structure and, notably, the presence of children however
appear more significant. Lone mothers appear significantly more likely to have been contaminated
than lone fathers. Women in households with multiple children and adults also appear to face higher

- 18 -
contamination rates. We therefore suspect that the household structure and the presence of children
partly explains the higher contamination rates among women aged below 50.
Examining contaminations, or more precisely “detected cases”, is important to understand the spread
of the virus and how men and women have been exposed to the risk of contracting COVID-19.
However, the vast majority of people who became infected did not develop any severe form of the
disease. Some cases remained asymptomatic –a number of which may therefore be missing from the
used in this study– and in most cases the symptoms remained relatively benign, albeit unpleasant
(and one could count the isolation measures imposed on infected people leading to limited work,
schooling or social relations as part of the nuisance of contaminations). Nonetheless, a significant
fraction of people developed much more worrisome, severe forms of COVID-19 and required
respiratory assistance in hospitals, sometimes required admission in intensive care units, and
sometimes did not survived. The next section explores gender differences in such hospitalisations and
deaths due to COVID-19.
4. Gender differences in severe forms of COVID-19:
hospitalisations and death
While international evidence has generally not established gender differences in contaminations by
the SARS-Cov-2, ample research has shown that men have been more severely hit than women. Most
notably, COVID-19 mortality rates have been shown to be higher for men than for women (see, e.g.,
Bhopal and Bhopal, 2020, Lakbar et al., 2020). Although Nielsen et al. (2021) argue that such excess
mortality among men may not be specific to contaminations by the SARS-CoV-2 and generally applies
to many infectious diseases, such a gender gap remains a source of concern for inequity in health
outcomes.
As explained in Section 1.2, our data allows us to estimate the share of the population that has been
hospitalised with a COVID-19 diagnostic, the share of these hospitalisations there required entry in an
intensive care unit, and the share of the population that has died in between March 2020 and October
27 2021 with COVID-19 identified as primary cause of death. All these conditions represent the most
severe forms of the health impacts of COVID-19. We are therefore able to assess whether the
patterns observed abroad were also seen in Luxembourg.
Figure 1.6 presents, by gender and age, the share of the population that has been admitted to hospital
at any point in time between March 1 2020 and October 27 2021 following a confirmed or suspected
COVID-19 case. As before the purple dot shows the rate for women, the green triangle shows the rate
for “adjusted” men –men adjusted to have the same socio-demographic characteristics as women
(income, employment, household structure, country of birth)– and the arrows emphasize the sign and
magnitude of the difference between the two. Two markers are added. The green circle shows the
“unadjusted” rate among men. The position of the circle relative to the green triangle gives a sense of
how important is the impact of the adjustment for socio-demographic characteristics. The green cross
shows a second version of “adjusted” men. The second adjustment not only corrects for differences in
socio-demographic characteristics between men and women, but it also corrects for differences in the
prevalence of pre-existing pathologies observed prior to the onset of the pandemic and which have
been shown to be factors of risk in the development of severe forms of COVID-19 (it includes
treatment for diabetes, cirrhosis, chronic respiratory diseases, cancer, etc. – see VKSA for details). To
the extent that such risk factors may be more prevalent among men, they could explain the higher
male mortality. Our second “adjusted” male rates eliminate this potential source of gender differences

- 19 -
in hospitalisations for COVID-19. The “adjusted” male rates correspond to rates that would be
observed if males experienced similar prevalence of pre-existing pathologies as women.
Unsurprisingly hospital admissions are largely determined by age. The share of hospitalised people
increases gradually from the 18-20 years old group, then starts going up more significantly from the
age of 40, then increases substantially from the age of 65. More than one in twenty people aged 85 or
above in March 2020 have been admitted to hospital because of COVID-19 (and more than half of that
number died of COVID-19, as we show below).
Beyond the relationship with age, the difference in rates between men and women is also clear. In
almost every age group, men were more likely to be hospitalised for COVID-19 than women.
Let us however highlight the exceptions first: the age ranges running from 21 through to 39 years old.
In absolute terms, hospitalisation rates are low in these ages – well below 5 per thousand people –,
however the relative differences between men and women are large. Between 25 and 34 years old,
the hospitalisation rates are twice larger for women than for men. This must partly be a consequence
of the higher infection rates shown in Figure 1.1 among women, yet these relatively small differences
in contamination cannot, alone, explain the differences in hospitalisations in this age range. While we
would like to stress this observation, we cannot offer any other explanation for this added
hospitalisation risk among women of prime child-bearing age. Note that the gap persists when we
examine the “adjusted” males that take pre-existing pathologies into account.
The risk of hospitalisation for men starts increasing from the age of 35 to 59 and then accelerates to
reach its peak above 85. Among women hospitalisation rates only grow slowly until the age of 65 and
then start increasing to peak beyond the age of 90. These different age-hospitalisation relationships
imply that hospitalisation rates for men are higher at every age from the age of 40 onwards. At age 80-
84 men are one percentage point more likely to be hospitalised for COVID-19 (at just above 4%
chance) than women (at just above 3% chance), and at age 85-89 the gap reaches almost two
percentage points. Perhaps surprisingly, the adjustment for differences in pre-existing pathologies has
only little influence on these results. The source of these gender differences in old age have to be
found in other, possibly more complex, biological factors.
Figure 1.6: Share of the population admitted to hospital with COVID-19 diagnostic, by gender and age

- 20 -
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
We highlighted above differences in contamination rates household structures. Figure 1.7 shows that
the higher contaminations of women observed in households with children does not translate into
higher hospitalisation risks. On the contrary, men have higher hospitalisation rate in almost all
household types. The gap is, unsurprisingly, larger in household structures related to old age –
echoing the higher hospitalisation rates of old age men. Hospitalisation rates of men remain higher in
households with children (except for lone parents with one child).
Figure 1.7: Share of the population admitted to hospital with COVID-19 diagnostic, by gender and
household composition

- 21 -
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
In their most severe forms, the COVID-19 disease required admission in intensive care units (ICU).
Figure 1.8 shows the rates of ICU admissions by gender and age. We only report rates for the
population aged above 50 (the number of cases for younger age groups is too small to report). Gender
differences in ICU admissions are striking. The risk of ICU admission for men is twice as large as
women’s in almost all age groups. Again, adjusting for pre-existing pathologies hardly makes any
difference to the gender gap in admissions.: all else equal, men were admitted in ICU much more
frequently than women.

- 22 -
Figure 1.8: Share of the population admitted to intensive care for COVID-19 complications, by gender
and age
Note: Calculations based on residents aged 50 and above.
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
The final picture is given by gender differences in mortality due to COVID-19, depicted in Figure 1.9.
These estimates confirm that Luxembourg is no exception and that COVID-19 killed many more men
than women. Similarly to ICU admissions, men’s risk of dying from COVID-19 have been about twice
as large as women’s. For example, COVID-19 has been the primary cause of death of 3.5% of women
aged 90 and above, and of 6% of men aged 90 and above.

- 23 -
Figure 1.9: Share of the population dead of COVID-19 infections, by gender and age
Note: Calculations based on residents aged 50 and above.
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
In sum, Luxembourg has been no exception: men (and notably old men) have been hit by severe
forms of COVID-19 much more frequently than women. Infections that lead to complications requiring
ICU admissions or that lead to the death of the victim where generally around twice more frequent
among men than among women. Such a pattern has been shown in many other countries.
These most severe forms of COVID-19 impacts are largely concentrated on the old-age population.
Much of the explanations for the greater vulnerability of men to fatal infections are therefore to be
sought in factors observed in old-age. While these patterns are generally true for all hospital
admissions caused by COVID-19, the gender gap is less striking when we consider hospitalizations
that did not require ICU admissions. Furthermore, the female advantage disappears at younger ages
and we observe that women of child-bearing age were in fact twice more likely to be hospitalized than
men of similar ages. Just like for contaminations, looking under the surface of aggregate rates reveals
patterns that are more complicated than what is expected.
5. Is vaccination the explanation? Gender differences in
vaccination rates
The vaccination campaign started in Luxembourg in December 2020, some ten months after the first
case officially detected in the country. By the Summer 2021, all residents have had the opportunity to
receive a complete vaccination protocol.
Vaccination has been shown to reduce drastically the risk of developing severe form of COVID-19
among people infected by the SARS-CoV-2. So, can the large gender differences in hospitalisation
rates between men and women be explained by differences in vaccination? The answer is no. First,

- 24 -
and most obviously, gender differences in mortality were observed in 2020, before the availability of
vaccines. Second, as we now show, men and women have adopted vaccination in very similar
proportions.
Figure 1.11 shows the share of the population fully vaccinated (as of October 27 2021) by age and
gender. Women have generally been more likely to be vaccinated –notably in the age age 25 to 50–
but the difference with men’s vaccination rates is small. In old ages, when vulnerability to COVID-19 is
the strongest, no gender difference in vaccination appears (with the only notable exception of the 90+
among whom men were much more likely to be vaccinated).
Figure 1.11: Share of the population vaccinated against COVID-19, by gender and age
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
Participation to the vaccination campaign is one of the key instruments available in the fight against
the pandemic. In Luxembourg, vaccines have been made available to all residents, free of charge. In
this context, being vaccinated is largely an individual decision. In spite of the ease of access to the
vaccines, vaccination rates remain comparatively low – at least in comparison with neighboring or
other European countries. VKSA document a “social gradient” in vaccination, with vaccination rates
varying with household income and country of birth, most notably.
The variation in vaccination by country of birth is particularly puzzling. Figure 1.12 shows these
vaccination rates separately for men and women. Clearly the variations in vaccination by country of
birth is observed for both sexes. Although women are somewhat more likely to be vaccinated than
men among residents born in Germany, Portugal or a country for the former Yugoslavian Republic, the
gender gap is small compared to the variations in vaccination across countries. The gender
perspective does little to explain the puzzle of the variations in vaccination rates by country of birth.

- 25 -
Figure 1.12: Share of the population vaccinated against COVID-19, by gender and country of birth
Source: Santé Pour Tous (calculations based on registers provided by IGSS and DiSa).
6. Conclusion
The answer to the question ‘Have men and women’s health been impacted in similar proportions by
the pandemic?’ is a clear “no”. The most dramatic consequences of infections by the SARS-CoV-2
appear to have been much stronger on men than on women, notably old age men. In Luxembourg,
like in other countries around the globe, women generally appear to have had a better biological
response to infections than men. This apparent better immune response (Conti and Younes,, 2020;
Nielsen et al., 2021) is fortunate because we show that women have been more frequently
contaminated than men, albeit relatively marginally and mostly among relatively young populations
who were at lower risk of developing severe form of the disease. The increased contamination risk of
young women appears to be related to household structures and the presence of children notably,
more than to employment characteristics.
The Lancet’s editorial on “the gendered dimension of COVID-19” stated that “The success of the
global response—the ability of both women and men to survive and recover from the pandemic's
effects—will depend on the quality of evidence informing the response and the extent to which data
represent sex and gender differences” (The Lancet, 2020). The analysis conducted here and
presented in this chapter can be seen as a contribution towards this goal, one which illustrates the
value of data accessibility for research purposes.
Although the results shown here are based on rich data, the analysis is not without limitations and
much critical information is also missing. Notably, the information on employment remains relatively
coarse. The fact that we do observe much gender differences in contaminations related to
employment may be because we are not able to see differences in the types of occupation at a

- 26 -
sufficiently detailed granularity. Also, to be able to design the most effective prevention measures, one
would ideally want to try and understand how much gender differences in severe outcomes of COVID-
19 are driven by differences in the prevalence of a compete set and detailed risk factors. The limited
information on pre-existing conditions prevents us from making detailed recommendations here.
Future more detailed epidemiological studies will certainly fill this gap. Finally, there is one outcome
which we have not examined at all: the prevalence of “long COVID” (generally viewed as the
persistence of symptoms such as fatigue or shortness of breath long after the initial infection). Some
research suggests that women are more likely affected by persisting symptoms (e.g., Bai et al., 2021)
– this is however a dimension that our study is not able to document.

- 27 -
References
Bai F, Tomasoni D, Falcinella C et al. (2021). Female gender is associated with long COVID
syndrome: a prospective cohort study. Clinical Microbiology and Infection, in press.
https://doi.org/10.1016/j.cmi.2021.11.002.
Baker MG, Peckham TK, Seixas NS (2020). Estimating the burden of United States workers exposed
to infection or disease: A key factor in containing risk of COVID-19 infection. PLoS ONE 15(4):
e0232452. https://doi.org/10.1371/journal.pone.0232452
Bhopal SS, Bhopal R (2020). Sex differential in COVID-19 mortality varies markedly by age. The
Lancet Correspondence. 396(101250), 532-533.
Conti P, Younes A. (2020). Coronavirus COV-19/SARS-CoV-2 affects women less than men: clinical
response to viral infection. Journal of Biological Regulators and Homeostatic Agents 34(2),339-343.
doi: 10.23812/Editorial-Conti-3. PMID: 32253888.
European Commission, Directorate-General for Research and Innovation, Oertelt-Prigione, S. (2020).
The impact of sex and gender in the COVID-19 pandemic: case study, Publications Office,
https://data.europa.eu/doi/10.2777/17055
Global Health 50/50 (2021). COVID-19 Sex-disaggregated Data Tracker.
https://globalhealth5050.org/the-sex-gender-and-covid-19-project/ (accessed February 14, 2021).
Lakbar I, Luque-Paz D, Mege JL, Einav S, Leone M. (2020). COVID-19 gender susceptibility and
outcomes: A systematic review. PLoS ONE, 15(11), e0241827.
https://doi.org/10.1371/journal.pone.0241.
Mongey S, Pilossoph L, Weinberg A (2021). Which workers bear the burden of social distancing?
Journal of Economic Inequality, 19, 509–526. https://doi.org/10.1007/s10888-021-09487-6
Nielsen J, Nørgaard SK, Lanzieri G, Verstergaard LS, Moelbak K (2021). Sex-differences in COVID-19
associated excess mortality is not exceptional for the COVID-19 pandemic. Nature (Scientific Reports)
11, 20815. https://doi.org/10.1038/s41598-021-00213-w
Peckham H, de Gruijter NM, Raine C. et al. (2020). Male sex identified by global COVID-19 meta-
analysis as a risk factor for death and ITU admission. Nature Communications 11, 6317.
https://doi.org/10.1038/s41467-020-19741-6
The Lancet (2020). The gendered dimensions of COVID-19 (Editorial). The Lancet, 395, 1168.
Van Kerm P, Salagean, IC, Amétépé, FS (2022). La COVID-19 au Luxembourg. Le gradient social de
l’épidémie. Rapport Santé Pour Tous #1. Ministère de la Santé, Luxembourg.
Ya'qoub L, Elgendy IY, Pepine CJ (2021). Sex and gender differences in COVID-19: More to be
learned! American heart journal plus: cardiology research and practice, 3, 100011.
https://doi.org/10.1016/j.ahjo.2021.100011

- 28 -
Chapter 2: Gender differences in attitudes towards
COVID-19 and health measures8
In this work package, we review the differences between women and men in terms of compliance with
COVID-19 measures. To this end, we have developed a questionnaire about attitudes towards
COVID-19 measures, behavioral traits, and concerns about COVID-19 and trust in institutions. This
module was incorporated in an online survey which was administered in March 2021 on a sample of
citizens from Luxembourg and its neighboring regions. It allows us to capture the attitudes of almost
700 individuals (mostly Luxembourgish nationals) towards physical distancing, mask-wearing, testing,
vaccination, and support and consciousness towards measures in general.
We first show that rates of compliance with these measures are rather high for both women and men,
though this depends on the measures considered. Large scale testing and avoidance of physical
contacts are adopted by more than 85% of respondents. More than 75% also intend to get vaccinated
and perceive that mask-wearing is a civic duty. About two thirds of respondents try to avoid public
places and always wear masks in such places, get tested if they feel sick or had a high-risk contact.
The same proportions of respondents remains as careful in applying safety measures as they were in
the pandemic’s beginning, and are globally supportive of the government’s long-term actions. The
domain in which individuals struggle the most pertains to being systematic in the application of safety
measures. Indeed, about 48% of respondents admit that they do not manage to apply these measures
consistently throughout the day as they tend to forget them once in a while.
We then analyze gender differences in these various outcomes. The proportion of women who comply
with health measures is higher than men’s for almost all measures. This is particularly true for mask-
wearing, consistent application of safety measures, proactive testing and general support of the
government’s actions, for which the proportion of women is about 7 to 11 percentage points higher
than men’s. The proportion of women who avoid contacts and public places and participate in large
scale testing is also larger, though only by 4 to 6 percentage points. The proportion of individuals
intending to get vaccinated is the same across genders.
In the main section, we introduce multiple individual characteristics as potential factors explaining
compliance with health measures. This allows us to refine our understanding of gender differences.
We find that about half of these differences results from the fact that women are more risk averse and
that they are more likely to perceive COVID-19 as dangerous for their health. On the other hand,
women’s lower average educational attainments negatively impact compliance.
When we compare of women and men with identical socio-demographic and personal characteristics,
we find that women’s compliance remains higher than men’s in several outcomes. In particular, the
proportion of women behaving responsibly and conscientiously in terms of mask-wearing and safety
measures is about 6 percentage points higher than that of men with similar education, risk aversion,
perception of danger and other individual characteristics. Women are also more involved in testing,
though the difference with men is only about 3 percentage points. On the other hand, proactive risk
avoidance through physical contacts and public places is identical between women and men with
similar traits. The only dimension in which women are less involved than men pertains to vaccination
intentions, which is 5 percentage points lower than men’s.
8 Author: Bertrand Verheyden.

- 29 -
We shed more light on the lower vaccination intentions of women in the last section. To this end, we
exploit the natural experiment provided by the wave of suspensions of Vaxzevria (Astrazeneca’s
vaccine) in the European Union during the first half of March 2021, right at the start of our data
collection process. We find that vaccination intentions of both women and men declined until March
18th, when the European Medicines Agency (EMA) provided public reassurances that the vaccine was
effective and safe. Although this decline was sharper for men, the EMA’s statement was more
effective on them than on women. Our data shows that women’s vaccination intentions remained more
dispersed after March 18th, suggesting that the EMA was less effective at restoring trust among
women. This is sensible, considering the fact that side effects and rare blood clots were
overwhelmingly reported on women, and that women are more risk averse than men in general.
Overall, this report provides evidence that women are by a reasonable margin more compliant than
men, and that this difference goes beyond the fact that they are more risk averse and more worried
about the impact of COVID-19 on their health. All else equal, only risk avoidance appears to be
identical between women and men, whereas vaccination is the only dimension in which women have
lower participation intentions.
These results suggest that women and men are not sensitive to COVID-19 and health policies in the
same way, and therefore communication about the importance of these policies might gain to be
gender-specific. For instance, men should be specifically alerted about the risks of COVID-19 and
about the benefit of policies, whereas concerns related to vaccination should concentrate on
reassuring women. Also, it appears that communications to women are more effective at stimulating
vaccination intentions if they are made by women (Chang et al., 2021).
1. Introduction
In order to mitigate the adverse effects of COVID-19 in terms of life losses, of pressure on health
systems, and of economic costs, policy makers have implemented a number of health policies. These
measures imply different forms of constraints on the population’s interactions (safety measures, mask-
wearing) and freedom of movements (quarantines, school and workplace closures, cancelling of public
and private events, curfews,...). Large scale testing as well as self-testing is important for assessing
the evolution of the number of infections and to prevent contaminations, but are tedious or
uncomfortable to many individuals. Last but not least, vaccination policy is particularly polarizing,
being perceived by some as a welcome solution, as a civic duty, but by others as an attempt on their
freedom of choice and on their physical integrity.
All these measures are demanding significant efforts from the population in already difficult times.
Furthermore, they are in many situations difficult, or impossible, to enforce, and may even face
opposition by some parts of the population. It is therefore crucial to encourage the population in
complying with these health measures so that they are integrated as norms. In a context of equal
opportunities, this is particularly critical since groups which are facing socio-economic disadvantages
in general tend to be those who suffer the most from this health crisis. In particular, other work
packages of this project show that women tend to be more affected by COVID-measures than men.
In this report, we study gender differences in attitudes towards a number of policies such as mask
wearing, social distancing, testing and vaccination, as well as perceptions towards some of these
measures and the government’s actions in general in this crisis. We also compare the degrees of
resilience in compliance, both in the short term (whether people remain equally careful throughout the

- 30 -
day) and in the long term (whether people tend to be less careful than in the beginning of the health
crisis).
Recent evidence establishes the existence of gender differences in compliance with most of these
policies. Galasso et al. (2020) use data from a multi-country study over the months of March and April
2020. They report that women are more likely to consider the virus as a serious risk than men. Based
on various public-health and social-distancing measures, the authors build an overall index of
compliance, which they use to show that the proportion of compliant women is about 5 percentage
point higher than men. These findings are in line with ours, though we find differences in terms of
vaccination and testing, which were not in place at the time of their study. Higher compliance by
women is consistent around the world across various types of measures (Lin et al., 2021; Rosha et al.,
2021; Alshammary, 2021). As in our analysis, the only exception to women’s higher compliance
pertains to vaccination. Indeed, women have lower intentions to get vaccinated (Green et al., 2021).
Explanations to women’s higher compliance are multidimensional. In this report, we stress that
women’s characteristics make them more prone to comply with health measures, in particular risk
attitudes and fear of COVID-19. The literature indeed shows that women tend to be more emotionally
affected by this health crisis than men (Alsharawy et al., 2021 ; Levkovich and Shinan-Altman, 2021 ;
Liu et al., 2020; Park et al., 2020 ; Wang et al., 2020 ; Xiao et al., 2020). Possible explanations to
these stronger emotional reactions are that women’s concerns actually pertain more to their family’s
health than financial aspects (McLaren et al., 2020), and that their perceptions of COVID-related
health risks are stronger than mens’ (Rodriguez-Besteiro, 2021), and that they are thus more averse to
being exposed to the virus (Kowalik and Lewandowski, 2021). Second, women may be more careful
since the objective risk that they face is higher than men’s. Indeed, they disproportionately operate in
high-risk occupations and sectors, such as education and health (Barbieri et al., 2020). Third, the
positive impact of education on compliance appears to be more pronounced among women (Algara,
2021). In line with our analysis, Clark et al. (2021) find that factors such as the feeling of vulnerability
to COVID-19 and trust in government play an important role in explaining compliance.
An interesting note of conclusion of this literature review is that gender differences appear to matter in
terms of leadership style in the face of this health crisis. Garikipati and Kambhampati (2020) indeed
show that the management of the pandemic in countries led by women was superior to that of
countries led by men. A potential explanation for this is that the overall approach of female leaders of
Northern Europe and New Zealand was democratic, inclusive and transparent in comparison with the
management of countries such as the US, the UK or Brazil.
2. Data and descriptive statistics
We exploit data from an online survey conducted among the residents of Luxembourg and the border
regions from early March to mid-April 2021. The survey was organized by Luxembourg Institute of
Socio-Economic Research (LISER) in collaboration with the University of Luxembourg and advertised
at the beginning of March on social media and on some local council websites.
After a general section on demographic characteristics, respondents were redirected to one of four
randomly assigned blocks of questions covering various themes. Our block of interest, which was
specifically designed for this report, concerns attitudes towards COVID-19 measures (social
distancing, testing and vaccination) as well as behavioral traits and beliefs. A total of 2,549 individuals
completed the main survey, about a fourth (689 individuals) completed the relevant block of questions
used in this analysis. However, some individuals did not provide answers to some of the questions

- 31 -
that we exploit in this analysis, in particular those pertaining to personality traits and beliefs. The
question which received the smallest number of answers (642) concerns the behavioral trait of self-
determination, or locus of control.
2.1 Descriptive statistics of respondents’ characteristics
Figure 2.1 (which is drawn from Table A2.1 in Appendix) provides descriptive statistics of individual
characteristics which are exploited in the analysis as determinants of the attitudes towards COVID-19.
These characteristics are split into three categories: socio-demographic characteristics, behavioral
traits, and beliefs. These three categories of individual characteristics will be incrementally introduced
in the statistical analysis (see Section 2.3).
Figure 2.1: Mean values of individual characteristics
Figure 2.2 provides a distinction by gender for the five characteristics which are significantly different
between women and men.9
We start with a description of respondents’ general socio-demographic characteristics. Individuals who
responded to our module are mostly women (67%). A majority of respondents are working (80%), a bit
more than half of the sample has tertiary education (55%), and 59% of individuals have the
Luxembourg nationality. In terms of age, 13% are below 35 years old and 44% are above 50 years
old. In terms of gender differences, the proportion of women with higher education is about 9
percentage points lower than men’s. This is also the case for the proportion of working women, which
is 6,6 percentage points lower than men’s (see Figure 2.2 below).
9 Statistical significance means that it is not plausible that the “true” difference between women and men in the
entire population is equal to zero. More specifically, it means that the probability to be wrong -when rejecting the
possibility that this difference is zero- is sufficiently small (e.g. 5%).
!"#$%
!"#&%
!"'(%
!")$%
!"*(%
!"&)%
!"$(%
!"+#%
!"$$%
!",#%
!"(!%
!"++%
!"#!%
!"+'%
!"&*%
0,0! 0,1! 0,2! 0,3! 0,4! 0,5! 0,6! 0,7! 0,8! 0,9! 1,0!
DOES!NOT!FULLY!TRUST!SCIENTISTS!
DOES!NOT!FULLY!TRUST!THE!GOVERNMENT!
ONLINE!JOURNALS!/!SOCIAL!MEDIA!
GETS!INFORMED!VIA!TV!/!RADIO!/!NEWSPAPERS!
CONSIDERS!COVID-19!AS!DANGEROUS!
BELIEFS!
PATIENCE!
SELF-DETERMINATION!
WILLINGNESS!TO!TAKE!RISKS!
BEHAVIOURAL!TRAITS!
AGE!ABOVE!50!
AGE!UNDER!35!
NOT!WORKING!
HAS!HIGHER!EDUCATION!
SINGLE!
LUXEMBOURGISH!
WOMAN!
SOCIO-DEMOGRAPHIC!CHARACTERISTICS!

- 32 -
Figure 2.2: main differences between women and men’s characteristics
The second category of individual characteristics pertains to behavioral traits, namely the willingness
to take risks, self-determination and patience. These variables are built thanks to questions specifically
designed and validated in the literature to capture these traits, and which have been standardized
here on a scale between 0 and 1. In line with the literature, the average willingness to take risks is
significantly higher for men (0,58) than for women (0,50). Women also exhibit a slightly higher locus of
control (0,43) than men (0,39), which captures the notion of self-determination, i.e. that events
occurring in life result more from one’s own actions than from random events imposed on oneself.
Women and men appear to be equally patient (almost 0,7).
The third category of individual characteristics pertains to beliefs and belief formation. The majority of
our sample considers COVID-19 as dangerous, with women being more concerned for their health
(75%) than men (66%). Women and men appear to get informed about the news through the same
channels, with 84% of the sample consulting every day traditional media (newspapers, TV or radio)
and 92% consulting online media or social media daily. Finally, there are no gender differences in
terms of trust in institutions, with 64% of respondents having a strong confidence in the
Luxembourgish government's action, and 66% having a strong confidence in the scientific community.
While women are similar to men in our sample in most dimensions, Figure 2.2 summarizes the key
dimensions in which gender differences exist. It shows that higher education, employment, and
willingness to take risks are less prevalent among women, whereas fear of COVID-19 is stronger
among women. These dimensions play an important role in attitudes towards COVID-19 measures,
and thus capture an important share of the gender differences which we are now presenting. It is also
worth noting that, conditional on gender, the proportion of Luxembourgish women (64%) is larger than
the proportion of Luxembourgish men (51%) in our sample.
0,4!
0,5!
0,6!
0,7!
0,8!
0,9!
1!
Luxembourgish! Higher!education! Working! Willingness!to!take!
risks!
Considers!
COVID-19!as!
dangerous!
Women! Men!

- 33 -
2.2 Descriptive statistics of respondents’ outcomes
Let us now provide a statistical description of the outcomes that we will study in Section 2.3, i.e.
respondents’ levels of compliance with COVID-19 measures. Figure 2.3.A and 2.3.B (which are drawn
from Table A2.2 in Appendix) provide comparisons of the differences in attitudes towards COVID-19
measures between women and men. A first look at these mean comparisons suggests that women are
significantly more compliant than men in virtually all dimensions.
Figure 2.3.A: Average levels of attitudes towards COVID-19 measures, by gender
Figure 2.3.A pertains to mask-wearing and social distancing and safety measures in general. Women
are more supportive of mask-wearing, with 81% of women considering it as a civic duty, compared to
71% for men. Also, 66% of women claim to always wear masks in public places, while this proportion
is only of 55% for men. Second, women appear to be more conscientious and supportive than men in
terms of social distancing and safety measures in general. Indeed, 54% of women (47% of men) claim
to never forget safety measures throughout the day, 69% of women (60% of men) consider
themselves as careful about applying safety measures in March 2021 as in the beginning of the
pandemic, and 73% of women (63% of men) support the government’s actions against the pandemic.
40,00!
50,00!
60,00!
70,00!
80,00!
90,00!
100,00!
Considers!wearing!
masks!as!a!civic!
duty!
Always!wears!
masks!in!public!
places!
Never!forgets!
safety!measure!
Is!as!careful!as!in!
the!beginning!of!
the!pandemic!
Supports!the!
governement's!
actions!
Woman! Man!

- 34 -
Figure 2.3.B: Average levels of attitudes towards COVID-19 measures, by gender
Figure 2.3.B pertains to testing, proactive risk avoidance and vaccination outcomes. First, it shows that
women are also more involved in testing. The Luxembourgish government put in place a large-scale
testing campaign in view of assessing the spread of the virus, and 88% of women in our sample
participated in it, compared to 82% of men. Also, spontaneous testing, in case of feeling sick or in
case of contact with an infected person, is more prevalent among women (67%) than men (59%).
Second, proactive behaviors of risk avoidance seem slightly more pronounced among women. Indeed,
89% of women and 84% of men avoid physical contacts (shaking hands, kissing, hugging,…) and
70% of women and 64% of men try to avoid public places since the start of the pandemic. Finally, the
only exception to the overall higher compliance of women pertains to vaccination. Indeed, about three
quarters of both women and men intend to get vaccinated.
Though quite informative, these comparisons of average compliance levels between women and men
however have an important limitation. Indeed, these comparisons are “unconditional”, in the sense that
they do not take into account the fact that women and men do not have the same characteristics, in
particular -as we have seen from Table A2.1- in terms of higher education, attitudes towards risk and
fear of COVID-19 and nationality. In the next Section, we apply a statistical method which allows to
measure gender differences that take these characteristics into account.
3. Empirical analysis
In this Section, we pursue the comparison between women’s and men’s compliance with COVID-19
measures in a way that is more robust than simple mean comparisons. Indeed, we use an “Ordinary
Least Squares regression” which allows us to account for the fact that women and men in our sample
have different observable characteristics affecting compliance. This approach allows us to identify the
extent to which the results presented in the descriptive statistics section are attributable to factors that
are correlated with gender, such as education, risk attitudes, and fear of COVID-19. We will see that
these factors indeed play an important role in explaining attitudes towards COVID-19 measures, but
that gender per se remains an important factor once these dimensions are taken into account.
40,00!
50,00!
60,00!
70,00!
80,00!
90,00!
100,00!
Participates!in!
large!scale!testing!
campaign!
Gets!tested!in!case!
or!contact!of!
feeling!sick!
Avoids!physical!
contacts!
Avoids!public!
places!
Intends!to!get!
vaccinated!
Woman! Man!

- 35 -
The results described here are to be interpreted as “ceteris paribus”. This means that, considering the
characteristics that are included in the regression, gender effects result from the comparison of
outcomes between women and men who have the same values of these characteristics, also called
control variables.
To make this interpretation concrete, let us consider a first regression in which the various compliance
outcomes (listed in Figures 2.3.A and 2.3.B) are explained by gender and by the main socio-
demographic characteristics (age, nationality, employment status, marital status and education). In
such a regression, the estimated effect of gender corresponds to the difference between the average
outcomes of women and men belonging to the same age group, having the same nationality, the
same employment status, the same marital status and the same education level. Unlike the
“unconditional” differences presented in Section 2.2, the gender effect estimated here is not affected
by the fact that men and women differ in these characteristics (e.g. nationality and education). It thus
provides a gender effect that is “net” of the differences in observable characteristics.
As we introduce additional sets of individual characteristics (behavioral traits and beliefs), we increase
our ability to explain the outcomes, and we refine our understanding of which variables are key in this
explanation. The sequential addition of potential drivers of compliance also allows us to understand
the roots of gender differences. In particular, this allows us to assess the extent to which gender
differences described in Section 2.2 are driven by the fact that women and men are different in
characteristics that also impact their compliance with COVID-19 measures, and conversely the extent
to which structural gender differences remain even after accounting for these additional
characteristics.
Results of this approach are summarized in Figures 2.4.A and 2.4.B (which are drawn from Table A2.3
to Table A2.7 in the Appendix). Figure 2.4.A reports the effects of being a woman (relative to a man)
on the first five outcomes, i.e. mask-wearing, conscientiousness and support towards safety
measures. Figure 2.4.B reports the effects of being a woman on the last five outcomes, i.e. testing,
proactive risk avoidance and vaccination. For each of the 10 outcomes, four orange dots (representing
the mean difference between women and men) and black lines (representing the corresponding
statistical 95% confidence interval) are presented.
- The first dot on the left represents the unconditional mean difference between women, i.e. a
difference between “raw” means, which does not take into account any other explanatory
factors. These first dots correspond to the differences between women’s and men’s bars in
Figures 2.3.A and 2.3.B.
- The second dot represents the “Woman” effect in a regression which includes basic socio-
demographic characteristics (age, education, nationality, marital status and employment
status). These coefficients (which are displayed in the first columns of Tables A2.3 to A2.7),
must be interpreted as the mean difference between women and men who belong to the same
age group, education category, nationality, marital status and employment status, but who
may still differ in terms of behavioral traits and beliefs.
- The third dot is based on the same approach, with a regression which also includes
personality traits (willingness to take risks, self-determination and patience) as variables
explaining the outcomes. The third dot thus represents the mean difference between women
and men who have the same set of socio-demographic characteristics as well as the same
levels of willingness to take risks, self-determination and patience.

- 36 -
- The fourth and last dot represents the mean difference between women and men having the
same socio-demographic characteristics, personality traits and also the same beliefs (fear of
COVID-19, type of media consumed and trust in government and in science).
The interpretation of each dot is thus different, and one may argue that since, for instance, women are
intrinsically less willing to take risks, comparing women and men who have the same willingness to
take risks is an artificial exercise. However, some men are more risk-averse than some women, and
one should typically avoid stereotypical generalizations when measuring gender effects. In particular,
Paramita et al. (2021) stress that the way gender is treated empirically significantly matters for the
interpretation of gender effects. They indeed show that gender psychology (feminine vs masculine)
and gender-role (traditional vs egalitarian) are more effective at explaining compliance towards health
measures than the dichotomous treatment of sex. Controlling for various personal characteristics is
therefore both informative and relevant.
Figure 2.4.A: Differences between women and men, by outcome and controls
-,!.%
-+.%
!.%
+.%
,!.%
,+.%
(!.%
Considers!mask!
as!a!civic!duty!
Always!wears!
mask!in!public!
places!
Never!forgets!
safety!measures!!
Is!as!careful!as!in!
the!beginning!!
Supports!the!
governement's!
actions!!

- 37 -
Figure 2.4.B: Differences between women and men, by outcome and controls
For each dot, gender differences are statistically significant (i.e. different from zero) if the black line
(the confidence interval) does not cross the blue horizontal axis representing 0%. We start the
discussion of the main results by making some general comments about the relative position of the
four dots which hold for almost all outcomes.
First, for all outcomes except vaccination, the first dot’s confidence interval (i.e. the specification which
does not account for any individual characteristic but gender) does not cross the horizontal axis. This
means that overall, women are significantly more compliant than men, though these men’s
characteristics may be very different from those of women.
Second, almost systematically, the second dot is higher than the first. This means once socio-
demographic characteristics are taken into account, the difference between women and men is even
stronger. This stems from the facts that (i) women are on average less likely to have higher education
than men, and (ii) less educated individuals tend to comply less with COVID-19 policies.10 Hence, the
basic “unconditional” gender difference (first dot) was underestimating the gender effect because it
was incorporating the negative effect of low education, which is more prevalent among women.
Indeed, the gender compliance gap is larger once comparisons are made on women and men with the
same education levels.11
Third, as we introduce additional characteristics (behavioral traits for the third dot, and beliefs in the
fourth dot), we observe that the difference between women and men tends to become smaller, and in
10 The gender difference in education is common, in particular among older cohorts, whereas the lower
compliance of less educated individuals has been vastly.
11 A second socio-demographic characteristic may counterbalance this effect for specific measures. It indeed
appears from our regressions that remaining careful throughout the pandemic and avoiding public places are
facilitated by not being in a work environment. Women, who are less active on the labor market on average, are
thus more likely to comply with these specific measures.
-,!.%
-+.%
!.%
+.%
,!.%
,+.%
(!.%
Participates!in!
large!scale!
testing!
Gets!tested!in!
case!of!contact!
or!feeling!sick!
Avoids!physical!
contacts!
Avoids!!!!!!!!!!!!!!!!!!!!!
public!places!
Intends!to!get!
vaccinated!

- 38 -
many cases statistically insignificant. The reason thereof stems from similar mechanisms, although
they apply here in the opposite direction. First, the third dot is generally lower than the second
essentially because it takes risk attitudes into account. Indeed, women tend to be more risk averse,
and both the literature and our analysis confirm that high risk aversion is associated with higher
compliance with COVID-19 measures. Hence, the inclusion of behavioral traits highlights the fact that
the first two dots were estimating gender differences which were incorporating the fact that women are
more compliant because they are on average more risk averse. This explains why the third dot is
lower than the first two dots, as it is based on a comparison between women and men with the same
levels of risk aversion (and the same socio-demographic characteristics). Second, we observe the
same drop in gender effects when comparing the third dot to the fourth. Belief-related factors greatly
contribute to improving the understanding of compliance. In particular, trust in the government and in
science and the intensive consumption of traditional media are strong positive determinants of
compliance. However, these characteristics are equally distributed between women and men, hence
they do not impact gender effects. The key channel which explains why gender effects are weaker in
the final regressions stems from the inclusion of the variable capturing individuals’ perception of
danger regarding COVID-19. The fear of COVID-19 is indeed much more prevalent among women,
and it provides a significant stimulus to comply with health measures. Hence, previous estimates,
which were not conditioning on beliefs, were overestimating gender differences, at least if we are
interested in comparing women and men who have the same beliefs.
Finally, comparing the first (no control variables) to the fourth dot (all control variables) shows us the
sensitivity of gender effects to important factors influencing compliance. Depending on the outcome,
we observe that unconditional gender effects are generally between 5 and 10 percentage points, and
the effects which account for all individual characteristics are about 5 percentage points lower. Hence,
for some outcomes, gender effects become small in magnitude and statistically insignificant. Let us
now identify nuances from these general observations by looking more specifically at the various
outcomes. Most of our focus here is on the full model represented by the fourth dots.
Whether one considers simple mean comparisons or uses the full model with all individual factors,
mask-wearing is a domain in which women are significantly more compliant. In the full model (fourth
dot), women remain 6 percentage points more likely than (comparable) men to consider masks as a
civic duty and to always wear them in public places.
Women are also more conscientious with respect to safety measures, they are very significantly more
likely to apply safety measures in a consistent way throughout the day than men, by about 7
percentage points. Although less significant, the proportion of women who are as careful with safety
measures as in the beginning is also larger than men’s (6 percentage points). Similarly, and again
weaker in significance, women are slightly more supportive than men towards the government’s
actions (4 percentage points difference).
Then, there are a number of outcomes for which conclusions on gender effects are strongly impacted
by the inclusion of variables such as education, risk aversion and fear of COVID-19 (among other).
First, participation in large scale and proactive testing are still slightly larger for women (2 percentage
points and 4 percentage points, respectively). However, these testing-related outcomes are no longer
statistically different between women and men who have the same characteristics. Furthermore,
average levels of avoidance of physical contacts and of public places are equal between women and
men once their education, risk aversion and fear of COVID-19 are taken into account (0% difference).
Last but not least, women’s intentions to get vaccinated become significantly lower than men’s. This
effect can be attributed, as discussed in the next Section to the fact that women are generally more

- 39 -
risk averse, and vaccination per se also entails a risk. Furthermore, the risks related to vaccination are
more prevalent among women. Indeed, adverse effects have been reported more frequently on
women than on men (Gee et al., 2021; Menni et al., 2021), in particular rare severe blood clots which
caused some vaccines to be suspended (LaPreze, 2021).
To conclude, women are more compliant with most health measures than men. About half of the
differences in compliance levels results from the fact that women are more risk averse and that they
are more likely to perceive COVID-19 as dangerous. While both risk perception and risk aversion
greatly contribute to women’s more careful attitudes, on the other hand women’s lower average
educational attainments negatively impacts compliance. When we compare the compliance levels of
women with identical socio-demographic characteristics, we find that compliance between women and
men still differs in several outcomes. In particular, women behave more responsibly and
conscientiously in terms of mask-wearing and safety measures than men with similar education, risk
aversion, perception of danger,... Gender differences are however smaller, for instance in terms of
testing participation. All characteristics equal, women and men have the same proactive risk
avoidance behavior. Finally, women are less willing to get vaccinated, which we investigate further in
the next section.
4. The impact of Vaxzevria suspensions and of EMA
communication on vaccination intentions
In this Section, we place a particular emphasis on vaccination intentions, which was the most timely
topic at the time in which the data was being collected, March 2021. Putting things in perspective, the
start of the vaccination campaign in early 2021 was being challenged by vaccine hesitancy and
antivax movements, despite months of lockdown and millions of deaths in 2020 (Williams, 2021). In
early March 2021, concerns related to vaccines were further fueled when European media reported
rare cases of blood clots among people who received Astrazeneca's Vaxzevria vaccine. This led
European governments to implement a large and uncoordinated wave of precautionary suspensions.
By March 15th, 18 countries had suspended Vaxzevria, pending an official statement by the European
Medicines Agency (EMA), the EU drug regulator. As a response, the EMA made gave a press
conference on March 18th stating that “the vaccine’s proven efficacy in preventing hospitalization and
death from COVID-19 outweighs the extremely small likelihood of developing” blood clots, and
recommended the vaccine's use. Within a few hours, 15 governments reintroduced Vaxzevria.
Since our data was collected in this particular period, we seized this opportunity to study how these
events affected vaccination intentions, and in particular whether the supranational health agency's
communication was able to mitigate the surge of vaccine hesitancy. In a recent research article
(Albanese, Fallucchi and Verheyden, 2021), we used a regression discontinuity design (RDD) which
exploits the information of the precise time at which individuals responded to the survey. This method
consists in comparing the intentions to be vaccinated between individuals who responded shortly
before and shortly after the EMA’s announcement, on March 18th at 17:00. Under the reasonable
assumption that characteristics of respondents did not vary around this particular moment (e.g. antivax
individuals did not choose to participate more or less in the survey after the EMA’s statement), this
method provides a causal estimation of the effect of the EMA’s statement. In our research, we found
an immediate and statistically significant effect of the EMA, which allowed to counterbalance the
severely declining trust in vaccines observed among respondents in the period preceding the EMA’s
statement.

- 40 -
In this Section, we extend the analysis of the impact of the EMA on vaccine hesitancy by providing a
specific focus on gender differences. Key results of this analysis are provided in Figure 2.5.A and
2.5.B, which describe the impact of the EMA statement on the willingness to bet vaccinated of men
(2.5.A) and women (2.5.B). The lines around the cutoff of March 18th represent the evolution of
vaccination intentions before and after the EMA’s statement.
First, the negative slopes of the lines before March 18th in both figures mean that the willingness to be
vaccinated of both women and men was severely declining in the days preceding the EMA’ statement.
However, the slope of men’s line is steeper, suggesting that men were more impacted (compared to
women) by the negative publicity induced by media reports and by the wave of suspensions across
Europe.
Secondly, men’s line after March 18th is both higher and flatter than the line before March 18th,
meaning that the EMA statement had an immediate and statistically significant positive impact on
men’s vaccination intentions. A qualitatively similar interpretation can be made for women, though the
post-EMA line has a downward slope, due to the stronger dispersion of the cloud of dots after March
18th. This suggests that the reassuring impact of the EMA was also stronger for men. While men
returned to stable vaccination intention levels comparable to the pre-crisis period, this cannot be said
for women. In the post-EMA period, the vaccination intentions of women are indeed significantly more
dispersed and on average lower than men’s.
Figure 2.5.A: the impact of the EMA statement on men’s willingness to bet vaccinated

- 41 -
Figure 2.5.B: the impact of the EMA statement on women’s willingness to bet vaccinated
These results suggest that while women were initially less impacted by the wave of suspensions, their
concerns might not have been totally alleviated by the EMA statement, as some women remained less
willing to get vaccinated. A plausible explanation for these findings is that most blood clot cases
reported in the media concern women, and women are more risk averse than men.
To conclude this Section, a cautionary remark is in order. Indeed, though these results are sensible,
they should be interpreted as descriptive rather than causal. The main reason for this is that the
sample size for each gender group is insufficient to draw robust conclusions with this methodology.
5. Conclusion
We review the differences between women and men in terms of compliance with COVID-19 measures
on a sample of almost 700 individuals from Luxembourg and its neighboring regions. We exploit data
from a survey administered online in March 2021 containing specific information on attitudes towards
physical distancing, mask-wearing, testing, vaccination, and support and consciousness towards
measures in general.
The majority of both women and men apply these measures. However, the proportion of women who
comply is generally larger than men’s, by 5 to 10 percentage points. This is particularly true for mask-
wearing, consistency in the application of safety measures, proactive testing and general support of
the government’s actions.
Out of this 5 to 10 percentage-point difference in compliance rates, about 5 percentage points can be
attributed to the fact that the average woman have different individual characteristics compared to the
average man. Women are indeed more risk averse and more likely to perceive COVID-19 as
dangerous, which captures a significant proportion of the gender difference. On the other hand,
women’s lower average educational attainments reduces the gender gap. Accounting for these factors
allows us to make “all else equal” gender comparisons, meaning comparisons between women and

- 42 -
men with the same individual characteristics. These all else equal comparisons show that gender
effects remain present in several domains, in particular conscientiousness in applying safety measures
and wearing masks, and to a lesser extent participation in testing. In contrast, proactive risk avoidance
through physical contacts and public places is identical between women and men with similar traits.
Finally, the only dimension in which women are less involved than men pertains to vaccination
intentions, which is 5 percentage points lower than men’s. This last conclusion is confirmed in our
analysis of the impacts of vaccine suspensions around Europe in early March 2021, and of the
ensuing reassurances about their safety brought by the European Medicines Agency (EMA). While
suspensions negatively impacted the intentions to get vaccinated of both women and men, the EMA
was less effective at restoring trust among women. This is sensible, considering that (i) women are
more risk averse than men in general and (ii) the ultrarare blood clots that caused the suspensions
were exclusively reported on women.
These results suggest that women and men are not sensitive to COVID-19 and health policies in the
same way, and therefore communication about the importance of these policies might gain to be
gender-specific. For instance, men should be specifically alerted about the risks of COVID-19 and
about the concrete benefits of policies, whereas women appear to be more naturally compliant with
safety measures. On the other hand, efforts to inform and quantify objectively the risks of vaccination
should be concentrated on women. Also, Chang et al. (2021) show that making such communication
gender-specific might also be effective, as it appears that communications made by women to the
attention of women stimulates vaccination intentions more than men-to-women communications.

- 43 -
References
Albanese, A., Fallucchi, F., and Verheyden, B. (2021) “Can a Supranational Medicines Agency
Restore Trust After Vaccine Suspensions? The Case of Vaxzevria”, IZA Discussion Paper, 14553.
Alsharawy, A., Spoon, R., Smith, A., and Ball, S. (2021) “Gender differences in fear and risk
perception during the COVID-19 pandemic.” Frontiers in psychology, 3104.
Algara, C., Fuller, S., Hare, C., and Kazemian, S. (2021). The interactive effects of scientific
knowledge and gender on COVID-19 social distancing compliance. Social Science Quarterly, 102(1),
7-16.
Alshammary, F., Siddiqui, A. A., Amin, J., Ilyas, M., Rathore, H. A., Hassan, I., ... and Kamal, M. A.
(2021) “Prevention knowledge and its practice towards COVID-19 among general population of Saudi
Arabia: a gender-based perspective.” Current pharmaceutical design, 27(13), 1642-1648.
Androsik, A. (2020) “Gendered understanding of ebola crisis in Sierra Leone. Lessons for COVID-19.”
Population and Economics, 4(2), 88–95. https://doi.org/10.3897/popecon. 4.e53301.
Baker, Alexander T., et al. (2021): "ChAdOx1 interacts with CAR and PF4 with implications for
thrombosis with thrombocytopenia syndrome." Science advances 7.49 (2021): eabl8213.
LaPreze, Jennifer, and CPP CDCES. "The Johnson and Johnson COVID-19 Vaccine, Women, and
VITT." Us Pharmacist HS2-HS6.
Barbieri, T., Basso, G. and Scicchitano, S. (2021) “Italian Workers at Risk During the COVID-19
Epidemic.” Italian Economic Journal.
Chang, J, M Shah, R Pramanik and S Shah (2021), “Financial Incentives and Other Nudges Do Not
Increase COVID-19 Vaccinations among the Vaccine Hesitant”, NBER Working Paper 29403.
Clark, C., Davila, A., Regis, M., and Kraus, S. (2020). Predictors of COVID-19 voluntary compliance
behaviors: An international investigation. Global transitions, 2, 76-82.
Galasso, V., V. Pons, P. Profeta, M. Becher, S. Brouard and M. Foucault (2020), “Gender differences
in COVID-19 related attitudes and behavior: Evidence from a panel survey in eight OECD countries”
Proceedings of the National Academy of Sciences of the United States of America 117(44): 27285–91.
Gee J., Marquez P., Su J., et al. (2021) “First Month of COVID-19 Vaccine Safety Monitoring — United
States, December 14, 2020–January 13, 2021.” MMWR Morb Mortal Wkly Rep 2021;70:283–288.
Green, M. S., Abdullah, R., Vered, S., and Nitzan, D. (2021) “A study of ethnic, gender and
educational differences in attitudes toward COVID-19 vaccines in Israel–implications for vaccination
implementation policies.” Israel Journal of Health Policy Research, 10(1), 1-12.
Kowalik, Z., and Lewandowski, P. (2021). The gender gap in aversion to COVID-19 exposure:
Evidence from professional tennis. PloS one, 16(3), e0249045.
Levkovich, I., and Shinan-Altman, S. (2021) “The impact of gender on emotional reactions, perceived
susceptibility and perceived knowledge about COVID-19 among the Israeli public.” International
Health.
Lin, T., Harris, E. A., Heemskerk, A., Van Bavel, J. J., and Ebner, N. C. (2021) “A multi-national test on
self-reported compliance with COVID-19 public health measures: The role of individual age and

- 44 -
gender demographics and countries’ developmental status.” Social Science and Medicine, 286,
114335.
Liu, N., Zhang, F., Wei, C., Jia, Y., Shang, Z., Sun, L., ... Liu, W. (2020) Prevalence and predictors of
PTSS during COVID-19 outbreak in China hardest-hit areas: Gender differences matter. Psychiatry
Research, 287, Article 112,921. https://doi.org/10. 1016/j.psychres.2020.112921.
McLaren, H. J., Wong, K. R., Nguyen, K. N., and Mahamadachchi, K. N. D. (2020) “Covid-19 and
women’s triple burden: Vignettes from Sri Lanka, Malaysia, Vietnam and Australia.” Social Sciences,
9(5), 1–11. https://doi.org/10.3390/socsci9050087.
Menni, Cristina, et al. (2021) "Vaccine side-effects and SARS-CoV-2 infection after vaccination in
users of the COVID Symptom Study app in the UK: a prospective observational study." The Lancet
Infectious Diseases.
Paramita, W., Rostiani, R., Winahjoe, S., Wibowo, A., Virgosita, R., and Audita, H. (2021) “Explaining
the voluntary compliance to COVID-19 measures: An extrapolation on the gender perspective.” Global
Journal of Flexible Systems Management, 22(1), 1-18.
Park, C. L., Russell, B. S., Fendrich, M., Finkelstein-Fox, L., Hutchison, M., and Becker, J. (2020).
“Americans’ COVID-19 stress, coping, and adherence to CDC guidelines.” Journal of General Internal
Medicine, 1–8. https://doi.org/10.1007/s11606-020- 05898-9.
Rodriguez-Besteiro, S., Tornero-Aguilera, J. F., Fernández-Lucas, J., and Clemente-Suárez, V. J.
(2021) “Gender differences in the covid-19 pandemic risk perception, psychology and behaviors of
spanish university students.” International Journal of Environmental Research and Public Health,
18(8), 3908.
Rosha, B. C., Suryaputri, I. Y., Irawan, I. R., Arfines, P. P., and Triwinarto, A. (2021) “Factors Affecting
Public Non-compliance With Large-scale Social Restrictions to Control COVID-19 Transmission in
Greater Jakarta, Indonesia.” Journal of Preventive Medicine and Public Health, 54(4), 221.
Wang, C. J., Ng, C. Y., and Brook, R. H. (2020) “Response to COVID-19 in Taiwan: big data analytics,
new technology, and proactive testing.” JAMA, 323(14), 1341–1342.
https://doi.org/10.1001/jama.2020.3151.
Williams, L., Flowers, P., McLeod, J., Young, D., and Rollins, L. (2021) “Social patterning and stability
of intention to accept a COVID-19 vaccine in Scotland: Will those most at risk accept a vaccine?.”
Vaccines, 9(1), 17.
Xiao, X., Xiao, J., Yao, J., Chen, Y., Saligan, L., Reynolds, N. R., and Wang, H. (2020) “The Role of
Resilience and Gender in Relation to Infectious-Disease-Specific Health Literacy and Anxiety During
the COVID-19 Pandemic.” Neuropsychiatric Disease and Treatment, 16, 3011.

- 45 -
Appendix
Table A2.1: Descriptive statistics of individual characteristics, by gender
(1)
(2)
(3)
(2)-(3)
All
Woman
Man
t-test
Mean
N
Mean
N
Mean
N
Difference
[SE]
[SE]
[SE]
Sociodemographic characteristics
Woman
0,67
689
1,00
461
0,00
228
NA
[0,018]
[0,000]
[0,000]
Not Luxembourgish
0,41
689
0,36
461
0,49
228
-0,122***
[0,019]
[0,022]
[0,033]
Single
0,30
689
0,29
461
0,33
228
-0,04
[0,018]
[0,021]
[0,031]
Has no higher education
0,45
689
0,48
461
0,39
228
0,091**
[0,019]
[0,023]
[0,032]
Not working
0,20
689
0,22
461
0,15
228
0,066**
[0,015]
[0,019]
[0,024]
Age under 35
0,13
689
0,13
461
0,12
228
0,01
[0,013]
[0,016]
[0,021]
Age above 50
0,44
689
0,41
461
0,48
228
-0,06
[0,019]
[0,023]
[0,033]
Behavioral traits
Willingness to take risks
0,53
653
0,50
430
0,58
223
-0,078***
[0,009]
[0,012]
[0,016]
Self-determination
0,42
642
0,43
426
0,39
216
0,033*
[0,009]
[0,011]
[0,016]
Patience
0,68
643
0,68
429
0,67
214
0,02
[0,009]
[0,010]
[0,016]
Beliefs and belief formation
Considers COVID-19 as dangerous
0,72
683
0,75
459
0,66
224
0,090***
[0,012]
[0,014]
[0,023]
TV / radio / newspapers
0,84
689
0,84
461
0,83
228
0,01
[0,014]
[0,017]
[0,025]
Online journals / social media
0,92
689
0,92
461
0,92
228
0,01
[0,010]
[0,013]
[0,018]
Does not fully trust the governm’t
0,36
689
0,36
461
0,38
228
-0,02
[0,018]
[0,022]
[0,032]
Does not fully trust scientists
0,34
689
0,35
461
0,33
228
0,03
[0,015]
[0,019]
[0,026]
The value displayed for t-tests are the differences in the means across the
groups.
***, **, and * indicate significance at the 1, 5, and 10 percent critical level.

- 46 -
Table A2.2: Summary statistics of attitudes towards COVID-19 measures, by gender
(1)
(2)
(3)
(2)-(3)
All
Woman
Man
t-test
Mean
N
Mean
N
Mean
N
Difference
[SE]
[SE]
[SE]
Considers wearing masks as a civic
duty
77,86
682
81,21
456
71,09
226
10,12***
[1,193]
[1,337]
[2,326]
Always wears masks in public places
62,25
687
65,91
460
54,85
227
11,06***
[1,416]
[1,650]
[2,619]
Never forgets safety measures
throughout the day
51,90
685
54,18
459
47,27
226
6,90**
[1,405]
[1,705]
[2,455]
Is as careful as in the beginning of the
pandemic
65,94
684
68,70
459
60,30
225
8,40***
[1,373]
[1,637]
[2,463]
Supports the government's actions
against the pandemic
69,64
685
72,80
459
63,20
226
9,60***
[1,262]
[1,434]
[2,428]
Participates in large scale testing
campaign
85,78
689
87,64
461
82,02
228
5,62**
[1,332]
[1,535]
[2,549]
Gets tested in case of contact or
feeling sick
64,08
689
66,81
461
58,55
228
8,26***
[1,437]
[1,720]
[2,566]
Avoids physical contacts
87,29
686
88,71
459
84,44
227
4,27**
[1,007]
[1,138]
[1,980]
Avoids public places
67,61
686
69,57
459
63,66
227
5,92**
[1,381]
[1,613]
[2,587]
Intends to get vaccinated
76,63
689
76,36
461
77,18
228
-0,82
[1,225]
[1,471]
[2,207]
The value displayed for t-tests are the differences in the means across
the groups.
***, **, and * indicate significance at the 1, 5, and 10 percent critical level.

- 47 -
Table A2.3: Regression results (1): Avoiding physical contacts and public places
Avoids physical contacts
Avoids public places
(1)
(2)
(3)
(4)
(5)
(6)
Woman
5,126**
3,111
0,088
7,453**
3,787
-0,508
(2,38)
(1,33)
(0,04)
(2,58)
(1,22)
(-0,18)
Not Luxembourgish
0,961
0,213
1,206
5,314*
3,305
3,466
(0,46)
(0,09)
(0,56)
(1,88)
(1,08)
(1,23)
Single
-2,845
-1,649
-2,220
-2,816
-3,863
-4,769
(-1,27)
(-0,67)
(-0,98)
(-0,94)
(-1,19)
(-1,62)
Has no higher education
-4,87**
-6,71***
-4,43**
-7,76***
-10,1***
-7,59***
(-2,35)
(-2,93)
(-2,06)
(-2,79)
(-3,33)
(-2,70)
Not working
0,0041
-0,202
-0,481
7,044*
7,136*
6,241*
(0,00)
(-0,07)
(-0,17)
(1,93)
(1,79)
(1,73)
Age under 35
-6,01*
-5,59
-3,70
-13,8***
-13,1***
-11,2***
(-1,87)
(-1,64)
(-1,17)
(-3,21)
(-2,90)
(-2,72)
Age above 50
5,227**
5,133**
0,932
9,75***
12,88***
7,167**
(2,28)
(2,05)
(0,39)
(3,17)
(3,88)
(2,31)
Willingness to take risks
-21,00***
-16,70***
-29,5***
-22,7***
(-4,50)
(-3,86)
(-4,76)
(-4,02)
Self-determination
-5,322
-3,367
5,773
6,899
(-1,10)
(-0,75)
(0,90)
(1,17)
Patience
3,284
-0,120
11,51*
7,320
(0,66)
(-0,03)
(1,74)
(1,20)
Considers COVID as
dangerous
30,84***
42,85***
(9,28)
(9,88)
TV / radio / newspapers
1,277
3,563
(0,46)
(0,99)
Online journals / social media
8,262**
0,750
(2,22)
(0,15)
Does not fully trust governm’t
-3,459
-7,89***
(-1,51)
(-2,64)
Does not fully trust scientists
-3,406
-2,610
(-1,21)
(-0,71)
Constant
85,02***
97,86***
71,14***
60,94***
68,61***
41,39***
(32,72)
(18,88)
(10,82)
(17,46)
(9,97)
(4,82)
R²
0,0335
0,0669
0,224
0,0746
0,122
0,286

- 48 -
Table A2.4: Regression results (2): Masks
Considers wearing masks as a
civic duty
Always wears masks in public
places
(1)
(2)
(3)
(4)
(5)
(6)
Woman
12,1***
9,77***
6,239***
12,71***
9,476***
5,835*
(4,84)
(3,65)
(2,78)
(4,24)
(2,91)
(1,90)
Not Luxembourgish
6,88***
6,483**
8,077***
9,723***
8,208**
9,616***
(2,80)
(2,45)
(3,65)
(3,30)
(2,56)
(3,17)
Single
0,315
0,909
1,309
-0,331
-2,481
-2,454
(0,12)
(0,32)
(0,56)
(-0,11)
(-0,73)
(-0,77)
Has no higher education
-
8,00***
-7,27***
-2,58
-5,89**
-6,31**
-2,59
(-3,32)
(-2,77)
(-1,17)
(-2,04)
(-1,98)
(-0,86)
Not working
1,015
1,663
0,446
6,695*
6,523
6,118
(0,32)
(0,48)
(0,16)
(1,77)
(1,57)
(1,58)
Age under 35
-1,06
-3,42
-1,37
-1,77
-1,79
0,71
(-0,28)
(-0,88)
(-0,42)
(-0,40)
(-0,38)
(0,16)
Age above 50
7,57***
6,670**
0,782
5,537*
6,516*
1,422
(2,83)
(2,32)
(0,32)
(1,74)
(1,88)
(0,43)
Willingness to take risks
-10,81**
-4,840
-13,98**
-8,847
(-2,02)
(-1,09)
(-2,16)
(-1,46)
Self-determination
1,924
3,623
5,194
6,424
(0,35)
(0,78)
(0,78)
(1,02)
Patience
17,53***
6,925
15,81**
8,164
(3,06)
(1,43)
(2,27)
(1,24)
Considers COVID as
dangerous
36,57***
32,34***
(10,69)
(6,93)
TV / radio / newspapers
8,183***
7,204*
(2,90)
(1,86)
Online journals / social media
4,103
9,118*
(1,07)
(1,72)
Does not fully trust governm’t
-14,24***
-10,75***
(-6,06)
(-3,34)
Does not fully trust scientists
-13,32***
-6,884*
(-4,60)
(-1,74)
Constant
67,1***
62,08***
40,23***
49,06***
46,01***
18,71**
(22,14)
(10,45)
(5,96)
(13,55)
(6,40)
(2,02)
R²
0,0655
0,0825
0,378
0,0531
0,0628
0,198