Tracing the gender wage gap: Income differences between ... · vestigating the gender wage gap, is used for the empirical analysis. The sample is ho-mogenous with respect to the measures

Tracing the gender wage gap:Income differences between male and femaleuniversity graduates in Germany

David Reimer and Jette Schröder*

The aim of this paper is to shed light on the causal mechanisms leading to the genderwage gap, drawing on neoclassical as well as sociological labor market theories. Aunique dataset from the 2001/2002 Mannheim University Social Sciences GraduateSurvey, which overcomes several limitations of standard population surveys when in-vestigating the gender wage gap, is used for the empirical analysis. The sample is ho-mogenous with respect to the measures normally used in income analyses Ð all of therespondents are university graduates, have a degree in the same field of study, and areobserved at career entry. Furthermore, the dataset includes detailed measures of hu-man capital, job search, and career attitudes, which are usually not included in standardpopulation surveys. The results of a sequence of nested regression models show thatnone of these measures reduces the gender wage gap substantially: on the contrary,the introduction of variables capturing human capital even leads to a small increase inthe gap. This indicates that the earnings differential between female and male gradu-ates in the study would be even larger if women had the same human capital endow-ment as men. Considering that a wage gap of almost 7 percent remains even with theextensive set of variables in the analysis, there is some indication that female universitygraduates are facing wage discrimination on the German labor market.

Contents

1 Introduction

2 Theoretical background

3 Data and measures

3.1 Descriptive statistics

3.2 Multivariate analysis

4 Conclusion

References

* Both authors contributed equally to this paper. We would like to thank Josef Brüderl, Johannes Giesecke,Irena Kogan, Sigrun Olafsdottir, Reinhard Pollak, Stephanie Steinmetz and two anonymous reviewers fortheir helpful comments. An earlier version of this paper was presented at the 2005 ISA Research Committee28 (Social Stratification) meeting in Oslo.

ZAF 2/2006, S. 235Ð253 235

Tracing the gender wage gap David Reimer and Jette Schröder

1 Introduction

Research has shown that in Germany Ð as in otherindustrialized nations Ð women’s wages are lowerthan those of men. Although the gender wage gaphas decreased in Germany in recent decades (Lauer2000), on average women still earn 24 percent lessthan men (Hinz/Gartner 2005). In contrast to resultsfrom a number of US studies which examined theearnings differentials between men and women, forexample Petersen and Morgan (1995), a new studyby Hinz and Gartner (2005) finds that the distribu-tion of men and women into different kinds of occu-pations, firms, and establishments cannot explain allof the gender differences in wages in Germany.When comparing men’s and women’s pay for thesame job, in the same occupation and the same firm,Hinz and Gartner (2005) still find that women earn15 percent less than men.1 Adding further controlsfor experience and education leads to a reduction ofthe wage gap to 12 percent and even in more nar-rowly defined occupational (ISCO) groups they findearnings differentials; the differences are smallestamong academics and upper management but stilladd up to 5Ð7 percent (Hinz/Gartner 2005: 34). Ina recent German study using data from the GermanSocio-Economic Panel (GSOEP), Lauer (2000)finds a gender wage gap of 15% in Germany for theGSOEP waves from 1994 to 1997, controlling forlevel of education, experience, training-occupationmatch and feminization of occupation.

When providing estimates of the gender earningsgap, most German studies only rarely address thecausal mechanisms responsible for the gender gap,mainly due to a lack of adequate measures in mostrepresentative population surveys. We try to com-plement existing research on gender earnings differ-entials by examining earnings for male and femaleuniversity graduates from the same field of study atthe same university. By analyzing wage differencesat career entry, potential gender differences in theaccumulation of human capital during working lifeand further market constraints between men andwomen are reduced to a minimum (e.g. Marini andFan 1997: 588). Furthermore, a unique dataset fromthe Mannheim Social Sciences Graduate Survey isused, which includes a number of detailed measuresconcerning skills and qualifications. These data en-able us to control for gender differences in humancapital acquired at university and outside universitystudies, as well as in ability and attitudes towardswork in a more refined way than is possible with

1 The analyses by Hinz and Gartner were restricted to full-timeemployees in 2001.

236 ZAF 2/2006

standard population surveys, thus reducing potentialproblems due to unobserved heterogeneity betweenmale and female graduates.

Only scant research focusing on gender differencesin wages for university graduates is available in Ger-many. A considerable number of international stud-ies, most of them from the United States, have dem-onstrated that among university graduates the gen-der wage gap can to some extent be attributed tomen and women choosing different fields of study:controlling for the field of study explains a sizeablepart of the gender wage gap (Daymont/Andrisani1984; Finnie/Frenette 2003; Loury 1997). However,earnings differences also exist between male and fe-male graduates within the same field. Hecker(1998), who analyzes earnings differentials betweenuniversity graduates with different major fields ofstudies in the US and differentiates between agegroups and qualificational levels (bachelor, master,doctoral degree), finds that women earn less at mostdegree levels in most fields of study, with some in-teresting variation between fields. Looking at the re-sults on earnings differentials between social scien-ces graduates, namely sociologists and psychologists,men earn more than their female counterparts at alldegree levels (Hecker 1998: 64).2 Nevertheless, ad-vanced tertiary degrees seem to reduce the genderdifferences in wages compared to lower degree lev-els (Montgomery/Powell 2003). In one of the fewGerman studies on income differentials betweenuniversity graduates, Machin and Puhani (2003),who use data from the German 1996 Labor ForceSurvey, find large gross differences in pay (28%).They show that differentiating between fields ofstudy helps to explain part of the wage gap betweenmale and female graduates. However, the set of ex-planatory variables used in their analyses is fairlylimited.3 Before turning to the analysis we give abrief overview of the mechanisms that have beenidentified in the literature as causing gender differ-entials in earnings.

2 Theoretical background

Human capital

Human capital theory suggests that women acquireless or different types of human capital than menand are therefore paid less (Becker 1991; Mincer/

2 With the exception of psychologists with a master’s degree inthe age group 25Ð34, where no gender wage gap was found.3 Unfortunately they do not report the size of the gender wagegap after controlling for the field of study.

David Reimer and Jette Schröder Tracing the gender wage gap

Polachek 1974). Earnings differentials stem fromproductivity differences associated with human capi-tal. It is assumed that initially women invest less inhuman capital than men because of the anticipateddivision of labor in the family and that after forminga family union they acquire less human capital be-cause of the actual division of labor between thespouses. Becker (1985) argues further that (married)women economize on the effort they expend onmarket work by seeking less demanding jobs. Fol-lowing these propositions, the gender wage gap canbe explained by human capital differences betweenmen and women and should disappear if humancapital is adequately measured. Consequently, it canbe argued that the gender wage gap found usingstandard population surveys could be a result ofpoor measures for human capital due to data limita-tions Ð usually only the level or years of educationand the amount of work experience are available.Because university graduates can have other skillsthat are highly valued by German employers Ð suchas internship experience, foreign language skills, orcomputer skills Ð it is likely that these additionalskills are important human capital endowments thatare relevant for the labor market. Gender differen-ces concerning investment in these endowmentsmight therefore result in income differences. Hence,differences in human capital which usually remainunobserved may possibly result in an overestimationof the gender wage gap. Consequently an existinggender wage gap for university graduates should dis-appear Ð or at least be reduced substantially Ð aftercontrolling for human capital with detailed meas-ures.

Job search

Differences between men and women with respectto the search effort prior to accepting employmentare another possible explanation for differences instarting pay. In neoclassical search models the dura-tion of the search depends on the reservation wageof the individual searching for a job Ð with higherreservation wages resulting in a longer job search.In the typical sequential model, job-seekers obtaininformation about one employer at every time-inter-val and if they receive a job offer they decidewhether to accept it or not, depending on the wageoffered. The distribution of wage offers is knownand the reservation wage is chosen so that the netincome of the search is maximized. Differences inthe expected tenure of the job therefore have aninfluence on the reservation wage (depending on themodel): individuals with a shorter time horizon havelower reservation wages (Pissarides 1985; Ziegler/Brüderl/Diekmann 1988). If women expect to workless due to childrearing or other family obligations,

ZAF 2/2006 237

these differences between the sexes as regardsexpected labor force attachment would result inwomen’s reservation wages being lower and conse-quently their job search being shorter and acceptedwages lower. Standard population surveys do notusually include information on reservation wagesand the duration of the job search. Therefore, partof the unexplained gender wage gap could be theresult of differences between men and women intheir search effort caused by differences in reserva-tion wages. Following these arguments, an existinggender wage gap should diminish if the job-searcheffort is adequately controlled for.

Career attitudes and job characteristics

Sociologists emphasize the role of gender differen-ces in career attitudes and aspirations as mechanismsresponsible for the sorting of men and women intodifferent kinds of jobs that offer different kinds ofcareer rewards. This means that gender-specific so-cialization patterns cause women to aspire to differ-ent jobs (Marini et al. 1996; Shu/Marini 1998). Evenmen and women with a similar amount and type ofeducation choose different jobs (Marini/Fan 1997:591). Thus, the distribution of men and women indifferent economic sectors and jobs or working timearrangements offering different levels of rewardscan be seen as the result of differing career attitudesand aspirations. Additionally, demand-side factorsmight play a role in the sorting of women into differ-ent sectors, occupations, functions and hierarchicallevels (see next section). Career attitudes and Ð irre-spective of the mechanisms at work Ð economic sec-tors and job characteristics can be expected to ex-plain part of the gender wage gap. Therefore, an ex-isting gender wage gap should be reduced when ca-reer attitudes and job characteristics are controlledfor.

Discrimination

A final explanation for the gender wage gap is dis-crimination. Altonji and Blank (1999: 3168) definelabor market discrimination as “a situation in whichpersons who provide labor market services and whoare equally productive in a physical or materialsense are treated unequally in a way that is relatedto an observable characteristic such as race, ethnic-ity, or gender.” Economic theories of discriminationcan be broadly classified into two categories: Thefirst is “statistical discrimination”, which refers tothe strategy of employers using group-level charac-teristics such as gender or race, which are easy toobserve (and at group level are correlated with pro-ductivity), to evaluate the productivity of applicants.Gender might be used as an indicator of lower pro-


ductivity or lower expected job tenure. The secondtheory, introduced by Becker (1971), relates dis-crimination to employer preferences: employerswith a “taste for discrimination” are prejudicedagainst members of a specific (minority) group. Forexample, employers with a taste for discriminationagainst women will employ less than the profit-max-imizing number of women. Instead, they will hire agreater number of equally skilled but more highlypaid men. Becker’s model suggests that in a per-fectly competitive market, discriminating employerscannot survive. The wage gap between men andwomen will therefore decline as discriminators areforced to leave the market altogether.4

In addition to these theories that explain employers’motivation for discrimination, it is useful to differen-tiate between the different mechanisms by whichdiscrimination can come about. Petersen and Sa-porta (2004) differentiate between “allocative dis-crimination” and “within-job discrimination”.5 Theydefine allocative discrimination as the differentialallocation of men and women to occupations andestablishments that differ in the wages they pay. Thisentails sorting men and women into different jobsat the point of hire and differences in the subse-quent rates of promotion and dismissal (Petersen/Saporta 2004: 853). Within-job discrimination on theother hand is differential payment of a man and awoman who are equally qualified and are doing thesame work for the same employer. This type of dis-crimination could also be labeled direct wage dis-crimination. Following Petersen and Saporta (2004)allocative discrimination at the point of hire pre-sents the most advantageous “opportunity structurefor discrimination” because discriminatory em-ployer practices are hard to trace or document atthis point.

While it can be investigated directly with surveydata whether the previously mentioned mechanismscontribute to the gender wage gap, it is very difficultto assess the extent to which the gender wage gap iscaused by discrimination. New approaches for meas-uring discrimination use non-standard research de-signs such as Goldin and Rouse’s (2000) analysis ofdata from (gender-) blind auditions for music or-chestras or Bertrand and Mullainathans’ (2004)analysis of employer responses to fictitious resumes.However, analyses of the gender wage gap usingsurvey data can only measure discrimination as a

4 For a review of economic theories of discrimination see Altonjiand Blank (1999: 3168Ð3191).5 They list “valuative discrimination” as a third discriminatorypractice, which is discrimination against a whole class of jobs heldprimarily by women.

238 ZAF 2/2006

residual Ð the unexplained portion of the genderwage gap (see Gerlach 1987: 590). Following this ap-proach, analyses based on standard surveys run therisk of overestimating wage discrimination becauseimportant gender differences might not be ob-served. However, given the extensive set of meas-ures available in the dataset used in the subsequentanalysis, we are fairly confident that the major partof a potentially remaining gap could be attributedto wage discrimination.

To sum up, gender differences in earnings can becaused by gender differences in human capital andgender differences in job-search effort. Further-more, the sorting of men and women into differenteconomic sectors, occupations and job functions be-cause of men’s and women’s differing gender prefer-ences or because of allocative discrimination couldlead to a gender wage gap. If the listed mechanismscannot explain gender differences in earnings thiscould be seen as an indication of direct wage dis-crimination, which, however, cannot be measureddirectly but only as a residual category.

3 Data and measures

We use unique data from the 2001/2002 MannheimUniversity Social Sciences Graduate Survey, whichovercome several of the previously mentioned limi-tations of standard population surveys in investigat-ing the gender wage gap. First, the sample is homog-enous with respect to the measures normally usedin income analyses Ð all of the respondents are uni-versity graduates, have a degree in the same fieldof study and are observed at career entry. Second,detailed measures of human capital, job search andcareer attitudes are available, which are usually notincluded in standard population surveys.

The 2001/2002 Mannheim University Social SciencesGraduate Survey is a mail survey of all former stu-dents of the social sciences department who gradu-ated between the winter semester 1994/1995 and thesummer semester 2000. All of them graduated witha Diplom or a Magister Artium degree, both ofwhich are roughly equivalent to a master’s degreeat an Anglo-Saxon higher education institution. Bymeans of intensive research it was possible to up-date the addresses of 895 of a total of 1100 studentswho graduated during that period. Overall, 629 ofthe 895 graduates (70.3%) who received the ques-tionnaire took part in the survey. Both the relativelyhigh response rate and a comparison of the demo-graphics between respondents and non-respondentsindicate the good quality of the data. The standard-


ized questionnaire included a wide range of ques-tions concerning the retrospective evaluation of thecompleted course of studies, the acquisition of quali-fications and work experience during the time atuniversity, information on job search, an extensivecareer biography with information on each employ-ment spell after graduation, current career aspira-tions, and attitudes toward the work-life balance aswell as detailed demographics.6

Measures

This section gives a short description of the meas-ures that were constructed in order to identify thedescribed explanatory mechanisms. The dependentvariable is the hourly wage at the beginning of thefirst regular job, and for the multivariate analysis thelogarithm of the hourly wage is used. This variableis constructed on the basis of the self-reportedmonthly income before tax at the beginning of thefirst regular job. Since the respondents entered thelabor market in different years, their starting salarieswere adjusted for inflation using the German con-sumer price index (Statistisches Bundesamt 2002:612). To account for working time we use hourlyincome, that is, we divide the reported monthly in-come by the reported contracted hours per weekmultiplied by 4.33 (the average number of weeks ina month).7 We will proceed by discussing our ex-planatory variables (Table 1).

Human capital

Since women might choose to acquire different Ðless highly valued Ð types of human capital thanmen, two measures indicating their chosen speciali-zation within the social sciences are used to measurethe human capital acquired at university: the degreesubject (psychology, sociology, social sciences majorand social sciences minor8) and whether businessadministration or economics was chosen as a minorsubject. Because women might put less effort intotheir studies than men due to a shorter expected la-bor market career, we include two additional meas-ures for human capital acquired at university: thefinal grade and the duration of studies. In Germany,employers typically associate short study times with

6 The questionnaire can be accessed at:http://www.sowi.uni-mannheim.de/lehrstuehle/lessm/absol/absol01_frabo.pdf.7 For 12 respondents who did not report contracted workinghours we used the self-reported real working hours that respon-dents were also asked to specify.8 Students with a social sciences major were graduates who hadone major in the Social Sciences Department and another majorin another department of the university. Students with a socialsciences minor only had a minor subject in the Social SciencesDepartment and a major in a different department.

ZAF 2/2006 239

a high level of ability and motivation of graduates.Students also acquire human capital outside theiruniversity studies. Again, following human capitaltheory, women might invest less in these additionalqualifications than men do. Therefore we utilizemeasures of students’ skills and competencies in ad-dition to their formal tertiary education. We includefour dummy variables that indicate whether the stu-dent completed vocational training before enteringuniversity, whether she/he completed an internshipwhile at university, whether she/he had a student jobequal or equivalent to that of a research or teachingassistant, and whether she/he spent time studying orworking abroad, a measure that indicates the acqui-sition of intercultural skills. Furthermore, we includemeasures of self-reported foreign language skillsand computer skills at the time of graduation.

Job search

The data include detailed information about the re-spondents’ job search. Respondents were asked toreport the duration of their job search as well as thenumber of times they applied for a job, which givesus a detailed picture of their job-search effort. Weuse these two dimensions of the job-search effort inour analysis by including one dummy variable indi-cating the number of times the respondent appliedfor a job (more than five times) and one dummyvariable for the duration of the job search (morethan two months, see Table 1).9 We decided againstusing a linear functional form for both measures dueto their skewed distribution. We also includeddummy variables for respondents with missing val-ues for one or both of the search variables.10

Job attitudes

Since our study is a cross-section, the informationavailable on job attitudes and career preferencesdoes not refer to the time of the labor-market entrybut to the time of the survey. Trying to explain thestarting salary with attitudes measured after the firstjob is problematic; nevertheless we think that atleast to some extent one can assume a stability ofcareer attitudes and preferences during the firstyears after graduation. If current career attitudescould explain gender differences in starting pay, thisfinding could be treated as a possible, albeit uncer-tain, hint of an effect at work. In the survey, respon-

9 Having measures of both job-search duration and the numberof job applications is advantageous because the assumption ofjob-search theory that the number of intervals of the job searchis equivalent to the number of employer contacts seems unrealis-tic: some job-seekers apply for 10 jobs per month whilst othersapply for only one per month.10 16 cases have missing values on both search variables.


240 ZAF 2/2006


dents were asked how “important” they considered17 different aspects of their working life (for exam-ple “flexible working hours”). For each item theyanswered on a scale ranging from 1 (not importantat all) to 5 (very important). We use only one of theitems Ð the importance of “high income” Ð as asingle variable in our analysis and combine the otheritems to five indices on the basis of a principal com-ponent factor analysis.11 The five indices used arelabeled: “Skill utilization”, “Job content”, “Careerprospects”, “Social environment/job security” and“Work-life balance”.

Economic sector and job characteristics

Respondents were also asked to indicate the eco-nomic sector of their job. We combine the originalcategory scheme to seven sectors based on the simi-larity of the sectors to each other as well as cell fre-quencies. In addition to the economic sectors, fourmeasures of job characteristics are included in theanalysis, in order to account for the allocation ofwomen into jobs associated with earnings disadvan-tages. We include dummy variables for whether therespondent had a temporary job or a part-time job,whether the job was a self-reported educational mis-match for a person with a tertiary degree, andwhether the respondents reported that real workinghours exceeded 120% of the reported contractedhours.

Control variables

A set of control variables is included in addition tothe explanatory variables. We include age at careerentry, especially because wage scales in the publicsector are a function of the position held (job title)and age, irrespective of work experience. Further-more, we include dummy variables for the year ofgraduation in order to capture possible differencesin the job market at the time of graduation. Finally,we include a dummy capturing whether the respond-ent had a child at the time of graduation, because inGermany public sector employees who have chil-dren receive child benefits as part of their salary.12

The dataset contains 629 cases. However, it was notpossible to use all of the cases in the analysis. Afterexcluding respondents with a teaching degree(“Lehramt”), as well as respondents who had not

11 We calculated the factor analysis with and without the “in-come” item. The same set of factors was identified irrespective ofthe inclusion of the income-attitude variable.12 Marital status could not be included as a control variable inthe regression models for starting salaries because the data con-tains only information about marital status at the time of the in-terview Ð and not at the beginning of the first job.

ZAF 2/2006 241

entered a regular job, were self-employed, workingabroad, or worked fewer than 16 hours per week13,465 cases from the basis for the analysis. Due tomissing values Ð mainly for the dependent vari-able Ð 399 cases remain for the analyses.

3.1 Descriptive statistics

As shown in Table 2, the mean hourly wage isDM 31.36 for women and DM 32.79 for men, thedifference being significant at the 5 percent level.Figure 1 illustrates the distribution of hourly wagesby gender. In the area of low hourly wages the distri-bution of women’s wages has a higher density thanmen’s. In the area of high hourly wages, on the otherhand, the distribution of men’s wages has a higherdensity than women’s. Comparing the wages of menand women at the 25%, 50% and 75% quantiles ofthe male and female wage distribution (not re-ported), one can see that the 25% and 75% quan-tiles of the female distribution are clearly lower than

13 Because the salary of teachers is fixed in Germany, these cases[43] are not used for the analysis. Of the remaining 586 graduates,522 had a regular job after graduation, but 35 of them were self-employed and 15 were working abroad; 7 of the respondents wereworking part-time with fewer than 16 hours per week. Examiningthe female/male distribution across the different categories showsthat more women than men (13 percent and 7 percent respec-tively) have not held a regular job yet and of those with a regularjob more men than women took up self employment (10 percentand 4 percent respectively) Thus the wage disadvantage for fe-males might be slightly underestimated in the subsequent analysisassuming that individuals with poor career prospects are morelikely not to enter the labour market. No substantial gender dif-ferences can be found in the other categories (working abroad,part-time work with fewer than 16 hours, missing in the depend-ent variable).


242 ZAF 2/2006


the respective quantiles of the male distributioneven though the median wage is fairly similar forthe two groups, at about DM 32.

Table 2 gives an overview of gender differences forthe independent variables. If the gender wage gap isdue to human capital differences one would expectmen to have more human capital than women ordifferent types of human capital which are valuedmore highly on the labor market.

The picture is not clear-cut, however. There are sig-nificant differences in the distribution of males andfemales across degree types, which might contributeto the explanation of the gender wage gap if thedegrees are valued differently on the labor market.Women’s duration of studies is shorter, their gradesare better than those of men (not significant) andwomen do not seem to have less additional humancapital than men. Women perform better than menon several of the variables capturing additional hu-man capital: a larger proportion of women havecompleted vocational training, internships, and haveexperience abroad (only vocational training is signi-ficant) and women have significantly more languageskills. Men, however, have worked more as research/teaching assistants (not significant) and possess sig-nificantly more computer skills.

With regard to the job search, results are in line withthe assumption of women making a smaller searcheffort: the proportion of women who submit morethan five applications is smaller than that of men;the same is true of the proportion of respondentslooking for a job for more than two months. How-ever, the differences are not significant.

No significant differences are found when looking atthe distribution of men and women across economicsectors. The biggest difference appears for social ser-vices and health care: the proportion of womenworking in this sector is seven percentage pointslarger than that of men. In contrast, significant dif-ferences exist regarding job attitudes. In general,women tend to have higher scores on all attitudedimensions Ð even income. But while the differen-ces for income, skill utilization and career prospectsare small and insignificant, they are larger for socialenvironment/job security, job content, and work-lifebalance, with the last two being significant. Further-more, differences appear concerning the jobs heldby men and women: women more often hold part-time jobs (significant), temporary jobs and jobs forwhich a university degree is not necessary (not sig-nificant). These job characteristics are likely to beassociated with lower remuneration, and could ex-plain part of the gender wage gap. Finally a greater

ZAF 2/2006 243

proportion of women compared to men work sur-plus hours in our sample (not significant).14

3.2 Multivariate analysis

Our strategy for investigating the wage gap is to esti-mate a sequence of nested regression models of thehourly wage at the beginning of the first job. Westart with the model containing the variable Female(Fi), the control variables (Xij) and an error term(ei). In our baseline model the hourly wage (Wi) isthus:

lnWi � α � γFi ��j�j Xij � ei

The effect of Female in Model A measures relativewage differences between men and women con-trolling for age, year of graduation and whetherthe respondent had a child at the time of gradua-tion. Expressed in percentages, women earn100* [1-exp(γ)] percent less than men. We then addto the model one by one the different sets of meas-ures that were identified as possible mechanisms ex-plaining the gender wage gap. The first column ofTable 3 gives an overview of the additional variablesincluded in the successive models. In Model B, hu-man capital variables are added. If the effect of fe-male declines after adding the human capital vari-ables, it can be assumed that part of the gender wagegap is caused by gender differences in human capi-tal. Namely: the decrease in 100* [1 -exp(γ)] wouldindicate how many percent women earn less relativeto men due to human capital differences. In Mod-el C, the job search variables are added with the in-tention of determining how much of the genderwage gap is due to a different search effort. In Mod-el D we introduce job attitudes. Because of theproblematic post hoc measurement of job attitudes,they are excluded in Models E and F. In Model Ewe add dummy variables for economic sector, andfinally, in Model F job characteristics are included.

In order to identify the unexplained part of the gen-der wage gap Ð and thus possible wage discrimina-tion more clearly Ð we additionally conduct a stand-ard Blinder-Oaxaca wage decomposition (Blinder1973; Oaxaca 1973) for our final Model (F). Thewage decomposition is based on separate regres-sions for males (M) and females (F):

lnWFi � αF ��j

�Fj XF

ij � eFi

14 This difference is partly due to the fact that more women holdpart-time positions and overtime is more likely in part-time posi-tions than in full-time positions.


lnWMi � αM ��j

�Mj XM

ij � eMi

The raw wage differential between men and womenis decomposed into the explained gap and the unex-plained gap.15

ln WM � ln WF � (αM � αF) ��jXF

j (�Mj � �F

j ) �

Raw gap Unexplained gap

� �j�M

j (XMj � XF

j )

Explained gap

The explained gap is the difference between menand women that is attributable to observed differen-ces in the independent variables. The unexplainedgap is the difference attributable to differing wageequations of men and women; this means the wagedifference due to differing returns on the independ-ent variables (� differing coefficients) plus the wagedifference between men and women, which is notcaptured by independent variables (� difference inshift parameter). The unexplained part of the gen-der wage gap is often labeled discrimination. How-ever, if important control variables are missing inthe model, the unexplained gap captures not onlywage discrimination but also these unobserved dif-ferences (Altonji/Blank 1999: 3156).

Turning to the results of the nested regression esti-mation, the model fit statistics for Models A to F

15 Because standard Blinder-Oaxaca wage decomposition analy-ses refer to the wage of the high-income group in relation to thelow-income group, we follow this standard and report the wageof men in relation to the wage of women in our wage decomposi-tion even though we report the income of women in relation tomen in the regression models.

244 ZAF 2/2006

are displayed in Table 3 and the regression coeffi-cients are displayed in Table 4. Model A, which in-cludes the control variables, confirms the bivariateanalysis from the previous section: women earn sig-nificantly16 less than men, the female disadvantagebeing 4.7% of men’s income. This difference can beconsidered substantial bearing in mind the homo-genous nature of our sample. In Model B the humancapital variables are introduced. The direction of ef-fects for grade, duration of studies and all the vari-ables capturing human capital acquired outside uni-versity are in line with expectations Ð that is, morehuman capital is associated with higher wages. Goodgrades, vocational training, internships and com-puter skills increase the hourly wage significantly.The effects for duration of studies, teaching assist-ance, experience abroad and languages, however,are not significant. Some significant effects appearconcerning the degree subject: graduates with a so-cial sciences minor degree Ð typically graduateswith a major in the humanities Ð earn significantlyless than the psychology graduates in the referencecategory. Furthermore, graduates who had chosen abusiness minor earn significantly more. The ex-plained variance of the logarithm of the hourly wageincreases from 5 percent in Model A to 21 percentin Model B. Apparently the human capital measuresused are important predictors for explaining vari-ance in the hourly wages of social sciences gradu-ates. Unexpectedly however, the gender wage gapdoes not decline with the inclusion of human capitalmeasures in the model; on the contrary, it even in-creases by two percentage points to 6.6%. This isbecause the women in the sample do better on many

16 Due to the small analytical sample we chose to report if coeffi-cients are significant at the 10 % level in the tables. In the discus-sion of the results, however, we report the 5 % level of signifi-cance only.


relevant human capital measures. Consequently, thefemale graduates in our sample would be even moredisadvantaged if they had the same human capitalendowment as men.

In Model C the indicators for job-search effort areintroduced. The search variables do not improve themodel fit significantly (Table 3). Nevertheless, grad-uates who applied for more than five jobs (keepingthe search time constant), have significantly higherwages than graduates who applied for fewer thanfive jobs. On the other hand a longer search time(over two months) has a significantly negative effecton income. The hypothesis that part of the genderwage gap is caused by lower investment in the jobsearch associated with a lower reservation wage forwomen is not confirmed here: the gender wage gapdoes not decline, but remains constant when job-search variables are included in the model.

In Model D we look at whether the gender wagegap could be a result of men’s and women’s differentjob attitudes. The model contains the indices meas-uring job attitudes. These variables improve themodel fit significantly; the explained variance in-creases by about five percentage points. Not surpris-ingly, graduates who place value on “high income”earn more while graduates who find the social envi-ronment of the job important earn less. The negativeeffect of career prospects might indicate that gradu-ates who value future career opportunities have en-try-level jobs offering low starting salaries but possi-bly higher returns later on in the career. The effectsfor the other three indices are not significant. Sur-prisingly, even with the introduction of attitudes, thegender wage gap does not decline substantially. Be-cause job attitudes were measured at the time of theinterview, i.e. after starting the first job, the meas-ured job attitudes might be a result of experiencesmade on the labor market, which would mean thatthe causal link could be reversed. However, the ab-sence of a substantial decline in the gender wagegap indicates that differences in job attitudes do notcontribute much to the explanation of the genderwage gap in our sample of graduates. Because of theproblematic post hoc measurement of job attitudeswe do not include them in the two subsequent mod-els.

In Model E, the economic sector of the job is intro-duced as an independent variable. Because job atti-tudes are not included in Model E (or Model F),Model E has to be compared with Model C in orderto judge the contribution of the economic sector toexplaining income variance and the gender wagegap. The comparison shows that only a small part ofthe gender wage gap is due to the sorting of men

ZAF 2/2006 245

and women into different economic sectors: the gen-der wage gap declines by about half a percentagepoint, from 6.7 to 6.1 percent.17

In a final step, job characteristics are added to themodel (Model F). The explained variance increasesby 5.5 percentage points, indicating the importanceof job characteristics for explaining the wages of so-cial sciences graduates. The educational mismatchvariable has the largest impact: graduates with a jobfor which a university degree is not the norm or notneeded at all earn 11 percent less than graduateswith a job for which a degree is the norm. Graduateswith temporary jobs also earn significantly less, andthose working more than 120 percent of the con-tracted working hours earn significantly more. Fromour point of view, jobs with a lot of overtime workare paid better because when the work contract isconcluded (with official working hours) an implicitwork contract already exists which entails overtimework. The last job characteristic Ð part-time work Ðhas a positive effect, which is not significant how-ever. Part-time positions for social sciences gradu-ates at career entry are apparently not associatedwith a per hour earnings disadvantage. Surprisingly,the gender wage gap does not decrease when jobcharacteristics are controlled for, but increases mar-ginally by about half a percentage point to 6.7 per-cent. Therefore, even job characteristics cannot ex-plain the gender wage gap. In an attempt to shedlight on this unexpected result, we introduced thefour job characteristics one by one into the regres-sion (not reported). These regressions show thatboth degree mismatch and a temporary job lead toa small decrease in the gender wage gap and thatthe small increase in the gender wage gap fromModel E to Model F can be attributed to the jobcharacteristics part-time job and surplus hours Ðthese job characteristics have a positive effect, andwomen hold jobs with these characteristics more of-ten than men. Following the argument above re-garding surplus hours it is evident Ð as already inthe case of human capital variables Ð that the gen-der wage gap for social sciences graduates would beeven larger if women did not invest more in theircareers than men do.18

17 We checked whether a finer classification of economic sectorswould explain more of the wage gap, but the wage differencebetween men and women remained constant regardless of differ-ent classifications.18 In Model F a dummy for surplus hours is included. However,to make sure that the estimated gender wage gap cannot be ex-plained by men’s unpaid overtime work, we ran Model F (exclud-ing the dummy for surplus hours) using the hourly wage com-puted on the basis of the real working hours as a dependent vari-able. The estimated gender wage gap was even one percentagepoint larger in this specification.


246 ZAF 2/2006


The result of the regression analysis presented hereis supplemented by the Oaxaca-Blinder wage de-composition. The Oaxaca-Blinder wage decomposi-tion is based on Model F estimated separately formen and women (Table 5). Looking first at the dif-ferences between the coefficients in the male modeland the female model, we obtain a mixed picture:the increase in wages due to vocational training, forexample, is larger for male graduates, but the in-come increase for a business minor is greater forwomen. The coefficients of almost all of the vari-ables point in the same direction for men andwomen. Exceptions are the economic sector andlanguage skills, although the coefficients in the lattercase are close to zero for men and women. To testwhether the effects of our explanatory variables dif-fered significantly between men and women Mod-el F is estimated including interactions of all varia-bles with the dummy female (not reported). Apartfrom economic sector none of the interactions is sig-nificant.19

Turning to the wage decomposition, Table 6 showsthat differing values between men and women forthe independent variables cannot explain the genderwage gap: On the contrary, the 3.0 percent gap ex-plained by differing values for the independent vari-ables is negative, indicating that if women main-tained their human capital and job characteristicsbut were given the male coefficients and shift pa-rameter, men would earn 3 percent less thanwomen. The unexplained gap due to differing coeffi-cients and shift parameters is 7.9 percent, which im-plies that if men and women kept their wage equa-tion, but men had the same values for the independ-ent variables on average as women, men would earn7.9 percent more than women. The unexplained gapis therefore larger than the raw gap, which is 4.7 per-cent. The result of the wage decomposition is thusconsistent with the outcome of the previous regres-sion models: the gender wage gap would be evenlarger if women had the same human capital and jobcharacteristics as men.

To find out how robust our findings are and to inves-tigate further the mechanisms influencing the wagegap we conduct several additional analyses. The firstanalysis addresses the question of whether themechanisms influencing the gender wage gap are thesame for the public and the private sector. Since full-time and part-time employees are included in theanalyses, we further investigate whether the results

19 Because of high collinearity in the model with full interactionswe tested whether interaction effects would turn out significant ifwe introduced them one at a time in Model F. However, the re-sults did not differ.

ZAF 2/2006 247


hold if the analysis is restricted to full-time em-ployees. Finally, we look at how the gender wagegap develops with work experience and whethercontrolling for human capital, economic sector andjob characteristics has the same result later in thecareer.

Public and private sector and full-timeemployees

A large percentage of the graduates in our sample,36.8%, works in the public sector. Empirical resultsshow that the gender wage gap is usually smaller inthe public sector than in the private sector (Gornick/Jacobs 1998). To check whether this holds true forour sample and to see whether the effects of ourtheoretically relevant independent variables are thesame in both sectors, we ran separate regressions forthe two sectors (Table 7).20 The models are equiva-lent to Model F in Table 4, apart from the economicsector variables, which are not included. The esti-mated gender wage gap for the private sector is9.8 percent, the estimated wage gap for the publicsector 6.3 percent. In the regression for the publicsector, the gender effect is not significant. The re-sults show interesting differences between the mod-els: first of all, the effect of educational mismatch isfar greater in the public sector. This makes sensesince in the public sector wages are firmly linked tothe educational requirements of a position. Regu-lated wage standards in the public sector could alsoexplain why jobs with surplus hours are only paidmore in the private sector: wages in the public sectorcannot be adjusted to an implicit work contract, atleast not at entry level. The described effects arethe only ones which differ significantly between thesectors (tested in a model with all interactions). Sur-prisingly, part-time work does not have a negativeeffect in the private sector.

20 A dummy variable for private and public sector was not in-cluded in the previous models because of the high collinearitywith economic sector.

248 ZAF 2/2006

In the next step Model F is estimated for full-timeemployees only (Table 7) in order to confirm thatthe estimated wage gap is not a result of job place-ment into full-time and part-time jobs. Despite thefact that this regression is based on only 73% of thecases used in the previous model, the estimatedwage gap Ð with a female earnings disadvantage of6.9% Ð is about the same as in the model with thefull sample, where it was 6.7%.

Development of the gender wage gap withwork experience

Since the dataset used includes a career biographywith all of the respondents’ employment episodesafter graduation, we capitalize on this feature in or-der to test whether the gender wage gap widens withwork experience. We also want to investigatewhether the result that the gender wage gap cannotbe explained by the set of explanatory variablesused can be replicated when income later in the ca-reer is used.

We therefore run two models resembling Model Aand Model F (from Table 4), except for the job-search indicators, using the logarithm of the hourlywage at the time of the survey as the dependentvariable. We additionally control for work experi-ence21 in these models because the respondents inour sample graduated in different years.22

All of the variables concerning the job refer to thejob held at the time of the interview. The child vari-able also refers to the time of the interview. We ex-clude respondents whose graduation was less thantwo years before the interview in order to guaranteea minimum of potential work experience. Table A1(Appendix) shows that the mean hourly wage at thetime of the interview was substantially higher thanthe mean of the first hourly wage, with women earn-ing on average DM 38.98 and men earningDM 42.34.

Turning to the regression results, Model Acurrent

shows that the estimated female wage disadvantageis 6.6 percent of men’s income (Table 8). Controllingadditionally for human capital, economic sector and

21 Work experience disregarding unemployment spells, childcare-leave and other non-work related spells.22 In the regression models for the current job, we could haveadditionally included marital status as the control variable. How-ever, to maintain comparability with the models of the first job(for which information about the marital status is not available),we did not. We did, however, check what effect the introductionof a dummy variable indicating whether the respondent was mar-ried at the time of interview has. There is no substantial changein the gender wage gap, the variable itself has an insignificantnegative effect.


ZAF 2/2006 249


250 ZAF 2/2006

job characteristics (Model Fcurrent), the gap increasesto 7.3 percent, the difference being significant inboth models. Therefore, the analysis using income atthe time of the interview confirms the result that thegender wage gap would Ð if at all Ð be bigger ifwomen had the same amount and type of humancapital as men and held jobs with the same charac-teristics. Comparing the results with the estimatedwage gap at job entry there is no sign of a reductionin the gender wage gap with work experience. Ifanything, the gender wage gap widens. In the modelwith controls only, the gender wage gap is about twopercentage points higher for the current job than forthe first job, and in the model with all explanatoryvariables (excluding attitudes) it is 0.6 percentagepoints higher. Overall, these results are in line withfindings which suggest that gender differences inearnings at career entry put women at a disadvan-tage that they cannot make up for later in their ca-reer (Gerhart 1990; Kunze 2005).

To sum up the results presented in this paper, re-gression models of the hourly wage at job entryshow that the gender difference in wages does notdisappear even if a full set of theoretically relevantmeasures is introduced into the models. On the con-trary, the introduction of variables capturing humancapital even leads to a small increase in the wagegap at job entry from 4.7 to 6.7 percent. The intro-duction of further measures for job-search effort,job attitudes, economic sectors and job characteris-tics hardly changes the size of the gender coefficientat all. This is remarkable considering that all the setsof variables Ð apart from the measures for jobsearch Ð increased the model fit significantly. In thefinal model (Model F, Table 4), 31% of the varianceof log hourly wages is explained, which is substantialconsidering that educational level and work experi-ence are held constant due to the nature of our sam-ple. A wage decomposition of the final model alsoconfirmed that the gap would even be larger ifwomen had the same human capital and job charac-teristics as men.

Further analyses revealed that some of the mecha-nisms we identified as influencing wages for univer-sity graduates operate quite differently in the publicand the private sector. When restricting the modelto full-time employees the gap remains significant ataround 7 percent. Furthermore, the additional ana-lyses using the current job show the same pattern:the gender wage gap does not decline but rather in-creases when all of the explanatory variables are in-cluded in the model.


4 Conclusion

By looking at a very homogenous population ofmale and female university graduates who all havethe same level and type of education this paper hasshed light on gender earnings differentials at careerentry. To analyze the impact of various explanatorymechanisms on the gender wage gap, different setsof measures were added one by one in a sequenceof nested regression models. Through the introduc-tion of detailed measures of human capital that areusually not available in standard population surveys,we tested the human capital explanation for genderdifferences in earnings. Contrary to the predictionsof human capital theory, the gender wage gap didnot decline but rather increased when human capitalmeasures were introduced. This implies that the fe-male graduates in the study would be even moredisadvantaged if they had the same human capitalendowment as men. We see this as an indication thatstudies looking at university graduates without tak-ing into account such detailed measures for skillsand competencies might underestimate the genderearnings differential. The introduction of measuresof the job-search effort, which were intended to cap-ture possible differences in income expectancies,also failed to reduce the gender gap. Furthermore,turning to the mechanisms that allocate men andwomen into sectors of the economy and jobs offer-ing differing rewards we found that although impor-tant for explaining variance in hourly wages, neitherattitudes toward the labor market nor the measuredjob characteristics could explain the earnings gapin our study. In line with these results a Blinder-Oaxaca wage decomposition showed that if thewomen in our sample had the same average charac-teristics as men the gender wage gap would be evenlarger.

Considering that a wage gap of almost 7 percent re-mains even with the extensive set of variables in theanalysis, there is at least some indication that femaleuniversity graduates are facing wage discriminationon the German labor market. The analysis does notpermit us to specify what type of discrimination thisdifference can be attributed to. Nevertheless Ðgiven that the introduction of the job-related vari-ables did not reduce the gender wage gap substan-tially Ð it seems unlikely that the allocation of fe-male and male graduates into different economicsectors or jobs is the mechanism behind the residualgap. Thus, it seems most likely that direct wage dis-crimination is the mechanism behind the earningsdifferential. Because of different “tastes” or produc-tivity expectations employers might choose to offerequally qualified female graduates lower salariesthan men. However, part of the unexplained gap

ZAF 2/2006 251

might result from differences in reservation wagesthat were possibly not fully captured by the job-search variables. Furthermore, differences in the ne-gotiation behavior of men and women, which mightpartly be related to gender differences in reservationwages, could play a role: male graduates might bemore likely to ask for more money and employersmight reward this boldness with higher wages. Newresearch by Babcock et al. (2003) showed for MBAgraduates in the United States that more womenthan men accepted the initial salary offer: only 7%of women but 57% of men had attempted to negoti-ate after the employer had made the initial offer.

Our findings are in line with the previously men-tioned findings of Hinz and Gartner (2005). Boththeir results and the results presented here suggestthat women in Germany have an earnings disadvan-tage compared to men which cannot be attributedentirely to human capital differences or the alloca-tion of women into less attractive jobs. As arguedbefore, wage discrimination seems to be the likelyexplanation. However, to ensure that differences inthe negotiation behavior of men and women are notthe reason for the earnings gap, future researchshould address this issue. Thus, future labor marketsurvey studies should try to incorporate measureson negotiation behavior and reservation wages. Fur-thermore, when looking at university graduates, de-tailed measures such as those we used are recom-mendable since they proved to be highly relevantfor the explanation of wages. An ideal survey designwould no doubt be a large-scale German panelstudy that includes information about ability, careeraspirations as well as job-search behavior, similar tothe National Longitudinal Study of Youth (NLS-72)or the High School and Beyond Study (HS&B) con-ducted in the United States.

References

Altonji, Joseph G./Blank, Rebecca (1999): Race and Gen-der in the Labor Market. In: Ashenfelter, O./Card, D. (ed.):Handbook of Labor Economics. Vol. 3C. Amsterdam:Elsevier, pp. 3143Ð3259.

Babcock, Linda/Laschever, Sara/Gelfand, Michele/Small,Deborah (2003): Nice Girls Don’t Ask Ð Women Negoti-ate less than Men Ð and Everyone Pays the Price. In: Hav-ard Business Review, Vol. 81, Number 10, pp. 14Ð16.

Becker, Gary S. (1971): The Economics of Discrimination.2nd. ed. Chicago: University of Chicago Press.

Becker, Gary S. (1985): Human-Capital, Effort, and theSexual Division of Labor. In: Journal of Labor Economics,Vol. 3, Number 1, pp. 33Ð58.


Becker, Gary S. (1991): A Treatise on the Family. EnlargedEdition. Cambridge: Harvard University Press.

Bertrand, Marianne/Mullainathan, Sendhil (2004): AreGreg and Emily more Employable than Lakisha and Ja-mal? A Field Experiment on Labor Market Discrimina-tion. In: American Economic Review, Vol. 94, Number 4,pp. 991Ð1013.

Blinder, Alan S. (1973): Wage Discrimination: ReducedForm and Structural Estimates. In: The Journal of HumanResources, Vol. 8, Number 4, pp. 436Ð455.

Daymont, Thomas N./Andrisani, Paul J. (1984): Job Prefer-ences, College Major, and the Gender Gap in Earnings. In:Journal of Human Resources, Vol. 19, Number 3, pp. 408Ð428.

England, Paula (1992): Comparable Worth: Theories andEvidence. Hawthorne, N.Y.: Aldine de Gruyter.

Finnie, Ross/Frenette, Marc (2003): Earning Differences byMajor Field of Study: Evidence from three Cohorts of Re-cent Canadian Graduates. In: Economics of Education Re-view, Vol. 22, Number 2, pp. 179Ð192.

Gerhart, Barry (1990): Gender Differences in Current andStarting Salaries Ð The Role of Performance, College Ma-jor and Job Title. In: Industrial and Labor Relations Re-view, Vol. 43, Number 4, pp. 418Ð433.

Gerlach, Knut (1987): A Note on Male-Female Wage Dif-ferences in West Germany. In: Journal of Human Resour-ces, Vol. 22, Number 4, pp. 584Ð592.

Goldin, Claudia/Rouse, Cecilia (2000): Orchestrating Im-partiality: The Impact of “Blind” Auditions on Female Mu-sicians. In: The American Economic Review, Vol. 90, Num-ber 4, pp. 715Ð741.

Gornick, Janet C./Jacobs, Jerry A. (1998): Gender, the Wel-fare State, and Public Employment: A Comparative Studyof Seven Industrialized Countries. In: American Sociologi-cal Review, Vol. 63, Number 5, pp. 688Ð710.

Hecker, Daniel E. (1998): Earnings of College Graduates;Women Compared with Men. In: Monthly Labor Review,Vol. 121, Number 3, pp. 62Ð71.

Hinz, Thomas/Gartner, Hermann (2005): The GenderWage Gap within Economic Sectors, Occupations, andFirms. In: Zeitschrift für Soziologie, Vol. 34, Number 1, pp.22Ð39.

Kunze, Astrid (2005): The evolution of the gender wagegap. In: Labour Economics, Vol. 12, Number 1, pp. 73Ð97.

Lauer, Charlotte (2000): Gender Wage Gap in West-Ger-many: How Far Do Gender Differences in Human CapitalMatter? ZEW Working Paper, Number 00Ð07.

252 ZAF 2/2006

Loury, Linda Datcher (1997): The Gender Earnings Gapamong College Educated Workers. In: Industrial and La-bour Relations Review, Vol. 50, Number 4, pp. 580Ð93.

Machin, Stephen and Puhani, Patrick A. (2003): Subject ofDegree and the Gender Wage Differential: Evidence fromthe UK and Germany. Economics Letters, Vol. 79, pp.393Ð400.

Marini, Margaret Mooney/Fan, Pi-Ling (1997): The Gen-der Gap in Earnings at Career Entry. In: American Socio-logical Review, Vol. 62, Number 4, pp. 588Ð604.

Marini, Margaret Mooney/Fan, Pi-Ling/Finley, Erica/Beu-tel, Ann M. (1996): Gender and Job Values. In: Sociologyof Education, Vol. 69, Number 1, pp. 49Ð65.

Mincer, Jacob/Polachek, Solomon W. (1974): Family Invest-ments in Human Capital Ð Earnings of Women. In: Jour-nal of Political Economy, Vol. 82, Number 2, pp. 76Ð108.

Montgomery, Mark/Powell, Irene (2003): Does an Ad-vanced Degree Reduce the Gender Wage Gap? Evidencefrom MBAs. In: Industrial Relations, Vol. 42, Number 3,pp. 396Ð418.

Oaxaca, Ronald L. (1973): Male-Female Wage Differen-tials in Urban Labor Markets. In: International EconomicReview, Vol. 14, Number 3, pp. 693Ð709.

Petersen, Trond/Morgan, Laurie A. (1995): Separate andUnequal Ð Occupation Establishment Sex Segregation andthe Gender Wage Gap. In: American Journal of Sociology,Vol. 101, Number 2, pp. 329Ð365.

Petersen, Trond/Saporta, Ishak (2004): The OpportunityStructure for Discrimination. In: American Journal of Soci-ology, Vol. 109, Number 4, pp. 852Ð901.

Pissarides, Christopher (1985): Job Search and the Func-tioning of Labour Markets. In: Carline, D. (ed.): Laboureconomics. London, New York: Longman, pp. 159Ð185.

Shu, Xiaoling/Marini, Margaret Mooney (1998): Gender-Related Change in Occupational Aspirations. In: Sociologyof Education, Vol. 71, Number 1, pp. 43Ð67.

Statistisches Bundesamt (2002): Statistisches Jahrbuch 2002für die Bundesrepublik Deutschland. Stuttgart: Metzler-Poeschel.

Ziegler, Rolf/Brüderl, Josef/Diekmann, Andreas (1988):Stellensuchdauer und Anfangseinkommen bei Hochschul-absolventen. Ein empirischer Beitrag zur Job-Search-The-orie. In: Zeitschrift für Wirtschafts- und Sozialwissenschaf-ten, Vol. 108, Number 2, pp. 247Ð270.


ZAF 2/2006 253

Documents

Tracing the gender wage gap: Income differences between ... · vestigating the gender wage gap, is used for the empirical analysis. The sample is ho-mogenous with respect to the measures