Occupational Prestige and the Gender Wage Gap

Occupational Prestige and the Gender Wage Gap
KRISTIN J. KLEINJANS, California State University, Fullerton
KARL FRITJOF KRASSEL,
Danish Institute for Local and Regional Government Research (KORA)
ANTHONY DUKES, University of Southern California – Marshall School of Business
ABSTRACT
Occupational segregation by gender remains widespread and explains a significant part of the
gender wage gap. We examine the explanation that heterogeneity in preferences for wages
and occupational prestige leads to gender differences in occupational choices, creating a
gender wage gap. In self-reports, women express a stronger preference than men for
occupations that are more valuable to society, which we hypothesize leads women to place
more importance than men on the occupational prestige of their occupation, resulting in lower
wages for occupations with higher prestige. Using Danish data, we find support for this
hypothesis, especially for individuals with lower ability.
Keywords: occupational choice, occupational prestige, social prestige, gender wage gap,
gender roles
JEL Codes: D13, J16, J24
1
INTRODUCTION
Women’s educational attainment has increased dramatically over the last decades, gender
roles have changed and women’s labor force attachment has increased in most industrialized
countries (Blau, 2012; Blau and Kahn, 2000; Goldin et al., 2006). Nevertheless, women still
earn less than men in most if not all countries (Anker, 1997; Blau, 2012). Up to one half of
this gender pay gap can be explained by gender differences in occupational choice,
commonly referred to as occupational segregation (Blau and Kahn, 2007; see also Hellerstein
et al., 2008; and Bayard et al., 2003). Women and men tend to choose different occupations
even with the same level and type of education (Shauman, 2006), even though young women
today expect to be working throughout their lives, albeit with intermittent absences for child
bearing and rearing (Goldin, 2006). These changes in educational attainment and patterns of
labor force participation make the traditional notions of differences in human capital and
expected labor force attachment insufficient to explain the gender wage gap, especially for
young people. 1
Our focus in this paper is on exploring whether young women and men choose
different occupations because of heterogeneity in preferences over attributes of occupations.
Up to now, research has mostly considered women’s traditionally stronger preference for
occupational attributes that make work more compatible with child rearing, such as shorter or
more flexible work hours. We consider a different type of attribute - occupational prestige and argue that part of the gender wage gap results from gender differences in preferences for
occupational prestige. Women express a stronger preference than men for occupations that
1
In addition to the explanations of differences in human capital and expected labor force attachment
(Polachek, 1981; see England, 1982 for an opposing argument) and social roles (Eccles, 1994), recent studies
have aimed at explaining occupational segregation with differences in non-cognitive skills (Antecol and CobbClark, 2013; Cobb-Clark and Tan, 2011; Grove, Hussey, and Jetter, 2011) and preferences for competition
(Kleinjans, 2009), but found generally only small statistically significant effects.
2
are deemed useful to society (see, for example, Fortin 2008; Grove, Hussey, and Jetter, 2011;
Marini et al., 1996). 2 Hence, women likely place more importance than men on the social
value of their occupation. 3 We argue that occupational prestige reflects the social value
bestowed on an occupation, which implies that women should be more likely to choose
occupations with higher prestige than men. Compensating variation leads to lower wages in
occupations with higher prestige, resulting in a gender wage gap. To investigate this
hypothesis, we analyze whether there are gender differences in the relative importance of
occupational prestige for occupational choice, and whether this difference can explain part of
the observed gender wage gap resulting from occupational segregation. 4
Our results improve our understanding of reasons for occupational segregation.
Furthermore, they shed light on the transmission mechanisms by which gender differences in
preferences lead to differences in economic outcomes. In particular, if gender differences in
preferences for occupational prestige are the result of gender roles (and we find some support
for this interpretation), our findings can explain the mechanism by which gender roles lead to
differences in occupational choices and, as a result, differences in wages.
In traditional economic models, occupational choice depends on expected wages and
the cost of attaining an occupation. Sociologists - and, more recently, economists - have
2
For example, Marini et al. (1996) report that in a survey of high school seniors on the importance of job
attributes women were 66% and 44% more likely than men to indicate as very important that a job is “helpful to
others” and “worthwhile to society”. Fortin (2008) finds that women’s greater expressed importance of
“working with people” and “usefulness of a job to others and society” contributes slightly if at all to the gender
wage gap. Because the analysis combines these two measures, her results are not comparable to ours.
3
Grove, Hussey, and Jetter (2011) find this for a national sample of MBAs in the US, and that it results in a
wage penalty for women.
4
This is also in line with the finding by Andreoni and Vesterlund (2001) that women are more altruistic than
men when altruism is expensive.
3
stressed the importance of other factors for occupational choice (Fershtman and Weiss, 1993
and 1998; Jacobs et al., 2006; Rothstein and Rouse, 2007). These include parental
expectations, social norms and non-pecuniary benefits, such as the occupational prestige of
an occupation.
Occupational prestige (sometimes also referred to as social prestige) is defined as the
social standing given to those holding a specific occupation (Hauser and Warren, 1997). The
prestige assigned to an occupation is stable over time and similar across countries and
population subgroups, including gender (Anker, 1982; Treiman, 1977; Warren, Sheridan and
Hauser, 1997). Although occupational prestige is highly correlated with wages, ability and
educational requirements (Chartrand et al., 1987), occupational prestige measures cannot be
explained solely by those variables. In some ways related to the notion of social status used
by economists to proxy relative social standing (see, e.g., Dolton, Makepeace, and van der
Klaauw, 1989; Fershtman and Weiss, 1998), 5 we interpret occupational prestige (net of the
effect of wage) as reflecting the occupation’s perceived contribution to society (see also
Anker, 1982). However, unlike social status, occupational prestige is non-rivalrous.
Specifically, the contribution to society results from positive externalities of occupations or
their contributions to public goods (e.g., teachers and artists) and not from the number or type
of workers in those occupations. Individuals benefit because contributing to society provides
altruistic rewards (Fortin, 2008), that is, direct utility. Consequently, since women express
stronger preferences for occupations that are deemed valuable to society we hypothesize that
women care more about occupational prestige than men.
5
Social status is generally a composite measure derived from occupational prestige, salary, and sometimes the
educational level of those holding the occupation (see Warren, Sheridan, and Hauser, 1998; Hauser and Warren,
1997). Few economists consider occupational prestige – a notable exception is Zhang (2012) who uses prestige
as a proxy for respect to examine the effect of cultural attitudes on occupational choice.
4
Our measure of interest in our empirical analysis (the social value of an occupation) is
thus occupational prestige net of the effects of wage and other confounding factors. We,
therefore, control for wage and ability, and conduct extensive robustness checks to assure that
our results regarding occupational prestige are not driven by confounding factors (such as job
characteristics).
In this paper, we are only able to speculate on the underlying reasons for the gender
difference in the importance of occupational prestige. Our findings are consistent with the
explanation that they result directly or indirectly from gender role socialization. This could be
through its effect on preferences (Eccles, 1994), discrimination by teachers or employers of
non-traditional choices, 6 or the ensuing lack of role models in non-traditional occupations
(Blau, Ferber, and Winkler, 2010). It is also possible, however, that the driver is the gender
difference in evolutionary advantages from altruistic behavior (Campbell, 2003). While
understanding the origins of the differences in preferences for occupational prestige is beyond
the scope of this paper, we discuss the results of some explorative analysis suggesting the
importance of expected discrimination and gender roles.
In what follows, we present a simple equilibrium model of equalizing differences
where occupational prestige is interpreted as an amenity. The model predicts, in particular,
lower wages in occupations with higher occupational prestige for a given skill level if some
individuals care about occupational prestige. Hence, if women derive higher utility from
occupational prestige, women will sort into lower paying but more prestigious occupations,
resulting in a gender wage gap. 7
6
We then use Danish data to estimate a model of
Compare, for example, the public perception of female police officers and male receptionists or parents’ and
especially fathers’ negative reaction to boys playing with dolls (Eliot, 2009; Fine, 2010).
7
Such sorting could also explain why daughters have a lower intergenerational correlation of income than
sons (Bowles and Gintis, 2002), and why parental income affects men’s but not women’s expectations of
educational achievement when parental education is controlled for (Kleinjans, 2010). If women opt for higher
5
occupational choice in which the probability of expecting to work in an occupation depends
on occupational and individual characteristics. 8 Our estimates indicate, indeed, that women
expect to work in occupations with higher occupational prestige and lower wages than men.
To examine the implications of these findings for the predicted wage gap, we study the
counterfactual question of how much the gender wage gap would change if women had
men’s coefficients of occupational prestige and wages. We find that this can explain about
half of the gender gap resulting from occupational segregation. Furthermore, the gender
differences are greater for individuals with lower ability, in line with an interpretation that the
gender differences are related to gender roles, which tend to be more traditional for
individuals with lower ability.
I. AN EQUILIBRIUM MODEL OF WAGES AND OCCUPATIONAL PRESTIGE
In this section, we describe a simple equilibrium model of a labor market with occupational
prestige, and show why gender differences in preferences for occupational prestige may lead
to a gender wage gap. The objective of the model is to provide the formalization of the theory
of equalizing differences (Rosen, 1986) as it regards the wage gap resulting from gender
differences in preferences for occupational prestige (see also Hamermesh, 1999). These
gender differences lead to gender segregation in occupations, which are accompanied by
prestige and wage differences. We then show how this difference can be moderated by
external factors, such as a worker’s parental background, affecting occupational preferences.
There are two types of jobs, 𝑗 = 0,1, which correspond to an occupation with a low (0)
and a high (1) exogenous level of occupational prestige. Workers’ preferences are
prestige but lower paying occupations than men, their wages have a lower correlation with parental income than
men’s.
8
In this paper, we cannot address gender wage differences resulting from differential sorting into firms within
occupations (Blau, 2012).
6
represented by the utility function 𝑢(𝐶, 𝑗), which is quasi-concave and increasing in 𝐶, the
level of consumption purchased by wages. Occupational prestige can be considered an
amenity, which provides additional utility for a given level of consumption for job 𝑗 = 1:
𝑢(𝐶, 1) > 𝑢(𝐶, 0) for all 𝐶. For each worker, there exists a unique 𝑧 ≥ 0 such 𝑢(𝐶 + 𝑧, 0) =
𝑢(𝐶, 1), which represents the compensating variation for job 𝑗 = 0 compared with 𝑗 = 1. As
discussed in the introduction, women may have a stronger relative preference for occupations
with higher occupational prestige. We interpret this gender difference as follows. Let
𝑖 = 𝐹, 𝑀 denote gender and suppose 𝑧𝑖 uniquely solves 𝑢𝑖 (𝐶 + 𝑧𝑖 , 0) = 𝑢𝑖 (𝐶, 1). Then
𝑧𝐹 > 𝑧𝑀 .
Using the above utility formulations, we can now derive the supply of workers.
Suppose there is a set of workers with mass of one. Let 𝑔𝑖 (𝑧) and 𝐺𝑖 (𝑧) be the p.d.f. and
c.d.f., respectively, corresponding to the distribution of workers’ compensating variations, 𝑧,
1
for each gender 𝑖 = 𝐹, 𝑀. Assume genders have equal mass so that 𝐺𝑀 (∞) = 𝐺𝐹 (∞) = 2.
The gender differences in relative preferences over occupational prestige assumed above
implies that:
𝐺𝐹 (𝑧) < 𝐺𝑀 (𝑧) for all 𝑧 ∈ (0, ∞).
(1)
Let Δ𝑤 = 𝑤0 − 𝑤1 be the difference in wages across the two professions. Then worker
𝑖’s occupational choice is 𝑗 = 0 if Δ𝑤 > 𝑧 and 𝑗 = 1 otherwise. Note that any equilibrium
with occupational segregation must have Δ𝑤 > 0 since otherwise all workers would choose
occupation 𝑗 = 1. Furthermore, we shall also assume that 𝑔𝑖 (𝑧) > 0 for all 𝑧 > 0 and some 𝑖
so that some workers (even with very small 𝑧) have sufficiently little preference for
occupational prestige and that an arbitrarily small wage differential makes occupation 𝑗 = 0
more attractive. The supply of workers 𝑁𝑗𝑆 for each occupation 𝑗 = 0,1 is then expressed as
Δ𝑤
𝑁0𝑆 = 1 − 𝑁1𝑆 = ∫0
[𝑔𝐹 (𝑧) + 𝑔𝑀 (𝑧)]𝑑𝑑 = 𝐺𝐹 (Δ𝑤) + 𝐺𝑀 (Δ𝑤).
(2)
7
1
The number of women and men in the market for occupation 𝑗 is 𝑁0𝑊 = 2 − 𝑁1𝑊 = 𝐺𝐹 (Δ𝑤)
1
and 𝑁0𝑀 = 2 − 𝑁1𝑀 = 𝐺𝑀 (Δ𝑤), respectively.
To fully characterize the equilibrium we specify the demand side of the market. We
assume that there are a set of employers that potentially hire in both occupations, 𝑗 = 0 or 1.
They differ, however, in the marginal products of each type of occupation. Formally, suppose
each employer’s output technology is represented by linear functions: 𝑥 = 𝑎𝑗 𝐿𝑗 , where 𝐿𝑗 ≥ 0
is the number of workers hired in occupation 𝑗 and 𝑎𝑗 is the marginal product of labor for
each occupation at a given employer. For instance, a hospital hires 𝐿1 doctors (high
occupational prestige) and 𝐿0 lab technicians. Define 𝑏 ≡ 𝑎0 − 𝑎1, which represents the
relative marginal benefit of hiring in a low occupational prestige profession. We assume that
each employer is characterized by its 𝑏 ∈ (−∞, +∞), whose distribution across employers is
represented by the p.d.f. and c.d.f. 𝑓(𝑏) and 𝐹(𝑏). Employers hire in occupation 𝑗 = 0 if
𝑏 > Δ𝑤 and 𝑗 = 1 otherwise. A necessary condition for some employers hiring in both
occupations is that 𝐹(0) < 1. Then, the demand for workers in each profession is expressed
as
Δ𝑤
𝑁1𝐷 = 1 − 𝑁0𝐷 = ∫−∞ 𝑓(𝑏)𝑑𝑑 = 𝐹(Δ𝑤).
(3)
The equilibrium condition is that the market clears in each occupation so that 𝑁𝑗𝑆 = 𝑁𝑗𝐷 for
𝑗 = 0,1. Using (2) and (3), it is directly shown that any equilibrium wage differential Δ𝑤 ∗
must satisfy 𝐺𝐹 (Δ𝑤 ∗ ) + 𝐺𝑀 (Δ𝑤 ∗ ) = 1 − 𝐹(Δ𝑤 ∗ ). Workers are employed in both
occupations as long as 𝐺𝑖 (Δ𝑤 ∗ ) ∈ (0,1/2) for some 𝑖. By the condition on 𝑔𝑖 mentioned
above, this is the case as long as Δ𝑤 ∗ > 0, which is implied by our assumption that 𝐹(0) <
1.
We now establish that in equilibrium, there is gender segregation with a corresponding
wage gap. Denote by 𝑁𝑗𝑖 the equilibrium level of employment for gender 𝑖 in occupation 𝑗.
8
The assumption on gender preference differences in (1) implies 𝑁0𝑊 < 𝑁0𝑀 and 𝑁1𝑊 > 𝑁1𝑀 , or
that a greater portion of women (men) than of men (women) is in the high (low) occupational
prestige profession, 𝑗 = 1. The weighted wage differential across the population is
∗
𝑤
� 𝐹∗ − 𝑤
�𝑀
=
𝑤0∗ 𝑁0𝑊 +𝑤1∗ 𝑁1𝑊
𝑁0𝑊 +𝑁1𝑊
−
𝑤0∗ 𝑁0𝑀 +𝑤1∗ 𝑁1𝑀
𝑁0𝑀 +𝑁1𝑀
< 2𝑤1∗ [(𝑁0𝑊 + 𝑁1𝑊 ) − (𝑁0𝑀 + 𝑁1𝑀 )],
where the inequality holds because Δ𝑤 ∗ = 𝑤0∗ − 𝑤1∗ > 0 > 𝑁0𝑊 − 𝑁0𝑀 . Since the RHS of the
∗
above inequality is zero (𝑁0𝑖 + 𝑁1𝑖 = 1/2, both 𝑖), we have 𝑤
� 𝐹∗ − 𝑤
�𝑀
< 0. Women earn less,
on average, than men. This also implies that occupations with greater female shares have
lower wages than those with greater male shares, in line with observed occupational pay
differences (Blau, 2012).
Finally, we argue how ability (or some other external factor) can affect worker
preferences for occupations and ultimately the degree of gender segregation and wage
differentials. Workers’ preferences for the degree of occupational prestige may be affected by
ability if, for example, gender roles are less traditional for individuals with higher ability. To
implement this notion formally, rewrite the utility of a worker of type 𝑖 = 𝐹, 𝑀 as 𝑢𝑖 (𝐶, 𝑗; 𝑃),
where 𝑃 represents ability and let 𝑧𝑖 (𝑃) uniquely solve 𝑢𝑖 (𝐶 + 𝑧𝑖 (𝑃), 0, 𝑃) = 𝑢𝑖 (𝐶, 1, 𝑃).
Since higher ability may lead to less traditional gender roles, we interpret this by saying that
𝜅(𝑃) = 𝑧𝐹 (𝑃) − 𝑧𝑀 (𝑃) > 0 is a decreasing function of 𝑃 for all 𝐶. It follows directly that
gender segregation in occupations and the wage gap Δ𝑤 decreases in 𝑃. If 𝜅(𝑃�) = 0 for
sufficiently large 𝑃�, then gender segregation and the wage gap disappear.
II. THE GENDER WAGE GAP IN DENMARK
In the empirical analysis, we use data from Denmark. Denmark is well-suited for our study
because its gender wage gap and level of occupational segregation is similar to other
countries. At the same time, gender differences in labor force participation and part-time
9
employment are relatively small, decreasing the importance of gender differences in expected
labor force attachment for occupational choice. In what follows, we briefly describe these
features.
The raw gender wage gap in Denmark, similar to many countries, is 16.3%. 9 Despite
women’s increase in educational achievement over the past decades (now women’s average
education is almost one year higher than men’s), 10 occupational segregation remains
widespread. Figure 1 shows segregation by gender and industry: four of the ten industries
have a 70% or higher gender concentration. Compared to other European countries, Denmark
is in the middle group with respect to occupational segregation (Bettio and Verashchagina,
2009). Denmark differs, however, from other industrialized (non-Scandinavian) countries in
its relatively high female labor force participation and lower prevalence of female part-time
employment. Denmark has strong dual-earner family policies with universal child care and
paid parental leave (Lambert, 2008), so, not surprisingly, the Danish gender gap in labor
force participation of 7.0 percentage points is smaller than in other countries, including
Germany (11.8 percentage points) and
the U.S. (13.5 percentage points). 11 Hence,
differences in labor force attachment are smaller than in many other countries.
9
Based
on
hourly
wage
in
2011
(Eurostat,
Gender
pay
gap
statistics,
http://ec.europa.eu/eurostat/tgm/table.do?tab=table&init=1&language=en&pcode=tsdsc340&plugin=1, accessed
1/17/2015. The gap is about 4%-points smaller for performed work hours, and fluctuates with the business cycle
(Larsen and Houlberg, 2013).
10
Own calculations based on Statistics Denmark and ISCED97 for 2011 (www.statistikbanken.dk, KRHFU1,
http://eng.uvm.dk/Education/Overview-of-the-Danish-Education-System, accessed 1/17/15).
11
Own
calculations
based
on
OECD
Labour
Force
Statistics
for
2010
(http://stats.oecd.org/Index.aspx?DataSetCode=LFS_D, accessed 1/17/15). Individuals in education are counted
as being in the labor force.
10
The gender gap in part-time work is also smaller in Denmark. In 2010, 24.1% of
women worked part-time versus 12.0% of men - in contrast, for example, to Germany, where
the shares were 30.8% and 8.4%. 12 Since part time workers tend to receive lower (hourly)
wages, the tradeoff between wages and occupational prestige is different for people expecting
to work part time from those who do not. This makes Danish data useful for the study of
occupational choice because of the smaller potential impact of lower expected work hours.
Construction
Agriculture, forestry and fishing
Agriculture, forestry and
fishing
Manufacturing, mining and
quarrying, and utility
Real estate
Trade and transport etc.
Other business services
Financial and insurance
Arts, entertainment and other services
Men
10
0%
90
%
80
%
70
%
60
%
50
%
40
%
30
%
20
%
10
%
0%
Public administration, education and health
Women
Source: Statistics Denmark.
FIGURE 1. Occupational Segregation by Sector in Denmark (full time employed), 2008
12
Own calculations based on OECD Labour Force Statistics
(http://stats.oecd.org/Index.aspx?DataSetCode=FTPTC_D, accessed 1/17/15). Part time is defined as working
fewer than 30 hours in the main job.
11
III. DATA
We use a unique data set from Denmark that combines individual characteristics from survey
and assessment data, occupational information drawn from population registries and the EU
Labour Force Survey, and a survey measure of occupational prestige. The data is combined
by linking parental information to youth, and by matching occupation-specific characteristics
to the expected occupations of these youth. A key advantage of this data set is that it allows
us to distinguish the importance of occupational prestige from other correlated occupational
and individual characteristics, such as wages and ability. Specifically, by using an exogenous
source of occupational prestige we avoid the potential pitfall of endogeneity of expectations
and occupational prestige.
The Danish PISA-Longitudinal Database is our source of individual-level data. It
combines information from the 2000 OECD Programme for International Students
Assessment (PISA) survey of nationally representative ninth graders and a follow-up survey
entitled Young people in job or education – values, choices and dreams for the future, which
re-interviewed the by then 19-years old PISA respondents in 2004. 13 The ability measure the PISA reading test score - comes from the 2000 PISA survey 14 and all other individual
variables from the follow-up survey. The exceptions are gender, parental income and
education, which are drawn from matched Statistics Denmark registers for the year 2003.
Occupational choice is derived from answers to a question about the expected
occupation at age 30. Teenagers’ expectations have been found to be predictive of outcomes
(Fischhoff et al., 2000) and occupational expectations to be predictive for professionals
(Schoon, 2001). Moreover, in Denmark the choice of occupation is closely related to
educational choices, such as college major or type of educational training, mitigating the
13
See Jensen and Andersen (2006) for more information on this data set.
14
There is no other ability measure available for our sample.
12
effect of the time difference between the age at the time of the survey (19 years) and expected
occupation. Since occupational expectations reflect plans and intentions they are a good
measure of the effect of occupational prestige and wages on occupational choice. This is
distinct from occupational attainment, which also reflects hiring decisions by firms (Antecol
and Cobb-Clark, 2013). Our analysis assumes that individuals know median wages or at least
the wage differences across occupations (Betts, 1996).
The source of the occupational prestige variable is a survey from 2006 conducted by
Ugebrevet A4, a Danish news media owned by LO, The Danish Confederation of Trade
Unions, in which a representative sample of 2,155 Danes was asked to score 99 occupations
according to their occupational prestige by assigning a number from 0 (lowest) to 10
(highest). 15 The survey was conducted in collaboration with Analyse Danmark, a Danish
market research institute, from whose multi-purpose web panel the respondents were drawn.
The occupational prestige of an occupation is measured as the mean score given by the
respondents of the survey. Because there are some (albeit small) differences in the scores by
age of the respondents, we use the scores of the youngest respondent category (ages 18-29).
The level of detail of this survey matches the type of occupations provided by the youth
well. 16 We derive the average number of actually performed work hours from the EU Labour
Force Survey from 2003 and use registry data from the entire Danish population in 2003 to
derive the other occupation-specific measures.
15
The question was: ”How would you assess the prestige of the following occupations in Denmark? You can
answer from 0 (no prestige at all) to 10 (very high prestige)”.
16
Because of this we prefer this survey measure to the Ganzeboom and Treiman’s scale. See Appendix C for
the results of robustness checks using this scale. There are also some differences in the prestige awarded to
certain types of occupations, most importantly lawyers and skilled trade occupations, which are much higher
ranked in Denmark.
13
Our working sample consists of 1,796 individuals matched to 74 occupations. Women
have 5.9% lower hourly wages in their expected occupation
17
and 4.6% lower occupational
prestige scores than men. As expected, wages and occupational prestige have a high
correlation with 0.73. Figure 1 in Appendix A displays the distribution of wages sorted by
occupational prestige. Appendix A also provides more detailed information about the data:
Table A1 shows the descriptive statistics by occupation, Table A2 the variables used in our
analysis, data source, definitions, and means and standard deviations of our working sample
by gender. Appendix B gives more details about the sample selection and data set
construction. Appendix C shows robustness checks for our results using a different measure
of occupational prestige as well as sensitivity analysis regarding our sample selection.
IV. ESTIMATION STRATEGY
Our objective is to assess whether occupational prestige differently affects women’s and
men’s choice of occupation, controlling for other occupation-specific and individual-specific
characteristics. Therefore, we estimate a conditional logit model (also called a fixed-effects
logit model) where the dependent variable equals one if an occupation is chosen and zero if
not. Note that to estimate this model, the data is arranged such that the number of
observations per individual is equal to the number of possible occupations. We maximize the
conditional likelihood with the following conditional probability (Boskin, 1974):
𝑃𝑃𝑃𝑃�𝑌𝑖 = 𝑗�𝑧𝑖1, 𝑧𝑖2 , … , 𝑧𝑖𝑖 � =
𝑒
𝛽′𝑧𝑖𝑖
𝛽′𝑧
∑𝐽𝑗=1 𝑒 𝑖𝑖
,
which is the probability that individual i’s choice, 𝑌𝑖 , is occupation j, and 𝑧𝑖𝑖 are the
occupation-specific characteristics as well as interaction terms of occupation- and individual17
Note that this gap results purely from occupational segregation. It implies that about half of the gender wage
gap per performed work hour can be explained by occupational segregation, similar in magnitude to previous
findings.
14
specific characteristics, such as ability. 18 All models are estimated using robust standard
errors. In order to allow for all possible interactions between gender and other effects and to
ease the interpretation of the results, we conduct separate regressions by gender and present
odds ratios.
In this paper, we are interested in the gender differences in the effect of occupational
prestige and wage on occupational choice. In order to identify this, we need to control for
potential confounders, that is, factors that affect occupational choice but are also correlated
with our variables of interest, that is, occupational prestige and wage.
Gender differences in occupational choices have been explained by women’s greater
desire for work compatible with child-bearing and rearing. We therefore control for average
work hours in each occupation. In a robustness check, we also include preferences for short/
convenient work hours, job safety, or a challenging job, to assess whether work hours capture
this potential confounder. Because women are on average more risk averse than men (e.g.,
Powell and Ansic, 1997; Croson and Gneezy, 2009, for experimental evidence), and risk
might be correlated with occupational prestige – which is possible if, for example,
occupations in the public sector have higher prestige and lower risk – we also include the
unemployment rate in all estimations. Since successful entry into occupations with higher
wages (and potentially those with higher prestige as well) requires higher ability, we control
in most regressions for ability, using dummies for the lowest and highest quartile.
Ability can also be seen as a measure of the cost of getting the education required for
occupations. There are two main kinds of cost: lost wages during the education, and the pain
18
In this model the outcome is equal to one if an occupation is chosen and zero otherwise, so it is not possible
to include non-interacted individual characteristics since they do not vary by occupation. As a result, the number
of individual characteristics that can be included given our sample size is limited, so instead we use in some
cases separate regressions to assess their impact.
15
or pleasure (Oreopoulos and Salvanes, 2011) of acquiring the education. While we are not
able to derive direct measures of these costs, lost wages are not as much of a concern for our
data since, in Denmark, there is no tuition and every student receives a stipend unless she
earns money (as is the case in most vocational training programs). For tertiary educations the
stipend level depends on residence (i.e. whether the individual lives alone or with parents)
while apprenticeship wages depend on age and whether the student has an education
agreement. The stipend for students not living at home is around 5,900 DDK/month while
initial apprenticeship wages (with typical work hours of about 37 per week) range between
8,600 and 10,000 DDK/month if the student has an education agreement with a firm;
otherwise the wage is around 6000 (2500) DKK/month if the student is at least 18 years old
(younger than 18 years). 19 To proxy for pain or pleasure of additional education, we conduct
regressions where we include variables of how individuals felt about school after
(compulsory) middle school, and specifically whether they were tired of school or felt the
need to earn money.
While illegal in Denmark, it is possible that women anticipate discrimination in hiring
or promotion. Our data do not allow us to directly assess whether a woman’s choice of
occupation is influenced by (anticipated) discrimination, but we are able to assess the
implications of two types of discrimination, described in more detail below. While this can
only give an idea of the importance of anticipated discrimination for occupational
segregation, our analysis yields important insights into the black box of why gender
differences in preferences for occupational prestige (whether they are the result of anticipated
discrimination or not) affect occupational segregation and result in a gender pay gap.
Furthermore, it is possible and maybe even likely that the 19 year olds we study may not
19
See https://www.ug.dk/flereomraader/maalgrupper/6til10klasse/elevloen .
16
anticipate facing much discrimination. According to a recent study commissioned by the
European Commission on gender equality in Denmark, the public discourse in Denmark
focusses on the role of traditional gender stereotypes for educational and occupational
choices and generally takes gender equality in opportunities as given (Agustin, 2011).
We explore potential evidence for gender roles as the source of the gender differences
in preferences for occupational prestige and wage by including measures of socioeconomic
status and parental occupations. Finally, we report the results of robustness checks with
which we explored additional explanations for our results but found no evidence that these
altered our conclusions.
While with our data we are not able to use exogenous variation as identification
strategy, we can justify the selection on observables because we are able to control for the
most important potential confounders, with the possible exception of discrimination which is
discussed above and below. 20 The remaining potential issue is reverse causality: It is possible
that the share of women in an occupation decreases its occupational prestige (a hypothesis
sometimes put forward in sociology) or that occupational prestige is higher for occupations
that are considered “underpaid” compared to their contribution to society. There is no
evidence for the former (England, 1979; Magnusson, 2009). Indeed, according to the survey
used in our analysis, mixed occupations (defined as having at least 20% of both genders in an
occupation) have the highest occupational prestige. Only three out of the top ten occupations
have more extreme gender differences, two of which are overwhelmingly male (pilots and
civil engineers) and one female (midwives).
20
One additional issue we cannot address in this paper is the role of the marriage market. Positive assortative
mating by education (Bruze, Svarer, and Weiss, 2015) might affect occupational choice through expectations
about a future spouse’s income and occupational prestige.
17
If occupational prestige is higher for “underpaid” occupations we might mistakenly
interpret our results as showing that occupational prestige affects women’s choices even
though it is women’s choices that affect occupational prestige. Consider the following
scenario: Women choose certain occupation because discrimination makes it harder for them
to work in others. This lowers wages, which in turn increases occupational prestige. As a
result, occupations with more women have higher occupational prestige. This is, however,
not the case - in fact, as mentioned above, occupations with relatively similar shares of
women and men have the highest prestige. Note that since the occupational prestige measure
is drawn from a different survey, justification bias (where individuals give higher
occupational prestige to desired occupations) is no concern.
V. RESULTS: THE ROLE OF OCCUPATIONAL PRESTIGE FOR GENDER
DIFFERENCES IN WAGES
Baseline Results
The results of our baseline estimation are presented in Table 1. Columns (1) and (4) show the
odds ratios when only wage, the unemployment rate and work hours are included. 21
Occupations with higher wages, lower unemployment rates, and – in the case of women –
lower average work hours are more likely to be chosen. Men are more influenced by wage
than women and women more by the unemployment rate, consistent with previous findings
that women are more risk averse than men. Women’s greater response to work hours and
unemployment can explain why women in our raw data have both lower wages and lower
occupational prestige. Since the results for work hours and unemployment rate remain almost
21
We also include a flag for occupations with work hours that the EU Labour Force Survey flags as unreliable
because of a small number of observations. Our results are robust to omitting the flag and to excluding those
occupations from the analysis.
18
unchanged through all regression results, we only report those here. Throughout, we test
whether odds ratios between women and men are statistically different using a Wald test.
TABLE 1
BASELINE RESULTS: OCCUPATIONAL CHOICE, CONDITIONAL LOGIT MODEL
(ODDS RATIOS SHOWN)
(1)
1.019*** †
(0.006)
Wage
Wage ×
low ability
Wage ×
high ability
Prestige
Prestige ×
low ability
Prestige ×
high ability
Unemployment
Work hours
Pseudo R2
Log Likelihood
# of individuals
Observations
Women
(2)
0.992
(0.007)
1.322***
(0.050)
0.813***
(0.025)
0.797***
(0.020)
0.048
-3,806.06
929
68,746
0.853*** †
(0.026)
0.756***
(0.023)
0.055
-3,777.57
929
68,746
(3)
1.001 ‡
(0.009)
0.893***
(0.015)
1.016
(0.014)
1.254***
(0.061)
0.974 ‡
(0.080)
1.350***
(0.117)
0.849*** †
(0.027)
0.744***
(0.023)
0.075
-3,698.55
929
68,746
(4)
1.038*** †
(0.004)
Men
(5)
1.019***
(0.005)
1.145***
(0.034)
0.920***
(0.019)
1.004
(0.013)
0.030
-3,618.31
867
64,158
0.923*** †
(0.018)
0.997
(0.014)
0.033
-3,609.82
867
64,158
(6)
1.024*** ‡
(0.008)
0.959***
(0.012)
1.009
(0.012)
1.183***
(0.050)
0.803*** ‡
(0.054)
1.336***
(0.109)
0.918*** †
(0.018)
0.987
(0.014)
0.050
-3,544.60
867
64,158
Notes
Robust standard errors are shown in parentheses. ^ p<0.15, * p<0.10, ** p<0.05, *** p<0.01. Bold (†, ‡) in
columns (1)-(6) indicates statistically significantly different odds ratios between men and women at the 1% (5%,
10%) percent level. Also included but not shown is a flag for imprecisely measured work hours.
Including occupational prestige in the regression (shown in columns 2 and 5) reduces
the effect of wage (now net of prestige) for women and men, and for women the odds ratio of
19
wage is not statistically different from zero. 22 Women are more influenced by an
occupation’s prestige, supporting our hypothesis. A one unit increase in prestige (about two
thirds of a standard deviation, and the equivalent of moving in terms of occupational prestige
from a physiotherapist to a police officer) increases women’s probability of choosing an
occupation 1.3 times and men’s 1.1 times.
Adding ability shows that there are differences across the ability distribution (columns
3 and 6). This is especially important since women have higher test scores in our sample.
Individuals in the lowest ability quartile choose occupations with lower wages, consistent
with the interpretation that it is more difficult or costly for them to acquire the necessary
education and to be successful in occupations with higher wages. This effect is stronger for
women. The odds ratio of prestige, on the other hand, does not differ for women in the lowest
ability range compared to those in the middle of the ability distribution, but is lower for men.
Prestige is more important for high ability individuals of both genders. This suggests that
occupations with higher wages or higher prestige are more difficult to get into, and that
women substitute prestige for wages to a higher degree than men at low ability levels.
Using this third specification, we compare women’s predicted wages with the
counterfactual prediction in which women have men’s preferences for occupational prestige
and wages but retain their own ability levels and aversion to unemployment and higher work
hours. This comparison shows that about half (48%) of the 5.9% wage gap in our data can be
explained by the gender differences in the effects of wages and occupational prestige on
occupational choice. Figure 2 shows the resulting changes in predicted probabilities with
22
While the lack of importance of wages for women’s occupational expectations at first might seem
surprising, it is in line with previous findings that, compared to non-pecuniary factors, earnings have only small
effects on postsecondary choice of major, especially for women (Zafar, 2013; Wiswall and Zafar, 2011).
20
60
40
20
0
-20
-40
occupations sorted by occupational prestige
FIGURE 2. Percentage change in predicted probabilities from the counterfactual (Table 1, specification 3) –
r r r k r t t t t r r r t r r r t n r r r r r ic y n h n r t r r r t k t r n e t r t e e n t t k r r y r t r r) ) g ic t t r r r t r r e st r r t l) r t
ne rke ive ler he tan tan tan tan lpe rke nte nis ato ke ite gis ria he me se be rke an tar so it so ice an ke ne he pis ler tan nte cia ye en he pis ye rs ia ies ltan oo ne he rm tho en ice to TV sin ed alisltan ge ito ne gis pe cto w if nti ee ssoitec ita ye ilo
ea o dr c utc is is is is he o ai tio er Ba Watolobra eacFar reslumwo ch cre Ma ksmper off end-ma rdeeacera e c sis rpe ctri plo ageacera plo Nuhnic Pr su C sig rape a Au ag off secvie/ erti am urn su /sinAudesig olovelo A id De ginofe rch ospLaw P
e
e
t t
e t
c
h
M en r A (h
n
c s
Cl n w ck use l b assassassass e e wg pcep op
e Li l t ir d P ial me e
p r
la s on at e Ga Th th ffi l a Ca Ele em nc ol ioth em tec -con ic d otogn th tat liceblic moadv par Jo-co cian n d yc de
e
io ru
ia
r m r n
a o
t O ta
sm hoo Ha oc to S B ale ris ht bin
vil te
o
ct /t ho tr rm les en e o ca ldi Re in
h Ph er i l es Po pu w ( in er/ IT usi hio Psem
nksur ch hystion ory HR
al
S P lig c a
n
S Au
p
Ci cia oct
e
(
a
t
rt uBus aredus Fa Sa itch achIn-H ild Bui ach
v
Co sc
t
e
s P a
s
a k re g ri
a fic
/
s
h
n
r
M
B
F
r
a
s
a
e
o
h
e
I
D
c
y
h
n
D
c
e
r
r
f
r
i d
e
i
s
n
W In
K lt
e
F
G
M
C
R
g
e
o
S
k
n
n
P
s
/
i
l
a
O
o
v
/
e
i
i
b
c
c er or c
A
H mu
er
oo ny
at
Jo
d
La
m
ad am n w lan
rn
ch an
m
lle
N
lte
ki
He C rso bu
es
am
Co
r
r
s
A
m
P
og
Un
Pe A
Pr
occupations sorted by occupational prestige. As expected given the gender differences in
odds ratios, women’s expected occupations change significantly.
21
TABLE 2
EXPLORATION OF COST AND DISCRIMINATION (ODDS RATIOS SHOWN)
Wage
Wage × low ability
Wage × high ability
Wage × low ability
(gender specific)
Wage × high ability
(gender specific)
Wage ×
Tired of school
Wage × Needed to
earn some money
Prestige
Prestige ×
low ability
Prestige ×
high ability
Prestige ×
low ability
(gender specific)
Prestige ×
high ability
(gender specific)
Prestige ×
Tired of school
Prestige × Needed
to earn some money
Own gender < 10%
Pseudo R2
Log Likelihood
# of individuals
Observations
(1)
1.006 ‡
(0.009)
0.902***
(0.016)
1.012
(0.014)
Women
(2)
1.001
(0.009)
(3)
1.007
(0.008)
0.902***
(0.014)
1.015
(0.013)
(4)
1.027*** ‡
(0.008)
0.963***
(0.013)
1.005
(0.012)
0.911***
(0.015)
1.020
(0.015)
0.980
(0.017)
0.942* ‡
(0.030)
1.309***
(0.068)
1.022
(0.087)
1.315***
(0.115)
1.306*** ‡
(0.065)
(6)
1.016**
(0.008)
0.956***
(0.013)
1.010
(0.013)
0.963***
(0.013)
1.021*
(0.012)
1.179***
(0.055)
0.970 ‡
(0.075)
1.355***
(0.112)
0.967**
(0.015)
1.013 ‡
(0.021)
1.268***
(0.058)
0.858**
(0.060)
1.261***
(0.106)
1.171*** ‡
(0.049)
0.911
0.803***
(0.071)
1.267***
(0.114)
(0.057)
1.278***
(0.097)
0.768***
(0.066)
1.036 †
(0.118)
0.079
-3,681.97
929
68,746
Men
(5)
1.017**
(0.008)
1.208***
(0.051)
0.807*** ‡
(0.054)
1.333***
(0.109)
0.851**
(0.061)
0.746*** †
(0.073)
0.073
-3,705.08
929
68,746
0.168*** ‡
(0.035)
0.092
-3,631.90
929
68,746
0.056
-3,521.43
867
64,158
0.050
-3,545.77
867
64,158
0.047*** ‡
(0.033)
0.059
-3,513.19
867
64,158
Notes
Robust standard errors are shown in parentheses. ^ p<0.15, * p<0.10, ** p<0.05, *** p<0.01. Bold (†, ‡) in
columns (1)-(6) indicates statistically significantly different odds ratios between men and women at the 1% (5%,
10%) percent level. Included but not shown: Unemployment, work hours, flag for imprecisely measured work
hours.
22
To evaluate whether our findings are related to a lack of more complete cost measures,
we repeated the third regression with further cost measures added (see Table 2, columns 1
and 4): Youth reporting in 2004 that their choice of what to do after (compulsory) middle
school was influenced by their need to earn some money and by being tired of school. We
find that mostly – as expected – being tired of school is associated with lower paying and
lower prestige occupations. Needing to earn money is negatively associated with wage for
women and positively for men, while it is the opposite for prestige, suggesting again that
women opt for occupations with higher prestige and men with higher wages. The results of
interest, odds ratios for prestige and wage, do not change in a significant manner.
To assure that our results are not driven by women choosing so-called “caring”
occupations, we repeated our analysis with added dummies for occupations in education and
health. The results (not shown, available upon request) indicate that this is not the case: in
this specification, gender differences in the odds ratios change only slightly. Women are
more likely to choose occupations in education and health than in other fields, and men less
likely to choose occupations in the health sector.
As mentioned earlier, we are not able to directly test whether expected discrimination
can explain the differential effects of prestige and wage for women and men. We are able to
address two consequences of discrimination, however (shown in Table 2), to assess the
potential importance of this explanation. First, suppose employers are biased against women.
They might nevertheless feel the need to hire some women for legal or public image reasons.
If this is the case, it is likely that women are evaluated separately, and compared only to other
potential female employees. To evaluate this implication of discrimination, we use genderspecific ability quartiles. Second, discrimination (or expected discrimination) in hiring is
more likely if the percentage of one’s own gender in an occupation is low. While this,
however, is endogenous to our parameters of interest (that is, it could also be the result of the
23
sorting by occupational prestige and wages) it is worthwhile to investigate whether
accounting for this affects our results. Hence, we include a dummy variable of whether the
percentage of one’s own gender in an occupation is less than 10%. Overall, we find that
neither of these explorations changes the odds ratios of interest in a statistically significant
manner. The dummy for gender segregation shows that it strongly reduces the probability of
an occupation being chosen, but its effect is not statistically different between women and
men. Both specifications decrease the percentage of the explained wage gap resulting from
different odds ratios for prestige and wage to 23% and 39%. These results point to a potential
additional role of discrimination in explaining occupational segregation. Given our data, we
are not able to explore this further.
The Importance of Gender Roles
One of the explanations for occupational segregation put forward is the influence of gender
role socialization on educational and occupational choices (see, e.g., Eccles, 1994).
Occupational choices are influenced by one’s socioeconomic background and ability (Turner
and Bowen, 1999), and parental income and parental education affect girls’ and boys’
educational expectations differently (Kleinjans, 2010). Gender roles are more traditional in
lower SES families (Dryler, 1998; Vella, 1994) and for lower ability individuals who may be
less likely to challenge gender roles (Ahrens and O’Brien, 1996; Fassinger, 1990), also
because parents’ approval is an important determinant for children’s occupational and
college-major choice (Jacobs, Chhin, and Blecker, 2006; Zafar, 2013). This is supported by
the high degree of occupational segregation of low-ability individuals in our data: The
majority of occupations (79%) in which individuals with low ability expect to work require
vocational training or less, and these occupations are highly segregated with 74% of
occupations having 75% or more workers of one gender compared to 21% of occupations
requiring a master degree. If the importance of occupational prestige and wages for
24
occupational choice is related to gender roles, then we would expect gender differences to be
most pronounced for low SES and low ability individuals. 23
We already found in results presented earlier that this is indeed the case for low-ability
individuals. Indeed, counterfactual predictions by ability show that 76% of the gender wage
gap disappears for the lowest ability quartile once we assume that women have men’s
coefficients for wages and occupational prestige, while it is 32% for the two middle quartiles
and only 17% for the highest ability quartile. To test whether this relationship also holds for
low SES individuals, we include interaction terms between low and high parental education
and income with wage and occupational prestige variables (see Table 3).
We find only weak support for this hypothesis, though this could be related to the
number of variables included for the sample size. Low parental education decreases the
importance of wage for women. Low parental income reduces the importance of occupational
prestige for women’s choices and low parental education the importance of prestige for
men’s choices.
As another way to investigate the role of parental attitudes, we constructed dummies for
a low, medium, or high share of women in the mother’s and the father’s occupation and
included those interacted with wage and occupational prestige in the baseline regression
(results not shown). While this makes the number of variables again rather high given the
sample size, we find that if there are less than one third of women in the father’s occupation
then men (but not women) put less weight on occupational prestige. None of the other odds
ratios are statistically significantly different for women and men. This could be a reflection of
more rigid gender roles for men (Valian, 1999).
23
This can also be thought of as controlling for the difference in culturally transmitted value systems (Corneo
and Jeanne, 2010) or different identity payoffs (Humlum et al., 2012) affecting occupational choice.
25
TABLE 3
GENDER ROLES – PARENTAL BACKGROUND (ODDS RATIOS SHOWN)
Wage
Wage × low ability
Wage × high ability
Wage × low parental income
Wage × high parental income
Wage × low parental education
Wage × high parental education
Prestige
Prestige × low ability
Prestige × high ability
Prestige × low parental income
Prestige × high parental income
Prestige × low parental education
Prestige × high parental education
Pseudo R2
Log Likelihood
# of individuals
Observations
(1)
Women
1.012
(0.015)
0.899***
(0.015)
1.012
(0.014)
0.992
(0.015)
1.013
(0.015)
0.978^
(0.014)
0.987
(0.021)
1.359***
(0.111)
1.010
(0.084)
1.303***
(0.115)
0.859*
(0.071)
1.027
(0.093)
0.908
(0.073)
1.286^
(0.207)
0.082
-3,671.66
929
68,746
(2)
Men
1.012
(0.011)
0.959***
(0.013)
1.008
(0.013)
1.008
(0.014)
1.013
(0.014)
1.007
(0.013)
1.007
(0.019)
1.357***
(0.093)
0.847**
(0.057)
1.267***
(0.106)
0.899^
(0.064)
1.005
(0.080)
0.813***
(0.057)
1.146
(0.144)
0.055
-3,526.62
867
64,158
Notes
Robust standard errors are shown in parentheses. ^ p<0.15, * p<0.10, ** p<0.05, *** p<0.01. Bold (†,
‡) in columns (1)-(2) indicates statistically significantly different odds ratios between men and women
at the 1% (5%, 10%) percent level. Included but not shown: Unemployment, work hours, flag for
imprecisely measured work hours. The odds ratios of Prestige x low ability are borderline statistically
significantly different at a p-value of 0.102.
26
TABLE 4
ROBUSTNESS CHECKS (ODDS RATIOS SHOWN)
Wage
Wage × low ability
Wage × high ability
Wage × Short/
convenient work
hours
Wage × Job safety
Wage × Effort and
perseverance
Wage ×
Not competitive
Prestige
Prestige × low ability
Prestige × high
ability
Prestige × Short/ convenient work hours
Prestige × Job safety
Prestige × Effort
and perseverance
Prestige ×
Not Competitive
Pseudo R2
Log Likelihood
# of individuals
Observations
(1)
1.007
(0.010)
0.895***
(0.015)
1.014
(0.014)
1.033
(0.035)
Women
(2)
1.001 ‡
(0.009)
0.896***
(0.016)
1.009
(0.014)
(3)
0.995 †
(0.010)
0.894***
(0.015)
1.014
(0.014)
0.960***
(0.015)
(4)
1.026***
(0.008)
0.960***
(0.012)
1.009
(0.012)
1.001
(0.034)
1.250*** ‡
(0.062)
0.950
(0.079)
1.353***
(0.118)
1.021***
(0.005)
1.017
(0.013)
1.205*** ‡
(0.065)
0.982
(0.081)
1.333***
(0.116)
1.215***
(0.054)
0.809***
(0.055)
1.330***
(0.109)
0.787
(0.141)
0.879*
(0.066)
1.011
(0.037)
0.078
-3,659.18
922
68,228
(6)
1.024*** †
(0.009)
0.959***
(0.013)
1.009
(0.012)
0.983
(0.014)
1.019***
(0.007)
1.298***
(0.069)
0.961
(0.080)
1.327***
(0.115)
0.881
(0.127)
0.873^
(0.072)
Men
(5)
1.021*** ‡
(0.008)
0.967***
(0.013)
1.004
(0.012)
0.077
-3,619.45
911
67,414
1.198*** ‡
(0.051)
0.788***
(0.055)
1.336***
(0.111)
0.998
(0.011)
1.117** ‡
(0.055)
0.817***
(0.055)
1.318***
(0.109)
0.988
(0.033)
1.141*
(0.085)
0.077
-3,690.60
929
68,746
0.051
-3,511.81
860
63,640
0.053
-3,443.12
845
62,530
1.127*
(0.069)
0.051
-3,541.60
867
64,158
Notes
Robust standard errors are shown in parentheses. ^ p<0.15, * p<0.10, ** p<0.05, *** p<0.01. Bold (†, ‡) in
columns (1)-(6) indicates statistically significantly different odds ratios between men and women at the 1% (5%,
10%) percent level. Included but not shown: Unemployment, work hours, flag for imprecisely measured work
hours.
27
Robustness Checks
Finally, in Table 4, we report the results of analyses using an additional measure of
preferences of short/ convenient work hours, job safety, or a job that is challenging, in case
that work hours do not adequately control for preference differences for such job
characteristics; a psychosocial trait that has been found to influence occupational choice and
may be related to gender differences in occupational choice: the non-cognitive trait of effort
and perseverance from the PISA data (Antecol and Cobb-Clark, 2013; Cobb-Clark and Tan,
2011); and preference for competition (Croson and Gneezy, 2009; Kleinjans, 2009), but find
no evidence that these alter our conclusions. In a different specification, we also included a
measure of self-confidence (results not shown) in case that women have lower selfconfidence (Niederle and Versterlund, 2007), and therefore expect it to be more difficult to
complete their education or to find work in occupations with higher salary or status. This
could affect especially low-ability women. However, we find no effect for self-confidence for
either gender once we control for ability. To ensure that our results are not driven by the
assumption that women and men have the same wage, we also conduct our baseline
estimation for women with wages that are 6% lower than the median wage (roughly
corresponding to the gender wage gap unrelated to occupational segregation). The results (not
shown) do not change.
VI. CONCLUSIONS
Despite women’s increased educational achievement, women and men often work in different
occupations. This occupational segregation can explain up to one half of the raw gender wage
gap, but it is not well understood why such segregation persists. In this paper, we investigate
an explanation that is based on job attributes, and more precisely, the idea that women place
more weight on occupational prestige than men. Women have been found to value an
28
occupation’s social contribution more than men, which we proxy with the occupational
prestige of an occupation. We employ the theory of equalizing differences in a labor market
to argue that if occupational prestige gives benefits to holders of occupations. If these
benefits vary by gender, then differences in occupational choices result in wage differences,
or a gender wage gap.
To investigate this hypothesis, we use a Danish data set that includes rich information
on individuals’ expected occupations, ability, parental background, and occupational
characteristics.
We find that women expect to work in occupations with higher occupational prestige
and lower wages than men, with the greatest difference for low-ability individuals. This is
consistent with the hypothesis that part of the gender wage gap can be explained by the
occupational segregation caused by women’s stronger preference for occupational prestige –
these occupations with higher occupational prestige have, in an equilibrium setting in a
competitive labor market, lower wages. Counterfactual predictions show that a significant
part (up to one half) of the gender wage gap caused by occupational segregation can be
explained by these preference differences. We conclude from this that an important fraction
of the gender wage gap results from different choices that women and men make that are
based on differences in preferences for occupational prestige. While we are not able to
identify the origin of these gender differences, we find some evidence in line with the
hypothesis of gender roles as a potential source. If this is indeed the case, then gender
differences in preferences for occupational prestige can help us understand why gender roles
(through their effect on occupational choices) lead to gender differences in wages. Our data
do not permit a causal analysis to identify whether preference differences result from gender
roles or anticipated discrimination. The analysis can, however, help explain why the
29
narrowing of the gender wage gap has slowed down and remains significant (Goldin, 2006;
Blau and Kahn, 2007) if gender roles are slow to change.
APPENDIX A
TABLE A1
OCCUPATION-SPECIFIC CHARACTERISTICS
(SORTED BY OCCUPATIONAL PRESTIGE)
Occupation
Pilot
Lawyer
Doctor (GP) *
Doctor (hospital)
Researcher in private company
Architect
Associate professor
Civil engineer
Soccer player *
Dentist
Midwife
Actor
Programmer/System developer
Psychologist
Fashion designer
Auditor
Politician
Musician/singer
IT-consultant
Journalist
Ambulance driver/paramedic
Person working in advertising
Camera crew (movie/TV)
Head clerk (public sector)
Police officer
Occupational Prestige Score Wage Unemployment
8.31
8.11
7.89
7.76
7.45
7.39
7.27
7.22
7.15
7.01
6.96
6.95
6.79
6.70
6.69
6.63
6.55
6.51
6.43
6.42
6.40
6.38
6.36
6.28
6.23
59.76
41.33
53.91
53.91
.
37.60
34.75
41.87
28.69
37.36
32.36
29.36
39.88
34.39
26.21
35.96
.
30.96
43.04
36.23
27.89
31.13
33.27
35.91
32.55
1.91
0.94
0.58
0.58
.
4.09
4.23
1.47
2.60
0.61
0.49
16.75
1.56
3.80
3.62
0.72
.
4.41
1.86
2.79
4.11
3.59
4.51
1.02
0.22
30
Occupation
Real estate agent
Author
Army officer
Photographer
Graphic designer
Cook
HR-consultant
Priest
Laboratory technician
Nurse
Communication employee
Physiotherapist
High school teacher
Insurance agent
Business high school teacher *
Bank employee
Electrician
Carpenter
Dental assistant
Office clerk
Alternative health therapist
Teacher
Gardener
Joiner/cabinet-maker
Flight attendant
Prison officer
Sales person
Nursing aide in a hospital *
Welder
Mason
Secretary
Auto mechanic
Social worker
Vocational teacher *
Glazier *
Plumber
Hair dresser
Farmer
Preschool teacher (children aged 3-6)
Train conductor *
Librarian
Cosmetologist
Occupational Prestige Score Wage Unemployment
6.20
6.16
6.03
5.96
5.72
5.69
5.67
5.60
5.50
5.39
5.33
5.20
4.98
4.87
4.86
4.85
4.70
4.40
4.35
4.32
4.30
4.28
4.25
4.25
4.19
4.10
4.08
4.06
4.01
4.01
4.00
3.99
3.98
3.97
3.97
3.91
3.90
3.84
3.83
3.79
3.78
3.77
40.26
36.23
26.59
33.27
33.65
20.41
28.56
30.86
26.82
28.77
36.23
26.06
38.20
45.51
38.20
28.58
28.15
25.28
20.67
26.11
19.48
32.74
23.35
25.28
32.28
27.84
35.09
23.66
27.44
25.64
26.11
26.41
29.10
38.20
25.91
28.18
19.43
19.56
25.70
35.10
32.04
19.43
1.09
2.79
2.60
4.51
1.91
4.46
1.53
2.54
2.26
0.66
2.79
2.87
2.78
1.42
2.78
1.33
3.12
4.34
3.57
3.66
2.19
1.88
5.22
4.34
2.85
1.14
1.84
2.54
3.30
7.05
3.66
1.67
1.06
2.78
3.33
3.77
2.19
1.12
3.03
0.79
2.63
2.19
31
Occupation
Occupational Prestige Score Wage Unemployment
Security guard *
Baker
Waiter
Machine operator
Receptionist
Building painter
Medical Orderly*
Mail carrier *
Nanny/Child care worker
In-Home caregiver
Preschool teacher assistant
(children aged 3-6)
Fisherman *
Kitchen assistant
Sales assistant
Butcher
Farm worker
Scaffolder *
Nursing home assistant *
Road worker *
Window cleaner *
Warehouse clerk
Taxi driver *
Trash collector *
Mover *
Bus/truck driver
Unskilled construction worker
Parking attendant *
Cashier *
Cleaner
Advertising delivery person *
Unemployment benefit recipient *
Welfare recipient *
Mean
3.65
3.40
3.40
3.28
3.25
3.16
3.13
3.07
3.01
2.94
2.89
2.88
2.88
2.84
2.74
2.74
2.74
2.72
2.71
2.58
2.58
2.49
2.46
2.45
2.41
2.29
2.22
1.87
1.57
1.31
0.68
0.43
4.55
27.19
24.26
22.09
28.51
24.35
25.57
23.66
25.54
21.27
20.45
23.66
3.55
3.71
6.25
2.50
4.31
5.65
2.54
2.93
6.17
4.92
2.54
12.96
20.41
21.14
21.42
19.56
29.20
23.66
25.46
25.69
25.48
22.49
28.75
25.48
22.49
28.39
27.19
21.14
21.96
25.11
.
.
29.36
0.95
4.46
3.22
5.91
1.12
8.32
2.54
8.84
2.18
5.82
2.99
4.03
5.82
2.99
8.67
3.55
3.22
6.36
2.24
.
.
3.25
* denotes occupations in which no one expected to work.
Wage is measured as median hourly wage and divided by 5 to approximate 1 U.S dollar. The occupational
prestige score is that of 18-29 year olds. Wage could not be determined for ‘researcher in private company’ and
‘politicians’.
32
TABLE A2
VARIABLES, DATA SOURCE AND SUMMARY STATISTICS OF WORKING SAMPLE
BY GENDER: MEANS (STANDARD DEVIATIONS IN PARENTHESES)
Variable
Derivation (Data source)
Attributes of expected occupation
Occupational prestige evaluated from 0-10 by 18-29
Occ. prestige score
years old (Ugebrevet A4).
Median wage divided by average work hours (see
Wage
below) and divided by 5 to approximate 1 U.S. dollar
(Statistics Denmark’s registers).
Unemployment
Percentage unemployment in occupation
(Statistics Denmark’s registers).
Work hours
Average work hours in occupation
(EU Labour Force Survey).
Work hours imprecise
An indicator variable for imprecise work hours
(EU Labour Force Survey).
= 1 if own gender is less than 10% in an occupation
Own gender < 10%
(Statistics Denmark’s registers).
Education occupations
Health occupations
% women in mother’s
occupation
% women in father’s
occupation
= 1 if education related occupations (associate
professor, nanny/ child care worker, any type of
teacher and teacher assistant), 0 otherwise.
= 1 if health related occupations (alternative health
therapist, ambulance driver/ paramedic, dental
assistant, dentist, doctor,
in-home caregiver,
midwife, nurse and nurse aides, medical orderly,
nursing home assistant, psychologist), 0 otherwise.
Percentage women in mother’s occupation
(Statistics Denmark’s registers).
Percentage women in father’s occupation
(Statistics Denmark’s registers).
Men Women
(1)
(2)
5.41
5.16
(1.52) (1.47)
32.27 30.37
( 8.07) ( 7.46)
2.64
2.55
( 1.93) ( 1.96)
40.46 39.13
(3.10) (2.44)
0.03
0.04
0.25
0.63
0.09
0.24
0.04
0.23
70.25
(22.76)
27.92
(26.41)
71.94
(20.32)
28.05
(26.71)
Individual characteristics
Ability
Affecting choice after
lower secondary school
Tired of school
Needed to earn money
Know which sector
Reading score (WLEREAD), lowest and
489.86 510.40
highest quartile used (PISA).
(99.95) (93.67)
Answer to “Here are some questions about what
influenced your decision about what to do
immediately after 9th or 10th grade. (Scale from 1
(none) - 5 (”extremely high”): What influence did the
following have” (2004 follow-up survey).
“You were tired of school”,
0.23
0.17
= 1 if “high” or “extremely high”.
“You needed to earn some money”,
0.12
0.07
= 1 if “high” or “extremely high”, 0 otherwise.
Answer to “You know today which sector or field you 0.72
0.71
want to work in. And you will probably stay there”,
= 1 if “partly agree” or “totally agree”, 0 otherwise
33
Job characteristics
Short/convenient work
hours
Job safety
Job is challenging
Competition
Effort and perseverance
Parental Characteristics
Parental income
(2004 follow-up survey).
Answer to “Which of the following three qualities do
you consider most important in a job?” (2004
follow-up survey).
= 1 if “short/convenient work hours”, 0 otherwise.
0.03
= 1 if “job safety”, 0 otherwise.
0.17
(omitted category)
Answer to “Outside the world of sports, people 0.48
should compete as little as possible” = 1 if “totally
disagree” or “partly disagree”, 0 otherwise (2004
follow-up survey).
Warm estimate of EFFPER (PISA).
-0.00
(0.94)
(Statistics Denmark’s registers)
Parental income / 50.000 to approximate US $10.000
11.89
(5.37)
0.03
0.18
0.34
0.02
(0.94)
11.94
(5.41)
Highest parental education
Low
= 1 if basic, high school or vocational, 0 otherwise.
0.59
0.61
Medium
= 1 if short or medium, 0 otherwise.
0.31
0.30
High
= 1 if long term, 0 otherwise.
0.11
0.09
Notes
Bold (†, ‡) in columns (1)-(2) indicates statistically significantly different means between men and
women at the 1% (5%, 10%) percent level.
34
FIGURE A1. Distribution of Wage – Occupations sorted by Prestige
35
lle
d
co
ns
tr
B uc
W us tionCle
In are/tru w ane
du ho ck or r
Pr
k
s
es
Fa triausedriver
ch
e
r l
oo K Sal m a bu cler r
Na l teitch es ss tch k
nn ac en ass istaer
y/ In he as is nt
Ch -H r s ta
ild o as ist nt
Bu came sistant
il r e h a n
M din w elpet
ac Re g or r
hi ce pa ke
n e p in r
optionter
er is
t
Co
B at
Pr sm Wakor
es
et a er
ch L ol ite
oo ib og r
l t rar ist
Ha Feac ian
ir ar he
S dr m r
Auoc Pluesser
to ial m er
m wo be
Seech rker
cr an r
e ic
SaBla Matary
c
Jo
le ks so
P
F
s
in li r is p m n
er gh o e it
Al
/c t n rs h
te
ab at of on
rn
in ten fic
at
et d er
ive
- a
he Ga ma nt
al T rde ke
th e n r
De O th ac er
nt ffi era he
al ce p r
is
a
C s cle t
Ba arsist rk
E
Co Hig Ins nk le penant
h ur emctri ter
m
c
s a
m
un P chonceplo ian
ica hy ol a ye
La
tio sio teage e
bo
n th c nt
ra emera he
r
to
ry plo pis
te N y e t
HR chn urse
G -co Piciae
ra
n
p n s r ie
O Phic ultast
ffi h d C n
ce o e o t
r i tog sig ok
H
n
R
n
Pe Cead ea therap er
h
le
am c
r
s
Am on e ler st Aarmer
P
a
bu w ra k (p oli te uth y
la ork cre ub ce ag or
nc in w li o e
e g (mc s ffic nt
dr in o e e
ive ad v cto r
r/p veie/T r)
ar rtis V)
a
Pr
I
M T- Joumeing
og
us co rn di
ra
Fa icia nsualisc
m
m
er shi n/s ltant
in t
o
/S
ys P n d Audge
te sy es it r
m ch ig or
de ol ne
ve og r
lo ist
As
A per
M
C
so i
id ctor
v
cia il D w
te e e ife
Do prnginntis
ct Aofe ee t
or r s r
(h chi so
os te r
c
Lapita t
wy l)
Pi er
lo
t
Un
sk
i
0
20
Wage
40
60
APPENDIX B: DATA CONSTRUCTION AND SAMPLE SELECTION
We match occupation-specific variables and expected occupations using the four digit
DISCO code, the official Danish version of the International Classification of Occupations
(ISCO) by the International Labour Organisation. This is generally straightforward though in
16 cases we are not able to distinguish occupations in the ISCO classification, implying that
these occupations are coded with the same occupation-specific characteristics. This applies to
304 individuals, including camera men and women/ photographers, journalists/ authors/
communication employees, cooks/ kitchen assistants, carpenters/ joiner/ cabinet-makers,
secretaries/ office clerk, alternative therapists/ hair dressers/ cosmetologists, and farmers/
farm assistants. Another side of this issue is that occupations from the occupational prestige
survey often share ISCO codes with occupations not in the survey. For instance, fashion
designers share an ISCO code with decorators, interior architects, and other types of
designers.
Table B1 summarizes the sample selection by gender. Our final sample excludes
individuals if the reading score is missing or if the individual has no parent in the registry
data. 50 men and 34 women are dropped because of missing answers to the expected
occupation question. Also excluded are individuals who answered “don’t know” (140 men
and 120 women) or “nothing” (20 men and 15 women).
Along with wage, our primary explanatory variable is occupational prestige derived
from the ranking of 99 specific occupations. Some individuals gave answers too vague to be
classified as a specific occupation (e.g., “something with people” or “trade”) leading to a
drop of 68 men and 105 women and leaving in total 2,491 observations. 72.4% of these
individuals expect to work in one of the ranked occupations. 346 men and 342 women expect
to work in occupations not included in the ranking. The ranking does not include managerial
level occupations and self-employed/owners, and 170 (78 men and 92 women) individuals
36
are dropped for this reason. The remaining 518 individual aspire to work in traditional – but
unranked – occupations. Among the largest unranked occupations are economists (19
individuals) and translators (13). Finally, dropped 7 individuals who aspire to work in
occupations that could not be matched to an ISCO code (“Researcher in a private company”
and ”Politician”). In total we match individuals to 74 of the ranked occupations.
TABLE B1
SAMPLE SELECTION
Sample Restriction
Danish PISA Longitudinal Data *
Reading score, parental income
or parental education missing
Occupational expectations
No answer recorded
Answer “don’t know”
Answer “nothing”
Answer too vague
Occupation not in occ. prestige survey
Occupation has no wage data
Number of individuals
Reshape of data: 1,796 ⋅ 99
Occupations in which no one expects
to work or without wage data
Estimation sample
Individuals dropped
Wome
Men
Total
n
Occupation
s dropped
Number of
observation
s
3,073
9
21
30
3,043
50
140
20
68
346
34
120
15
105
342
84
260
35
173
688
7
2,959
2,699
2,664
2,491
1,803
1,796
1,796
177,804
25 ⋅ 1,796 =
44,900
132,904
132,904
* This includes individuals who were tested in PISA and answered the follow-up survey.
Note that Statistics Denmark does not allow the reporting of characteristics for fewer than four individuals.
If a respondent answered more than one occupation we use the first occupation
mentioned. In the sample of 1,796, 123 individuals mentioned more than one expected
occupation. As a robustness check the baseline estimations were conducted dropping
individuals mentioning more than one occupation. Results are in line with our baseline
results.
37
Table B2 compares the estimation sample (1) with all individuals dropped (2) and by
reason for dropping (3)-(5). Comparing columns (1) and (2) shows no difference in parental
education between the individuals in and out of the sample while parental income and ability
are somewhat lower for individuals in the sample. Column (4) shows that the differences
between individuals in and out of the sample are driven by those expecting to work in
occupations not in the ranking. A likely explanation is that managerial occupations are not
included in the prestige ranking.
TABLE B2
COMPARISONS OF OBSERVATIONS IN AND OUT OF SAMPLE
Female
Reading score /100
Parental income
Parental education
Low
Medium
High
N
(1)
In
Sample
(2)
Not in
sample
0.517
5.005
(0.973)
11.917
(5.393)
0.496
5.106
(0.929)
12.286 ‡
(6.172)
0.601
0.303
0.095
1,796
0.578
0.308
0.114
1,247
(3)
(4)
(5)
Sub groups of dropped observations
Too
Not in
“Nothing” or
vague
survey
“don't know”
0.458 ‡
0.607 †
0.497
5.031
5.076 ‡
5.338
(0.960)
(0.783)
(0.945)
11.917
13.144 ‡
12.040
(5.475)
(8.476)
(5.475)
0.617
0.281
0.102
295
0.543
0.306
0.150 ‡
173
0.580
0.313
0.108
688
Table B3 shows the marginal effects of a generalized ordered logit estimation with the
outcome variable given by the respondents’ answer to the question: “You know today which
sector or field you want to work in. And you will probably stay there”. Of the 3,043
individuals, 6 did not answer the question while 76 answered “don’t know”, leaving 2,961
individuals answering from “totally disagree” to “totally agree”. The results show a clear
relationship between uncertainty about which sector to work in and answers given with
respect to specific occupation at age 30. Vague answers are primarily given by the
respondents who are uncertain about the sector.
38
TABLE B3
MARGINAL EFFECTS OF GENERALIZED ORDERED LOGIT:
“YOU KNOW TODAY WHICH SECTOR OR FIELD YOU WANT TO WORK IN. AND YOU
WILL PROBABLY STAY THERE”
(1)
(2)
(3)
Totally/ partly disagree
Partly/ totally
Both/ and
agree
Female
0.011
0.005
-0.016
(0.012)
(0.005)
(0.017)
Reading Ability
0.009
0.004
-0.013
(0.007)
(0.003)
(0.009)
Parental income
0.004***
-0.002**
-0.001
(0.001)
(0.001)
(0.002)
Low parental education
-0.041***
-0.017***
0.059***
(0.013)
(0.005)
(0.018)
High parental education
0.032^
0.014^
-0.046^
(0.021)
(0.009)
(0.030)
Short/convenient work
0.056**
0.024**
-0.080**
hours
(0.028)
(0.012)
(0.040)
Job safety
0.003
0.001
-0.004
(0.015)
(0.006)
(0.022)
Too vague
0.146***
0.061***
-0.207***
(0.023)
(0.010)
(0.032)
Not in survey
0.018
0.007
-0.025
(0.014)
(0.006)
(0.020)
Don't know and nothing
0.299***
0.077***
-0.375***
(0.018)
(0.018)
(0.027)
Pseudo R2
0.060
Log likelihood
-2,431.09
Observations
2,961
Notes
Standard errors shown in parentheses, * p<0.10, ** p<0.05, *** p<0.01.
Overall, the evidence implies that the working sample consists of individuals with
slightly lower ability and SES. As discussed in the results section, these are the individuals
most affected by the gender differences in occupational prestige and wages. Since we do
control for ability in our analysis and, in some cases, for parental background, we do not
expect that the loss of sample affects our results in an important way. We also report further
robustness checks in Appendix C.
39
APPENDIX C: ROBUSTNESS CHECKS – THE TREIMAN SCALE OF
OCCUPATIONAL PRESTIGE AND SAMPLE SELECTION
In our main specification, we use the occupational prestige scores from the Danish
Ugebrevet survey. We prefer this measure to the Ganzeboom and Treiman scale because it is
based on a recent assessment by Danish individuals aged 18 to 29 of specific occupations
relevant in a Danish context, and as such a better measure of the prestige of occupations
assigned by Danish youth. Hence we consider the ranking to be a better representation of the
relevant occupational prestige for Danish youth when choosing education and career path. In
what follows, we present robustness checks that show that a) the use of the Ganzeboom and
Treiman prestige scale does not change our results in any significant way, and b) that the
non-random sample selection resulting from the omission of occupations not included in the
Danish survey does not affect our results.
The Treiman Scale
Ganzeboom and Treiman (1996) assign prestige to occupations using the International
Standard Classification of Occupations (ISCO). In what follows, we report the results using
this updated version of Treiman’s (1977) Standard International Occupational Prestige Scale
(SIOPS).
Table C1 reproduces Table 1 with the SIOPS prestige scale. There is a difference in
the number of observations (but not in the number of individuals) since a few of the
occupations we are able to match in the Danish prestige survey are categorized as the same
occupation in the SIOPS scale, and the SIOPS scale includes a few occupations that are
broader categories than in the prestige survey. In total, the second effect is larger, and the
number of observations in the results shown in Table C1 is larger than in Table 1 as each
individual has more occupations to choose from. The results are comparable to the results
presented in Table 1 indicating that the selected prestige ranking is not a major concern with
40
respect to the generalizability of our results.
TABLE C1
REPRODUCTION OF TABLE 1 USING THE SIOPS PRESTIGE SCALE
(1)
0.997
(0.006)
Wage
Wage ×
low ability
Wage ×
high ability
Prestige
Prestige ×
low ability
Prestige ×
high ability
Unemployment
Work hours
Women
(2)
0.910***
(0.008)
1.073***
(0.004)
0.794***
(0.020)
0.867***
(0.010)
0.063
-3937.04
929
85,468
0.859***
(0.024)
0.817***
(0.011)
0.097
-3791.65
929
85,468
(3)
0.898***
(0.011)
0.870***
(0.022)
1.088***
(0.017)
1.080***
(0.005)
0.986^
(0.009)
1.009 †
(0.007)
0.860*** †
(0.025)
0.802***
(0.011)
0.126
-3671.55
929
85,468
(4)
1.027***
(0.006)
Men
(5)
1.028***
(0.007)
0.999
(0.003)
0.948**
(0.021)
1.000
(0.010)
0.071
-3643.97
867
79,764
0.947**
(0.022)
1.001
(0.010)
0.071
-3643.94
867
79,764
(6)
1.031***
(0.009)
0.947***
(0.016)
1.017
(0.013)
1.002
(0.005)
0.979***
(0.007)
1.030*** †
(0.008)
0.933*** †
(0.023)
0.990
(0.010)
0.089
-3572.00
867
79,764
Pseudo R2
Log Likelihood
# of individuals
Observations
Notes
Odds ratios shown. Robust standard errors are shown in parentheses. ^ p<0.15, * p<0.10, ** p<0.05, ***
p<0.01. Bold (†, ‡) in columns (1)-(6) indicates statistically significantly different odds ratios between men and
women at the 1%, 5%, 10%) percent level. Also included but not shown is a flag for imprecisely measured work
hours.
Sample Selection: Omission of Occupations from our Sample
The prestige measure used in our estimations is based on a ranking of 99 representative
occupations, but we were only able to match 74 expected occupations. Hence, there is some
concern of the effects of this non-random sample selection on our results. Since all of those
are, however, included in the Ganzeboom and Treiman (1996) SIOPS prestige scale, we can
use it to assess whether the omission of occupations affects our results.
Table C2 reproduces Table 1 using the larger sample. The number of observations
increases significantly for two reasons: First, the number of individuals increases since we
now have 825 more individuals in the sample who expect to work in occupations not
included in the Danish prestige survey; and second, since each individual can choose among
41
more occupations. The results presented in Table C2 are qualitatively similar to the results
presented in Table C1 and Table 1, and we conclude from this that there is no evidence that
the omission of occupations in our analysis drives our results.
TABLE C2
REPRODUCTION OF TABLE 1 WITH A LARGER SAMPLE,
USING THE SIOPS PRESTIGE SCALE
Wage
(1)
1.003
(0.004)
Wage ×
low ability
Wage ×
high ability
Prestige
Prestige ×
low ability
Prestige ×
high ability
Unemployment
Work hours
Women
(2)
0.920***
(0.006)
1.068***
(0.003)
0.808***
(0.016)
0.938***
(0.007)
0.043
-5852.35
1,353
124,476
0.847*** †
(0.018)
0.894***
(0.008)
0.076
-5656.00
1,353
124,476
(3)
0.905***
(0.009)
0.919*** †
(0.019)
1.079***
(0.014)
1.072***
(0.004)
0.992
(0.008)
1.005 †
(0.005)
0.849*** ‡
(0.019)
0.884***
(0.008)
0.043
-5852.35
1,353
124,476
(4)
1.028***
(0.004)
Men
(5)
1.028***
(0.005)
1.000
(0.003)
0.908***
(0.017)
1.006
(0.007)
0.053
-5428.46
1,268
116,656
0.908*** †
(0.018)
1.006
(0.007)
0.053
-5428.46
1,268
116,656
(6)
1.030***
(0.006)
0.964*** †
(0.011)
1.013^
(0.009)
1.003
(0.004)
0.981***
(0.006)
1.023*** †
(0.006)
0.899*** ‡
(0.018)
1.000
(0.007)
0.064
-5364.49
1,268
116,656
Pseudo R2
Log Likelihood
# of individuals
Observations
Notes
Robust standard errors are shown in parentheses. ^ p<0.15, * p<0.10, ** p<0.05, *** p<0.01. Bold (†, ‡) in
columns (1)-(6) indicates statistically significantly different odds ratios between men and women at the 1%
5%, 10%) percent level. Also included but not shown is a flag for imprecisely measured work hours.
ACKNOWLEDGMENTS
We would like to thank Ugebrevet A4 and Henrik Feindor Christensen at Analyse Denmark
for access to the occupational prestige survey, and Julie Cullen, Mona Larsen, Helena Skyt
Nielsen, Stefanie Schurer, Edward J. Schumacher, Stephan Thomsen, and participants at the
AEA, SOLE, ESPE, and WEA meetings, the IZA/CEPR European Summer Symposium in
Labour Economics, as well as seminar participants at KORA, Aarhus University, and
42
Binghamton University for helpful comments. Kleinjans thanks the Milton A. Gordon Fund
for Scholarly & Creative Activities at CSUF for funding.
REFERENCES
AGUSTIN, L. R. (2011). The Policy on Gender Equality in Denmark. Directorate General for
internal Policies. Policy Department C: Citizen’s rights and constitutional affairs,
gender equality. European Parliament.
AHRENS, J. A. and O’BRIEN, K. M. (1996). Predicting Gender-Role Attitudes in
Adolescent Females: Ability, Agency, and Parental Factors. Psychology of Women
Quarterly, 20. 409-417.
ANDREONI, J. and VESTERLUND, L. (2001). Which is the Fairer Sex? Gender differences
in altruism, Quarterly Journal of Economics, 116(1), 293-312.
ANKER, R. (1997). Theories of Occupational Segregation by Sex: An overview.
International Labor Review, 136(3),315-339.
ANTECOL, H. and COBB-CLARK, D. A. (2013). Do Psychosocial Traits Help Explain
Gender Segregation in Young People’s Occupations? Labour Economics, 21, 59-73.
BAYARD, K., HELLERSTEIN, J., NEUMARK, D. and TROSKE, K. (2003). New Evidence
on Sex Segregation and Sex Differences in Wages from Matched Employer-Employee
Data. Journal of Labor Economics, 21(4), 887-922.
BETTIO, F. and VERASHCHAGINA, A. (2009). Gender segregation in the labour market:
Root causes, implications and policy responses in the EU. Luxemburg: European
Commission Directorate-General for Employment, Social Affairs and Equal
Opportunities.
BETTS, J.R. (1996). What do Students Know About Wages? Evidence from a survey of
Undergraduates. Journal of Human Resources, 31(4), 27-56.
BLAU, F. D. (2012). Gender, Inequality, and Wages. Oxford: Oxford University Press.
____ , FERBER, M. A. and WINKLER, A. E. (2010). The Economics of Women, Men, and
Work. Upper Saddle River, NJ: Prentice-Hall.
____ and KAHN, L. M. (2007). The Gender Pay Gap. The Economists’ Voice, 4(4), article 5.
____ and KAHN, L. Gender Differences in Pay. (2000). Journal of Economic Perspectives,
14(4), 75-99.
BOSKIN, M. J. (1974). A Conditional Logit Model of Occupational Choice. Journal of
Political Economy, 82(2) part 1, 389-398.
BRUZE, G., SVARER, M. and WEISS, Y. (2015). The Dynamics of Marriage and Divorce.
Journal of Labor Economics, 33(1), 123-170.
CAMPBELL, A. (2003). A Mind of Her Own: The evolutionary psychology of women. New
York: Oxford University Press.
CHARTRAND, J. M., DOHM, T. E., DAWIS, R. V. and LOFQUIST, L. H. (1987).
Estimating Occupational Prestige. Journal of Vocational Behavior, 31, 14-25.
COBB-CLARK, D. A. and TAN, M. (2011). Noncognitive skills, occupational attachment,
and relative wages. Labour Economics, 18, 1-13.
43
CORNEO, G. and JEANNE, O. (2010). Symbolic values, occupational choice, and economic
development. European Economic Review, 54, 237-251.
CROSON, R. and GNEEZY, U. (2009). Gender Differences in Preferences. Journal of
Economic Literature, 47(2), 1-27.
DOLTON, P. J., MAKEPIECE, G. H. and VAN DER KLAAUW, W. (1989). Occupational
Choice and Earnings Determination: The role of sample selection and non-pecuniary
factors. Oxford Economic Papers New Series, 41(3), 537-595.
DRYLER, H. (1998). Parental Role Models, Gender, and Educational Choice. The British
Journal of Sociology, 49(3), 375-398.
ECCLES, J. S. (1994). Understanding Women’s Educational and Occupational Choices:
Applying the Eccles et al. model of achievement-related choices. Psychology of Women
Quarterly, 18, 585-609.
ELIOT, L. (2009). Pink Brain, Blue Brain: How small differences grow into troublesome
gaps - and what we can do about it. Boston, New York: Mariner Books, Houghton
Mifflin Harcourt.
ENGLAND, P. (1979). Women and Occupational Prestige: A case of vacuous sex equality.
Signs, 5(2), 252-265.
____ (1982). The Failure of Human Capital Theory to Explain Occupational Sex
Segregation. Journal of Human Capital, 17(3), 358-370.
FASSINGER, R. E. (1990). Causal Models of Career Choice in Two Samples of College
Women. Journal of Vocational Behavior, 36, 225-248.
FERSHTMAN, C. and WEISS, Y. (1993). Social Status, Culture, and Economics
Performance. Economic Journal, 103 (July), 946-959.
FERSHTMAN, C. and WEISS, Y. (1998). Social Rewards, Externalities, and Stable
Preferences. Journal of Public Economics, 70, 53-73.
FINE, C. (2010). Delusions of Gender: How our minds, society, and neurosexism create
difference. New York, London: W. W. Norton & Company.
FISCHHOFF, B., PARKER, A. M., BRUINE DE BRUIN, W., DOWNS, J., PALMGREN,
C., DAWES, R. and MANSKI, C. F. (2000). Teen Expectations of Significant Life
Events. Public Opinion Quarterly, 64, 189-205.
FORTIN, N. M. (2008). The Gender Wage Gap among Young Adults in the United States:
The importance of money versus people. Journal of Human Resources, 43, 885-918.
GANZEBOOM, H. B. G., DE GRAAF, P. M., and TREIMAN, D. J. (1992). A standard
international socio-economic index of occupational status. Social Science Research, 21,
1 -56.
GOLDIN, C. (2006). The Quiet Revolution that Transformed Women’s Employment,
Education, and Family (Richard T. Ely Lecture), American Economic Review Papers
and Proceedings, 96(2), 1-21.
____ , KATZ, L. F. and KUZIEMKO, I. (2006). The Homecoming of American College
Women: The Reversal of the College Gender Gap. Journal of Economic Perspectives,
20(4), 133-156.
GROVE, W. A., HUSSEY, A. and JETTER, M. (2011). The Gender Pay Gap Beyond
Human Capital: Heterogeneity in Noncognitive Skills and in Labor Market Tastes.
Journal of Human Resources, 46(4), 827-874.
44
HAMERMESH, DS. (1999). Changing Inequality in Markets for Workplace Amenities.
Quarterly Journal of Economics, CXIV(4), 1085-1123.
HELLERSTEIN, J., NEUMARK, D. and MCINERNEY, M. (2008). Changes in Workplace
Segregation in the United States: Evidence from the 1990 and 2000 Decennial
Employer-Employee Datasets. In S. Bender, J. Lane, K. Shaw, F. Andersson and T. von
Waechter, T. (eds), The Analysis of Firms and Employees: Quantitative and Qualitative
Approaches. Chicago: University of Chicago Press, pp. 163-195.
HUMLUM, M. K., KLEINJANS, K. J. and NIELSEN, H. S. (2012). An Economic Analysis
of Identity and Career Choice. Economic Inquiry, 50(1), 39-61.
JACOBS, J. E., CHHIN, C. S. and BLEEKER, M. M. (2006). Enduring links: Parents’
expectations and their young adult children's gender-typed occupational choices.
Educational Research and Evaluation, 12(4), 395–407.
JENSEN, T. P. and ANDERSEN, D. (2006). Participants in PISA 2000 – Four Years Later.
In j. Mejding and A. Roe (eds), Northern Lights on PISA 2003 - A Reflection from the
Nordic Countries. Denmark: Nordic Council of Ministers.
KLEINJANS, K. J. (2010). Family Background and Gender Differences in Educational
Expectations, Economics Letters, 107, 125-127.
____ (2009). Do Gender Differences in Preferences for Competition Matter for Occupational
Expectations? Journal of Economic Psychology, 30(5), 701-710.
LAMBERT, P. A. (2008). The Comparative Political Economy of Parental Leave and Child
Care: Evidence from Twenty OECD Countries. Social Politics, 15(3), 315-344.
LARSEN, M. and HOULBERG, H. S. B. (2013). Lønforskelle Mellem Mænd og Kvinder
2007-2010. Copenhagen, Denmark: SFI-Det Nationale Forskningscenter for Velfærd
13:24.
LØNKOMMISSIONEN (2010). Lønkommissionens Redegørelse: Løn, Køn, Uddannelse og
Fleksibilitet. Copenhagen, Denmark.
MAGNUSSON, C. (2009). Gender, Occupational Prestige, and Wages: A Test of
Devaluation Theory. European Sociological Review, 25(1), 87-101.
MARINI, M. M., FAN, P. L., FINLEY, E. and BEUTEL, A. M. (2006). Gender and Job
Values. Sociology of Education, 69(1), 49-65.
NIEDERLE, M. and VESTERLUND, L. (2007). Do women shy away from competition? Do
men compete too much? Quarterly Journal of Economics, 122(3), 1067-1101.
OREOPOULOS, P. and SALVANES, K. (2011). Priceless: The Nonpecuniary Benefits of
Schooling. Journal of Economic Perspectives, 25(1), 159–184.
POLACHEK, S. W. (1981). Occupational Self-Selection: A Human Capital Approach to Sex
Differences in Occupational Structure. The Review of Economics and Statistics, 63(1),
60-69.
POWELL, M. and ANSIC, D. (1997). Gender differences in risk behavior in financial
decision-making: An experimental analysis. Journal of Economic Psychology, 18, 605828.
ROSEN, S. (1986). The Theory of Equalizing Differences. In O. Ashenfelter and R. Layard
(eds), Handbook of Labor Economics 1, Elsevier Science Publishers, pp. 641-692.
ROTHSTEIN, J. and ROUSE, C. E. (2011). Constrained After College: Student Loans and
Early Career Occupational Choices. Journal of Public Economics, 95(1), 149-163.
45
SHAUMAN, K. A. (2006). Occupational Sex Segregation and the Earnings of Occupations:
What causes the link among college-educated workers? Social Science Research, 35,
577-619.
SCHOON, I. (2001). Teenage Job Aspirations and Career Attainment in Adulthood: A 17year follow-up study of teenagers who aspired to become scientists, health professionals,
or engineers. International Journal of Behavioral Development, 25(2), 124-132.
TREIMAN, D. J. (1977). Occupational Prestige in Comparative Perspective. New York:
Academic Publishers.
TURNER, S. E. and BOWEN, W. G. (1999). Choice of Major: The Changing (Unchanging)
Gender Gap. Industrial and Labor Relations Review, 52(2), 289-313.
VALIAN, V. (1998). Why so slow? The advancement of women. Cambridge, Massachusetts,
London, England: The MIT Press.
VELLA, D. (1994). Gender roles and human capital investment: the relationship between
traditional attitudes and female labour market performance. Economica, 61(242), 191211.
WARREN, J. R., SHERIDAN, J. T. and HAUSER, R. M. (1998). Choosing a Measure of
Occupational Standing: How Useful Are Composite Measures in Analyses of Gender
Equality in Occupational Attainment? Sociological Methods & Research, 27(1), 3-76.
WISWALL, M. and ZAFAR, B. (2011). Determinants of College Major Choice:
Identification Using an Information Experiment. Federal Reserve Bank of New York
Staff Reports; no. 500.
ZAFAR, B. (2013). College Major Choice and the Gender Gap. Journal of Human
Resources, 48(3), 545-595.
ZHAN, Y. (2012). Money vs. Prestige: Cultural Attitudes and Occupational Choices.
Mimeo, University of California, San Diego.
46