1Progress towards gender equality is held back by numerous obstacles, both in the private sphere, with the unequal division of domestic tasks, and in the occupational sphere, with the disparities between men’s and women’s wages and the glass ceiling over women’s careers. There are multiple reasons – linked to women’s education, working hours, employment sector and family constraints – why women tend to hold jobs that are less qualified, less valued and less well paid than those of men. Yet even when men and women share the same characteristics, a wage difference of around 12% is still observed. Based on this observation, Isabelle Bensidoun and Danièle Trancart examine the various components of the gender wage gap, focusing especially on work preferences and attitudes. Applying a wage gap decomposition model to data from the “Génération 1998 à 10 ans” survey, they show that women’s lower wages are explained in part by differences in these preferences and attitudes.
2Women’s status has changed substantially since the time when Schopenhauer laughed at the very idea of women holding a position of power.  Women are now more educated than men, and go out to work not only, as in the past, to provide a second source of income for their family, but also, in many cases, to achieve personal fulfilment and a satisfying career. Despite their education and labour market investment, women’s wages are still lower, on average, than those of men. There are several explanations for this: women more often work part-time; they are also more qualified, but the subjects they study prepare them for careers in less well paid lines of work. Yet even after taking these factors into account, a non-negligible share of the gender wage gap remains unexplained. This means either that women face discrimination on the labour market, or that other less obvious or less measurable factors are at play. For example, gender differences in the priority given to work, in personality, values or attitudes may play a role. In the final summary on gender questions in the Handbook of Labor Economics, Bertrand (2010) suggested exploring this angle, after describing the results of laboratory experiments that revealed gender differences in negotiating skills and in attitudes to risk and competition.  Psychological research has also identified gender disparities in personality traits and preferences. In her theory of preferences, Hakim (2004) highlights the importance of values and attitudes in employment decisions and career choices, but also in pay levels, when personal goals and preferences are involved rather than general moral stances or opinions. Hence, while women’s place on the labour market, and in society more generally, has been transformed since Schopenhauer’s time, the social norms which shape our preferences still bear the stigmata of long-standing past beliefs. The purpose of this article is to assess the influence of these gender differences in preferences and attitudes on the French labour market, and on the gender pay gap in particular. It follows on from work by Filer (1983), Mueller and Plug (2006), Fortin (2008), Grove et al. (2011), Cobb-Clark and Tan (2011), and Nyhus and Pons (2012) on the role of these factors qualified as “non-cognitive” in the international economic literature.
3Most studies, excepting that of Cobb-Clark and Tan (2011), consider the direct effect of non-cognitive variables (preferences and personality traits) on wage gaps – i.e. their effect on individual productivity – and measure the contribution of these variables using a traditional decomposition method (Blinder, 1973; Oaxaca, 1973). However, these variables may also determine individuals’ career choices  and employers’ recruitment decisions (Chantreuil and Epiphane, 2013), thus explaining, in part, the occupational segregation between men and women observed on the labour market. Indeed, this is the conclusion drawn by Filer (1986), Ham et al. (2009), Falter and Wendelspiess Chávez Juárez (2012), John and Thomsen (2012): alongside more traditional explanatory variables (education, work experience), non-cognitive aspects are factors of heterogeneity between individuals which influence occupational choices, notably via their effect on preferences. To take account of this indirect mechanism whereby preferences and attitudes may influence wages, but also of the potentially discriminatory nature of occupational segregation, wage gaps are decomposed using the method proposed by Brown, Moon and Zoloth (1980). The wage gap is thus decomposed into an inter-occupational component (linked to the differences between male and female distributions across different occupations) and an intra-occupational component (linked to wage differences within occupations), each component being split into an explained gap and an unexplained gap by the gender differences in characteristics.
4By using this decomposition method, we apply an approach similar to that of Cobb-Clark and Tan (2011). But as well as a different country of observation and different non-traditional variables, our study adopts a different method to capture the influence of non-cognitive factors on wage gaps. While Cobb-Clark and Tan (2011) assess the contribution of preferences and attitudes to the wage gap by comparing estimates with and without non-cognitive variables, we propose to make a detailed decomposition. While certain technical precautions are necessary, as we shall see below (Section II), this method provides an accurate measure of the share of the wage gap attributable to these factors.
5Ours is the first study to explore how gender differences in preferences and attitudes are liable to influence the gender wage gap in France. The CEREQ  survey used here, “Génération 1998 à 10 ans” (The 1998 cohort, 10 years on), includes subjective questions that provide insights into the potential effect of the level of priority given to one’s career, attitudes to risk and optimism about future career prospects on wage differences between young men and young women.
I – Literature review
6Over the last 10 to 15 years, a growing body of research has explored the effects of non-traditional factors on labour market behaviours. After considering the potential impact of education, experience, cognitive skills (which mobilize memory, language, reasoning or problem-solving) on employment decisions or individual pay levels, attention is now turning to the role of non-cognitive capacities, notably personality traits, but also social preferences or norms. Psychologists and sociologists have long understood the key importance of these factors in decision-making, and they now form part of economists’ standard “toolbox”. For sociologists, there is nothing new in the idea that the gendered roles socially assigned to men and women shape their preferences and personality traits, and that their occupational choices and career aspirations are influenced accordingly. In economics, research on the contribution of non-cognitive variables to the gender wage gap is more recent. We will examine the results obtained by Filer (1983), Mueller and Plug (2006), Fortin (2008), Grove et al. (2011), Cobb-Clark and Tan (2011) and Nyhus and Pons (2012). Their studies cover different populations (samples), use different non-cognitive variables, and apply different decomposition methods to the results obtained (Table 1).
Literature review of the influence of non-cognitive variables on the gender wage gap
Literature review of the influence of non-cognitive variables on the gender wage gap
7In most of the listed studied (four out of six), the unexplained component of the wage gap is large, representing between 63% (Nyhus et Pons, 2012) and more than three-quarters of the total gap (Fortin, 2008; Cobb-Clark and Tan, 2011). The contribution stemming from gender differences in non-cognitive variables is very low and negative in Cobb-Clark and Tan’s study, and positive, at around 4%, in Filer’s study, 7.3% in Mueller and Plug, 8.4% in Fortin, 11.5% in Nyhus and Pons and as much as 17.4% in Grove et al. Only two studies – those of Fortin and of Grove et al. – give the significance levels of the various components of the explained wage gap. In the first, the contribution of the non-cognitive variables is lowered from 8.4% to 7.4% as a consequence, and in the second from 17.4% to 8.2%. All in all, the range extends “at best” from a small negative quantity to 8.2%.
8The differences in market returns to these variables (“o/w NCV” line in the unexplained total of Table 1), account for 13% of the wage gap in Fortin and 10% in Filer. Their contribution is very low (0.4%) in Nyhus and Pons, and negative (–4.5%) in Mueller and Plug. However, the various authors do not estimate the significance of this component.
II – Method
9After briefly describing the selected wage gap decomposition, we will present our methodological improvements with respect to existing studies, notably that of Cobb-Clark and Tan (2011).
10The wage gap decomposition proposed by Brown, Moon and Zoloth (1980) has several advantages. It allows us to consider that the gender-based occupational segregation observed on the labour market is the result of individual preferences, but also of discriminatory behaviours (unlike the more widely used Oaxaca-Blinder decomposition method). It also allows us to determine the way in which preferences and attitudes can directly influence wage gaps, i.e. their effect on individual productivity, but also their impact on individuals’ choices of occupational category (OC) and employers’ recruitment decisions, and hence their effect on occupational segregation. This decomposition is expressed as follows:
12Where and are the mean of the log of men’s and women’s wages, and is the mean of the log of men’s wages in the OCj.
13The first component of the equation represents the intra-OC wage gap which is explained by average differences in male characteristics , and female characteristics , (the fact that they do not have the same educational level, work experience, working hours or preference, for example), while the second measures the unexplained part, stemming from differences in the returns to wages of these characteristics for women , and men , in other words, the gender differences in the contributions of the chosen characteristics to wage levels.
14In this wage gap decomposition, we see that the “justified” (explained) wage gaps are those stemming from gender differences in productivity; for example, men’s higher pay levels are justified by their greater average work experience. The unexplained differences, for their part, are linked to differences in returns to these characteristics – the fact that men with a given qualification are paid more than women for example, a situation that is totally unjustified.
15Likewise, the inter-OC wage gap is decomposed into two components, the first of which represents the explained part, i.e. the difference between the observed distribution of men by OC, pmj, and the counterfactual distribution of women , i.e. that which would exist if women with equivalent characteristics had the same access as men to the different occupational categories. The second component measures the difference between this counterfactual distribution of women and their observed distribution pwj, thereby capturing the unexplained part of the inter-OC wage differences, those attributable to the differential access of men and women to the various occupational categories.
16To make this decomposition, the returns of men’s and women’s characteristics must be estimated, along with the counterfactual distribution of women in the various OCs.
17The equations of wages by OC for men and women have the following standard form:
19It is assumed that the choice of OCs is determined by the interaction of supply factors (individual skills and preferences for an occupation compatible with family constraints) and demand factors (employers’ decisions to hire an individual based in his/her productive characteristics). These interactions are summarized in reduced form as follows:
21Where pij represents the probability that individual i is employed in the OC j determined by the variables Xo and the estimated coefficients γj.
22These OC choices are modelled by a multinomial logit for men to evaluate the counterfactual situation for women in terms of job distribution .
1 – Detailed decomposition
23Previous studies that applied this decomposition (Chamkhi and Toutlemonde, 2015; Cobb-Clark and Tan, 2011; Meng and Meurs, 2001; Reilly, 1991) simply measured the four overall components, i.e. the explained and unexplained components of wage gaps between and within OCs. Yet measures of the characteristics (explained parts) or their returns (unexplained parts) that contribute to these various components of the gender wage gap provide key information not only to guide policy-makers – they identify the factors to be acted upon in order to reduce wage gaps – but also to assess the relative contribution of attitudes and preferences to this gap.
24There are several reasons why these detailed decompositions were not made. First, overall BMZ decompositions, distinguishing between wage gap components within OCs and between OCs, already enable us to determine whether wage inequalities are due to unequal pay for equal work, or to unequal work despite equal qualifications.
25Second, when the factors explaining the gender wage gap include qualitative variables, the results of estimations made using a reference group for these variables cannot be used directly to detail the unexplained parts, given that these parts are dependant on the reference groups used in the estimations (Oaxaca and Ransom, 1999). Hence, to obtain decompositions that are invariant to the choice of reference groups, Yun (2005) suggested transforming the estimated coefficients by expressing them as a difference with respect to the average and adding the coefficient of the reference group (Bensidoun and Trancart, 2015). This is the approach applied here.
26Another difficulty arises, linked to the use of a non-linear model to estimate the OC choices. To circumvent this problem and obtain a detailed decomposition of inter-OC wage gaps, and hence be able to assess the influence of preferences and attitudes on overall gender wage gaps, a linear model was used to estimate OC choices. 
27In this case, the equation
29is replaced by
III – Description of data
31The “Génération” survey used here was conducted by CÉREQ and includes subjective questions to determine career preferences, attitudes to risk and perceptions of future career prospects. Its aim was to analyse the first years of working life of a cohort of young people leaving the education system at the same time, whatever their age, educational level or skillset. The Génération 1998 survey concerns young people who left the education system in 1998, and who were interviewed in 2001 2003, 2005 and 2008. The survey weightings are always fitted to the 1998 cohort of school leavers.
32The “Génération 1998 à dix ans” survey (Generation 1998, ten years on), conducted in 2008, is used here for all the variables except the preferences and attitudes variables which, as will be explained below, are based on the first Génération 1998 interview in 2001. The analysis concerns individuals in employment in 2008, excluding the self-employed,  who answered the question on working hours  (full time versus part-time). The sample comprises 9,422 individuals, of which 4,625 men and 4,797 women. As actual working hours are not available in the survey, monthly wages (including bonuses and 13th month, if applicable) are modelled and information on working hours by job category is used as a control variable.
33The two modelled variables, employment and wages, are shown in Table 2 by OC, while the independent variables are presented at overall level (Table 3) to show the average differences observed between men and women.
1 – Individual, familial and occupational characteristics
34Table 2 shows that men and women are distributed differently across the ten selected occupational categories:  women are significantly overrepresented in intermediate occupations in the social and health sectors, and among sales and clerical workers, while men are over-represented among engineers, supervisors, and above all among manual workers. The differences are especially pronounced for the clerical/sales worker and manual worker categories. This is consistent with the results of Brinbaum and Trancart (2015) and Meron et al. (2006) who found substantial gender segregation in employment when educational levels are low. However, beyond educational level itself, it is the specific skillset (Table 3) that doubtless contributes to this gender segregation between clerical and manual workers: women tend to train for jobs in the service sector (65%) and men for jobs in the industrial sector (67%). 
Distribution of jobs and wages by occupational category
Distribution of jobs and wages by occupational categoryThe abbreviations and codes used in the INSEE classification are given in parentheses.
Statistical significance: * p < 0.10; ** p < 0.05; *** p < 0.01.
35Ten years after completing their education, the mean monthly wage of young men (€1,963 in 2008) is 27.6% higher (0.24 log points) than that of young women (€1,538 in 2008, Table 2).  At the bottom of the wage hierarchy (manual and clerical workers), the gender gap is very wide, at between 25% and 45%, while at the other extreme, only administrative and commercial professionals have a gap of similar magnitude (25%). The wage gaps are significant and lower for engineers or technical professionals (18.6%), intermediate occupations (7-16%) and civil service professionals (9%)
Individual, family and employment characteristics in 2008 and preferences in 2001(a)
Individual, family and employment characteristics in 2008 and preferences in 2001(a)(a) The “qualifications” and “family characteristics” variables determine the choices of OC and wages. For wages, the model includes experience, occupational characteristics and working hours, and for choice of OC, age is added.
(b) Non-response is higher for women as it mainly corresponds to public-sector jobs where the proportion of women is higher. In farming, non-response accounts for 90% of the total.
Statistical significance: * p < 0.10; ** p < 0.05; *** p < 0.01.
36The characteristics used to explain these differences in the distribution of jobs and wages by occupational category of the individuals in our sample are summarized in Table 3. They show that the women, on average, are slightly older than the men (6 months) but that their experience  on the labour market is shorter (a difference of slightly less than 4 months). Experience measured here by means of a month-by-month description in a work diary of all positions occupied from the date of leaving the education system up to the survey date reflects the actual experience of individuals and not a potential experience, as is often the case. While women’s experience is shorter than men’s, their educational level is higher: 40% of men have a qualification no higher than a lower secondary certificate, versus just 26% of women, while almost a quarter of women have at least three years of post-secondary education, versus 19% of men. Moreover, one quarter of men had already retaken a school year by the time they reached secondary school, versus less than 18% of women.
37With a shorter working history than men, women differ from their male counterparts mainly in terms of mean working hours. While 97% of men work full time, among women the proportion is just 72% (18% work four-fifths time). Differences in experience and weekly presence on the labour market may be linked to differences in family characteristics. For example, ten years after leaving education, women are more often in a union than men and, above all, a larger number are already parents: almost 2/3 already have at least one child versus less than half of the men.
38In terms of occupational characteristics, gender segregation is similar to that described above, with men being more present in industrial sectors and women in services, above all in the administrative, education, health and social domains (almost 50% of women, of which two-thirds in the public sector). Not surprisingly, women more often work in the public sector (37% versus 20%) and are slightly less inclined to work non-standard hours. They slightly less often have an open-ended contract and, above all, less frequently have supervisory responsibilities (one woman in five versus more than one-third of men). There are also fewer women in large companies.
39The set of characteristics presented here concerns young people, aged 31.5 years on average, i.e. 10 years younger than employees in France as a whole (INSEE, Labour Force survey 2008). For this reason, educational levels are higher for both sexes: only 40% of young men and 26% of young women do not have an upper secondary qualification (baccalauréat), compared with 55% of men and 42% of women in general. The young men’s family situation is also atypical: only two-thirds are in a union and just 23% have a dependent child, versus 72% and 36% of male wage employees in general. The young women’s family situation, on the other hand, is similar to that of female wage employees overall. Doubtless for this reason, women’s working hours are very similar in both samples: 27.5% of the young women work part-time versus 32% of female wage earners in general. The gender wage gaps are comparable in both populations, with the young women earning 21.6% less than the young men, compared with a gap of 24.3% for the entire population of wage employees.  Ten years after leaving the education system, the wage differences between men and women are already well entrenched.
2 – Work preferences and attitudes
40The database used here gives an initial insight into the possible role of preferences and attitudes – the priority given by individuals to their career, their appetite for risk and their optimism about future career prospects – in employment decisions and wages. To limit the risks of endogeneity, i.e. that these variables may be influenced by individuals’ situations on the labour market (notably their wage), the responses given in 2001 were used although the analysis was conducted for 2008.
41Respondents were asked a first question about their career priorities: “Is your priority over the next three years mainly to: 1) find a stable job; 2) get ahead in your career; 3) ensure a good work-life balance?”. A dichotomous variable based on the response “get ahead with your career” was constructed. This preference expressed by individuals – probably influenced by gender stereotypes or social norms, with women more often feeling a “duty” to invest in the family sphere and men in the work sphere – may lead to certain career choices over others, and to higher wages, by encouraging those who invest in the work sphere to press harder for a job change or wage increase in order to satisfy their ambition (Fortin, 2008; Grove et al., 2011).
42A second question on perceived future career prospects was also used. Based on the answers to the question “How do you feel about your career prospects 1) quite worried; 2) quite optimistic; 3) don’t know”, a dichotomous variable was constructed which sets the “quite optimistic” response against the two others. Several studies have shown that employment insecurity affects wage levels as the individuals concerned tend to moderate their wage demands or limit external mobility that might lead to a better paid job at the start of their career (Aaronson et Sullivan, 1998; Campbell et al., 2007; Hakim, 2004; Simonnet, 1996).
43Last, the third question: “Do you plan to set up your own business one day 1) Yes, I do; 2) Yes, maybe; 3) No; 4) Don’t know”, was used to construct a dichotomous variable with, on one side, those who replied “Yes I do” and “Yes, maybe”, and on the other, those who chose the two other options. The variable constructed in this way is considered as a marker of a positive attitude to risk, since individuals who report plans to set up their own business have a lower level of risk aversion than the others. Being self-employed involves risk – not only financial but also personal – and social insurance is less generous. Numerous studies have shown that the least risk-averse individuals are more likely to become self-employed. Cramer et al. (2002) for the Netherlands, Ekelund et al. (2005) for Finland, Brown et al. (2011) and Ahn (2010) for the United States show that appetite for risk is a key determinant of self-employment. For France, based on experimental studies, Colombier et al. (2008) and Masclet et al. (2009) also showed that self-employed workers are significantly less risk-averse. Individual attitudes to risk may influence career choices, with the most risk-averse opting for occupations where earnings variance is low (Bonin et al., 2007), or in the public rather than private sector (Jung, 2013). Risk aversion may also result in lower pay due to the earnings differential associated with lower risk-taking (Bertrand, 2010).
44In 2001, three years after leaving education, the distributions of these three variables show significant differences between men and women, in line with other study findings. Women are, on average, significantly less optimistic, less frequently express career ambitions, and are more risk-averse than men (Table 3). These gender differences persist even after controlling for differences in educational level and skillset.
45The variables of preferences and attitudes were measured seven years before the date concerned by our analysis, but after the respondents’ labour market entry (i.e. after 1998), so they may reflect individuals’ labour market situation and not their “true” preferences. This means that the results of our analysis in terms of gender wage gaps may be contaminated by the labour market situation if this factor influences men and women differently. To test for this potential bias, we used logistic models to estimate the influence of the number of months of unemployment between 1998 and 2001 on work preferences and attitudes (controlling by educational level) and tested whether the impact was different for men and women.
46Table 4, which gives the marginal effects of unemployment, sex and the interaction between these two variables, shows that optimism and giving priority to one’s career are affected by unemployment: all other things being equal, the longer individuals remains unemployed, the less optimistic they are about their career prospects and the less priority they give to their working career. Moreover, and in line with the results of the descriptive statistics, women are less frequently optimistic, less frequently express the desire to have a career, and are more risk-averse than men. But above all, we observe that the interaction of the gender effect and of months spent unemployed is not significant for any of the variables studied: while unemployment does indeed affect the respondents’ responses about their career prospects or the desire to give priority to their career, there is no difference between men and women in this respect
Preferences and attitudes to work and unemployment
Preferences and attitudes to work and unemploymentLogit model, standard deviations in brackets.
Note: Marginal effects of educational level not given here.
Statistical significance: * p < 0.10; ** p < 0.05; *** p < 0.01.
IV – Results and discussion
47The decomposition of wage gaps used here enables us to identify the share of these gaps that is linked to differences in male and female characteristics (explained gap), and the unexplained share. It also allows us to determine the share stemming from work segregation, i.e. the fact that men and women are not equally distributed across occupational categories (inter-OC wage gap). The results are first presented at overall level, then at detailed level, to identify the factors behind the overall wage differences. The section will conclude with a discussion of the reasons why our results on the gender wage gap in France differ from those published elsewhere.
1 – Overall wage gap decomposition
How much can be explained?
48Table 5 gives the results of the BMZ  decomposition for the estimates without (column 1) and with (column 2) variables of preferences and attitudes, and for those of the Oaxaca-Blinder decomposition  (column 3).
Decomposition of the gender wage gap
Decomposition of the gender wage gapThe differences in log wages were multiplied by 100 to make the table more legible.
Statistical significance: * p < 0.10; ** p < 0.05; *** p < 0.01 based on 200 bootstrap sample replications, except for the Oaxaca-Blinder column.
49Comparison of the second and third columns shows that, as expected, the explicit inclusion of labour market gender segregation in the decomposition of the gender wage gap (column 2) reduces the explained component. While the explained part accounts for 60% of the wage gap when the gender distribution across occupational categories is considered as stemming solely from personal choice (column 3), it accounts for no more than 40% when the decomposition assumes that these distributions may also reflect discriminatory behaviour on the part of employers. Introducing preferences and attitudes, on the other hand, reduces the unexplained component, which falls from 70% without preferences and attitudes (column 1) to 62% (column 2). The unexplained intra-OC gap is reduced, but also the unexplained inter-OC gap, thereby justifying the use of a decomposition that takes into account the indirect effects of preferences and attitudes on careers.
What is the role of occupational segregation?
50Looking now at the shares of the wage gap attributable to intra-OC and inter-OC differences (column 2), we see that almost 80% of the wage gap is due to wage differences between men and women within the different occupational categories, with 46% stemming from differences in characteristics and 33% remaining unexplained. The wage gaps attributable to gender segregation on the labour market – the fact that men and women are not equally distributed across OCs – account for just 20%,  although this figure is the sum of a negative explained component (–8.2%) and an unexplained component of almost 30%. This first negative component signifies that if women had the same opportunities as men to work in the various occupational categories, their characteristics would enable them, on average, to obtain jobs with higher wages than men. Figure 1 illustrates this situation. It shows three distributions: that of men, that of women and the counterfactual distribution that would be observed if women had the same opportunities to work in the various OCs as men. The OCs are ranked by decreasing level of mean male wages.
51We see that many OCs should account for a larger share of women’s employment than that of men.  Only the OCs of engineer and, above all, manual worker are “legitimately” masculine, in the sense that even if women had equal access to the various OCs, these two categories would still represent a larger share of male jobs than female jobs due to the characteristics which more frequently orient men towards them. Yet, given that the OCs where they are legitimately more strongly represented (among manual workers) are also those where mean wages are low, the gender wage gap should favour women, given their characteristics, if they have the same level of access to the various OCs as men. This graph also shows that the counterfactual distribution of women is very different to the actual one. There should be higher proportions of women who are administrative and commercial professionals, engineers, civil service professionals, manual workers or associate professionals in fields other than health, social services and the civil service. Conversely, the proportions who are skilled or unskilled clerical workers or associate professionals in health, social services and the civil service should be much smaller. The wage gaps linked to occupational segregation remain unexplained here. They may be linked to determinants other than those used here, or to discrimination against women in access to various occupations.
Observed distribution of women’s and men’s jobs by occupational category and counterfactual distribution of women (%)
Observed distribution of women’s and men’s jobs by occupational category and counterfactual distribution of women (%)Note: CAC: Administrative and commercial professional; CFP: Civil service professional; Other PI: Administrative and commercial associate professional, technician and supervisor, etc.; PI_SS: Health and social work associate professional; PI_FP: Civil service associate professional.
Statistical significance: * p < 0.10; ** p < 0.05; *** p < 0.01.
2 – Detailed decompositions: the factors behind gender wage gaps
52The detailed decompositions of wage gaps within and between occupational categories enable us to identify the characteristics that account for the overall gaps analysed so far.  In this respect, the first columns of Table 6 show that more than one-third of the total wage gap is explained by higher levels of part-time working among women,  and 13.5% by gender differences in occupational characteristics. For example, the fact that more men than women occupy managerial positions justifies their higher pay levels.  Their greater professional experience also explains 3.5% of the wage gap in favour of men. Conversely, women’s level of education and the fact the more women than men live with a partner and have children at these ages plays in the opposite direction.
53The decomposition by characteristic of the unexplained intra-OC wage gap (columns 3 and 4) shows first that most of this gap is linked to differences in the constants estimated for men and women, and second, that the returns of men’s and women’s characteristics are not significantly statistically different, apart from the return of their family characteristics, which is unfavourable to women. This tallies with the conclusions of Filer (1983) and of Nyhus and Pons (2012) whereby most of the returns of characteristics explaining wages are similar for both sexes.
Detailed decomposition of the gender wage gap(a)
Detailed decomposition of the gender wage gap(a)(a) The differences in log wages were multiplied by 100 to make the table more legible.
(b) The detailed decomposition on the inter-OC wage gap is based on a linear model of choices of OC, whereas the overall decomposition (Table 5) is based on a non-linear model (multinomial logit). This change leads to relatively modest differences in the respective contributions of the explained and unexplained components:–2.0 / 7.1 (for the non-linear model) versus –2.5 / 7.6 (for the linear model).
Statistical significance: * p < 0.10; ** p < 0.05; *** p < 0.01 based on 200 bootstrap sample replications.
54Columns 5 and 6 show that it is women’s level of education – notably the fact that a much smaller proportion of women have no qualifications or a lower-secondary industrial qualification – that would justify their presence in occupations with higher wages than those where they are actually employed (negative contribution). For the unexplained component of the inter-OC wage gap (columns 7 and 8), as for the intra-OC wage gap, most returns are not statistically different for men and women, so most of the gap stems from differences between the estimated constants.
55Hence, most of what is sometimes qualified as discrimination is not linked to the fact that the factors explaining wages or choice of OC have higher returns for men (excepting family characteristics, which penalize women), but to gender differences in treatment that extend beyond these factors.
56Regarding the influence of preferences and attitudes (Table 6, shaded lines), we observe that optimism and giving priority to one’s career explain 3.3% of the overall wage gap – almost as much as experience. The difference in attitudes to risk, on the other hand, has no influence on the wage gap.
57Overall, preferences and attitudes explain more than 6% of the observed gender wage gap, more than the traditional variables of human capital, experience and educational level, especially since women’s educational advantage should also give them a wage advantage. A similar proportion is found in other studies. In those examined in this article (Section I), non-cognitive factors explain 8.2% of the wage gap at most. Taking account of non-traditional factors linked to individual preferences and attitudes thus makes it possible to reduce the unexplained component by rendering observable what is usually counted as unobservable. Within the unexplained component, only the factor linked to preference for a career gives men a significant advantage over women by facilitating access to well-paid OCs. This unexplained component remains strong however, representing more than 60% of the observed gender wage gap.
3 – Comparison with other studies of France
58Recent analyses (Bozio et al., 2014; Meurs and Ponthieux, 2006) on the origin of the gender wage gap in France produce results that differ from those presented here for several reasons.
59The first concerns the decomposition method used. These two studies are based on an Oaxaca-Blinder decomposition, and therefore consider that the choice of OC is exogenous; in other words, they do not consider the potentially discriminatory nature of occupational segregation. Consequently, the explained part of the wage differences – 76.2% in 2002 for Meurs and Ponthieux (2006) and 71.6% in 2012 for Bozio et al. (2014) – is much higher than the proportion found by us (36.1%).  As described earlier, if this decomposition method was applied to our data it would likewise result in a much higher explained component (60.3%).
60The second reason is linked to the populations concerned. While our study is based on young people who left the education system ten years earlier, in 1998, these two studies are based on all wage-earners. Our results therefore concern a younger population. In fact, if our results are compared with those of the two other two studies using the same decomposition method (Table 7), we observe that working hours and occupational characteristics are the two main sources of explained wage gaps in all cases, but that they are smaller in our own estimation.
61In terms of working hours, the higher contribution observed in these studies (48% and 44.3 % versus 36.1 % in ours) is linked to the fact that they take differences in weekly working hours into account in addition to differences in percentage of full-time working. Differences in occupational characteristics, for their part, represent between 30% (Bozio et al., 2014) and 34 % (Meurs and Ponthieux, 2006) of wage differences, versus 26% in our study, due to the age differences of the populations concerned. As the individuals in our sample are younger, the differences in the structure of employment are smaller than for the population in general. The more negative contribution of educational level in our study is, here too, linked to the differences in age of the study populations and reflects women’s strong investment in education. Differences in the contribution of experience doubtless reflect differences in the ways this variable is measured, i.e. real experience in our study versus potential experience in the two others.
Comparison of results for France: Oaxaca-Blinder decomposition of the monthly gender wage gap(a) (%)
Comparison of results for France: Oaxaca-Blinder decomposition of the monthly gender wage gap(a) (%)(a) The variables used in Bozio et al (2014) and Meurs and Ponthieux (2006) are identical but different from ours. Experience is real in our study but potential in the two others. Occupational characteristics, occupational categories (OC) and job characteristics overlap, with the exception of managerial functions and firm size which are present in our study only. Working hours include the percentage of full-time working in all three studies, and hours per week only in 2002 and 2012. Given that female labour force participation is high in our study (92%) because the women in the sample are relatively young, selection linked to female labour force participation was not used as the results are not significant.
62In sum, the differing results for the scale of the explained component of gender wage gaps are due essentially to the decomposition method used, differences in the way working hours are taken into account and differences in the populations concerned. In all three studies, however, the characteristics which contribute most to this explained gender wage gap are the same, namely gender differences in working hours and in occupational characteristics.
63Differences between men’s and women’s preferences and attitudes have been put forward as one of the reasons behind the stagnation of the gender wage gap over the last two decades. Based on data from the United States, Australia and the Netherlands, it has been shown that in most cases, the wage gap is partly explained by differences in personality traits or in preferences. In France, no surveys have been conducted to measure individual preferences or personality traits in as much detail as in other countries. The data used here nonetheless provide a number of pointers, and we hope that these new insights will attract sufficient interest to prompt the inclusion of questions on preferences and attitudes in future surveys.
64First insight: ten years after leaving the education system, the fact that women’s wages are 21.6% below those of men is only 20% attributable to their presence in different occupational categories, and 80% to the fact that within an identical OC, their wages are lower than those of their male counterparts.
65Second insight: while 40% of the wage difference can be explained by gender differences in characteristics, more than 60% remains unexplained. These differences in characteristics should result in female wages that are just 8% below those of men.
66Third insight: differences in preferences and attitudes – optimism, giving priority to one’s career and appetite for risk – matter (6.3% of the total wage gap, i.e. almost twice as much as experience), with as much (direct) influence on wages as on choice of OC. Indeed, regarding the choice of OC, differences in career priorities and attitudes exert a significant influence, in a direction which explains why men opt for better-paid OCs than women.
67While these characteristics reduce the unexplained component of the gender wage gap, it nonetheless remains large, especially since wage differences across OCs are generally unjustified, unlike differences within OCs, which are partly explained by gender differences in working hours, occupational characteristics or preferences and attitudes.
68Among the unexplained components of wage gaps, only the differences in returns of family characteristics are significant (to women’s disadvantage), which signifies that most of these unexplained differences are not explained by the specific variables identified in the analysis (the explanatory variables selected to determine wages and occupational choices). The unexplained differences are thus not linked to men making better use than women of their characteristics, i.e. their qualifications, their experience, their occupational characteristics or their work preferences and attitudes, but to unexplained reasons which result in men receiving higher wages.
69Once potential discrimination against women has been eliminated, these gender differences in preferences and attitudes, like the differences that explain the largest share of their wage gaps – time spent in the labour force or managerial responsibilities – reflect the differences probably stemming from the gendered roles attributed to each sex. In this context, Akerlof and Kranton (2000) have shown that the identity of individuals, their desire to comply with the social norms prevailing in their social group, may influence their economic decisions via the utility associated with them.  Hence, beyond questions of discrimination which, while important in our results, must always be viewed with caution in this type of exercise, our work points to the importance of taking measures to change mentalities. Government policies designed to deconstruct gender prejudice and to foster gender equality from an early age are key to progress in this area.
“The mere idea of seeing women sitting on the judges’ bench raises a smile.” The Basis of Morality, Chapter VI.
See Bertrand (2010) and Eswaran (2014) for a summary of this research.
The meaning of the word “choice” in this article does not rule out the notion of constraint; likewise for preferences.
Centre d’études et de recherches sur les qualifications (Centre for research on qualifications), Marseille.
See Bensidoun and Trancart (2015) for a discussion of the advantages and drawbacks of the linear probability model in this case.
The sample includes 443 artisans and traders of whom more than 80% are self-employed.
For this reason, 125 individuals were excluded.
A detailed list comprising 10 categories was established on the basis of the INSEE list of 24 categories, ensuring that sufficient numbers were included in each category (2% in the distributions by sex). The “unskilled clerical worker” category was constructed using the scale developed by Chardon (2001).
The study options in the last “class” attended were recoded in accordance with the specialities in INSEE’s NSF list. Codes 100 and 136 cover general specialities, 200 to 255 industrial specialities and 300 to 346 service specialities.
This corresponds, for young women, to a wage 21.6% lower than that of young men, i.e. slightly less than the difference observed for all wage-earners in France (24%).
i.e. the number of months in employment, including maternity or paternity leave but not parental leave.
Note that the difference between the two populations is greater if men’s wages are considered in relation to women’s wages: young men’s wages are 27.6% higher on average than those of young women, versus 32.1% for all wage-earners.
The results presented here consider that choice of OC and wage are independent, given that taking account of possible occupational selection linked to unobservable characteristics affecting both OC choices and wages gives non-significant results (Bensidoun and Trancart, 2015).
Selection linked to the fact that only wages of employed individuals are observed was tested for both sexes, taking parents’ social origin and mother’s occupation as exclusion variables. However, the inverse of the Mills ratio was not significant in either the men’s or the women’s wage equation. This is probably due to the fact that the women in our sample have a labour force participation rate that is high (92%), because they are relatively young, and similar to that of men (98%).
This low share of the inter-OC component is even more pronounced in the study by Cobb-Clark and Tan (2011) presented in Table 1, but also, in a similar proportion, in that of Meng and Meurs (2001) on the gender wage gap in France in 1992.
The OCs for which the female counterfactual bar is higher than the male bar in the figure.
The detailed decompositions were performed using the STATA Oaxaca program developed by Jann (2008) to each OC. The mean contributions of each variable were then obtained by weighting the different contributions per OC by the distribution of female jobs by OC for the intra-OC part and the distribution of male jobs by OC for the inter-OC part.
By considering working hours and managerial responsibility as elements that justify the wage gap, the model used here assumes that the observed situation is the result of individual choice. Bensidoun and Trancart (2015) consider that in 25% to 36% of cases, women in part-time work did not choose their working hours. Our decomposition therefore underestimates the unexplained part of the gender wage gap.
The glass ceiling on women’s careers suggests that the low proportion of women in managerial positions is not simply the result of individual choice, but also of discriminatory behaviour. If so, as is the case for involuntary part-time working, our decomposition underestimates the unexplained part of the gender wage gap.
Or the 35.6% obtained by Chamkhi and Toutlemonde (2015), resulting from the use of a decomposition that takes account of the partially discriminatory nature of occupational segregation.
Or the disutility that would result from behaviour that deviates from the norms of the group to which the individual belongs.