Employment discrimination against the disabled: an analysis of men and women on the Dutch labour market

(1)

Employment discrimination against the

disabled: an analysis of men and women on

the Dutch labour market

Author: J.E. Doornbos

(2)

Master’s Thesis Econometrics, Operations Research and Actuarial Studies

Supervisor: Dr. A. Postepska

(3)

Employment discrimination against the

disabled: an analysis of men and women on

the Dutch labour market

Author: J.E. Doornbos

July 1, 2019

Abstract

This study investigates the presence of labour market discrimination against disabled men and women in the Netherlands. It makes use of LISS data from 2008 till 2018. An Oaxaca-Blinder decomposition is used to decompose the difference in real hourly wages between non-disabled and disabled workers in an explained and an unexplained part. Next to that, differences in participation rates are also decomposed into explained and unexplained parts. The explained part is attributable to differences in human capital related characteristics, whereas the unexplained part might either be a consequence of discrimination on the labour market or is due to unobserved/omitted factors. It is found that more than 75 % of the difference in participation rates is unexplained and might hence be attributable to labour market discrimination. This results applies to both men and women. However, the real hourly wage differential is largely explained by differences in human capital characteristics.

(4)

Acknowledgements

I gratefully acknowledge CentERdata for giving me access to the LISS panel for scientific use. Especially, I would like to thank dr. A. Postepska from the University of Groningen for her constructive feedback and unconditional support.

1 Introduction

In 2003 the Law of Equal Treatment (Ministerie van Volksgezondheid, Welzijn en Sport, 2003) based on a handicap or chronical disease was introduced in the Netherlands. Among other things, unequal treatment on the labour market is forbidden by this law. Employers are forced to adapt their workplace such that it is accessible for everyone. However, this adaption is only forced to be executed if it is necessary in order to grant access to all employees and visitors. Next to that, it needs to be financially achievable for the company. On top of this law, there has been introduced a consequence for not conforming to it. Unequal treatment of disabled persons can be prosecuted under criminal law since January 2006. This may serve as a threat towards employers to conform to the Law of Equal Treatment more strictly. The Participation Law (Ministerie van Sociale Zaken en Werkgelegenheid, 2015) was introduced in January 2015. Everyone who is able to work but is not self-reliant on the job market belongs to the target audience of this law. This law intends to help more job-seekers find a job, especially disabled people. Municipalities are responsible to help those who are not self-reliant on the job market. They can provide wage cost subsidies to encourage employers to hire disabled people. They also offer sheltered work to people with such severe limitations (physically, mentally or psychic) which makes it unrealistic for them to get paid work.

(6)

I will decompose the gap in participation rates both before and after the introduction of the Participation Law. Any difference observed might be a direct consequence of the Participation Law. However, note that the Participation Law was introduced in 2015, whereas data is available up to and including 2018. It might be that the impact of the Participation Law is not yet visible in the first two or three years after its introduction. Therefore, the evaluation of the impact of the Participation Law will not be the main focus of this study, it will only be adressed in the Appendix (Section 8.1.2). The main focus of this study will be on detecting possible discrimination in the Dutch labour market in the entire period 2008-2018. Another improvement over the research done by Kidd et al. (2000) is the use of sickness absence as a productivity measure. Though it is a good measure of productivity, it is only measured for one week in the Labour Force Survey, which makes it highly unreliable. This study measures sickness absence for a month, which will make it more reliable.

(7)

Other types of discrimination, such as gender discrimination and racial discrimination, have been examined a lot in previous research. For example, Fransen, Plantenga, and Vlasblom (2012) decomposed the gender pay gap in the Netherlands in order to find out why women earn less than men. Next to that, Andriessen, Nievers, Dagevos, and Faulk (2012) investigated ethnic discrimination of immigrants in the Dutch labour market using field experiments. A lot of attention has been paid to discrimination of the disabled as well (Johnson and Lambrinos, 1985; Baldwin and Johnson, 1994; DeLeire, 2000; Kidd et al., 2000; Jones, Latreille, and Sloane, 2006). However, up to my knowledge, Malo and Pagán (2012) were the first investigating this issue with respect to the Dutch labour market. These authors investigated wage discrimination against the disabled. However, they did not pay attention to the causes of the gap in participation rates between non-disabled and disabled members of the Dutch labour force. This study pays attention to the latter as well. All in all, this research addresses a topic which is not extensively investigated before in the Netherlands. This is surprising, since 15% of the labour force in the Netherlands has a disability which effects their capacity to work (Lautenbach and Sanders, 2010). Next to that, in 2006 approximately 20 % of the labour force had a non-Dutch background (Oerlemans, 2009). These people are either born outside the Netherlands, or have parents who were born outside the Netherlands. Since the percentage of disabled workers and the percentage of workers with a non-Dutch background is quite comparable, it is important and relevant to also investigate the discrimination against disabled on the Dutch labour market. This research will contribute to the existing literature by investigating this issue.

The remainder of this paper is organized as follows. In Section 2 a literature review is presented, after which a description of the data is given in Section 3. Section 4 discusses the econometric model which is used to perform the analysis, followed by Section 5 in which the empirical findings are presented. Then, in Sections 6 and 7 a conclusion and discussion will follow. Section 8 contains some theoretical notes and tables presenting the empirical outcomes.

2 Literature Review

(8)

disabled people this exogenous characteristic might be "having a long-standing disease or hand-icap". This may be (but is not necessarily) related to unobserved factors which are influencing the wage or the probability of employment, such as the quality of experience/education or the motivation to work. These factors influence productivity and hence indirectly wage and the probability of employment. According to the definition of statistical discrimination of Phelps (1972), employers assume equal adjustments to the returns to the unobserved factors for any individual having the same exogenous characteristic. Hence, employers take this exogenous characteristic as a proxy for relevant data (e.g. quality of education/experience or motivation to work) which is not sampled. Relying on this proxy, there might be an under- or overestimation of the adjustment to returns to human capital characteristics. This under- or overestimation, which is thus based on asymmetric information, is called statistical discrimination. Statistical discrimination is one of the most common forms of labour market discrimination. However, since distinguishing different types of discrimination is impossible, I use the most general term, ’labour market discrimination’, throughout this study.

(9)

gap can be explained by human capital characteristics, whereas the other half is unexplained. Hence, they found that the disabled might suffer from significant wage discrimination, assuming that unobserved effects or omitted factors do not completely cover the unexplained part of the wage differential. DeLeire (2000) analysed the impact of the Americans with Disabilities Act (ADA) on the employment rates and wages of disabled people. Surprisingly, they found that participation rates among disabled dropped after the ADA was initialized. Also, wages did not increase for disabled workers as a consequence of the ADA.

Jones et al. (2006) analysed the British labour market using the same data as Kidd et al. (2000) did. These authors found that, especially for those with mental health problems, substantial wage differences continue to exist. They separated the unobserved effect of disability on pro-ductivity from discrimination by distinguishing between work-limiting and non-work-limiting disabilities. Jones et al. (2006) assumed a constant rate of discrimination across different groups of disabled in order to estimate the work productivity consequences of a disability. This is a rather strict assumption, since it is unlikely that there is a constant rate of discrimination across different groups of disabled. These authors looked into the consequences of the Disabil-ity Discrimination Act for both men and women, and found an (largely unexplained) improved situation for men. However, the ’penalty’ for a work-limiting disability has increased for women since the implementation of the DDA. Using an Oaxaca-Blinder decomposition they found that the part of the wage differential attributable to unobserved characteristics and discrimination dropped from 67 % to 53 % for men after the implementation of the DDA. However, for women this share of the wage differential raised from 55 % to 76 %.

(10)

2.1 Definitions of disability and productivity

All studies on this subject used different definitions of disability and productivity, since these concepts are hard to define. The first decomposition of the wage gaps between disabled and nondisabled workers was done by Johnson and Lambrinos (1985). They assume that discrimina-tion is more severe once a disability is visible. Therefore, they define people with impairments that affect the ability to communicate or that visibly alter bodily movements as disabled. They define people with a disability which is (almost) not visible, and therefore less subject to preju-dice, as nondisabled. In this way, Johnson and Lambrinos (1985) do not measure discrimination against people with an ’invisible’ disability. An ’invisible’ disability might be a long-standing disease which is not yet visible, but will be in the (near) future. It is not strange to assume that employers will rather not hire someone who will for sure get sick within a limited amount of time. Saying this, I implicitly assume that either the employer has access to the medical records of the applicant or the applicant informs the employer about his or her long-standing disease. This might however not be the case, in which case the employer has an informational disadvantage. The definition of productivity Johnson and Lambrinos (1985) use largely depends on the number of impairments and an objective measurement of their severity, but lacks the employee’s own perception of the impact of a certain impairment on his or her work activi-ties. To account for health-related impacts on productivity it is best to combine measures of self-reported productivity with objective measures of productivity (Longhi et al., 2012). Hence, Johnson and Lambrinos (1985) can improve upon their definition of productivity by including self-reported productivity measures.

Baldwin and Johnson (1994) use three different categories of disabilities: impairments, disabil-ities and handicaps. They split up their sample into three different groups according to these categories: those without impairments (non-disabled, i.e. abled), those with impairments sub-ject to little or no prejudice (disabled) and those with impairments subsub-ject to greater prejudice (handicapped). In this way, they are able to measure discrimination in a similar way as Johnson and Lambrinos (1985) did. Furthermore, they measure productivity by years of education, a number of factors representing health limitations and three work experience measures.

(11)

effects. They mention a number of day-to-day activities of which at least one has to be affected in order to classify an individual as disabled. Five categories of health problem are made in order to analyse the impact of a certain type of disability on work activities: some problems related to limbs, some to sight and/or hearing, some to skin, breathing, heart and other organs, some mental, and other. Using those five categories, Kidd et al. (2000) are able to take the severity of a disability into account. Kidd et al. (2000) measure productivity by experience, education, tenure, marital status, occupation, industry, region and two dummy variables indicating the presence and length of any period of sickness. The disadvantage of the data Kidd et al. (2000) are using is that information regarding health problems is only available for the disabled. Hence, any health problems that non-disabled individuals might encounter can not be observed.

DeLeire (2000) defines people with functional limitations as disabled. These functional limita-tions are self-reported by the respondents and might therefore contain some bias. Individuals may have financial and social incentives to either exaggerate their functional limitations or to report more functional limitations than they actually have. Contrary to Johnson and Lambrinos (1985), DeLeire’s definition of disability also includes the ’invisible’ disabilities, since ’invisible’ disabilities may have an impact on functional limitations. DeLeire (2000) measures productivity by the presence (or absence) of self-reported work limitations. Furthermore, DeLeire (2000) also includes a subjective five-point measure of health as explanatory variable in the wage equations. Since there are known to be significant differences in the way people perceive their own health and in the way they use this five-point measure to report it (Bound, 1989), this does not con-trol for health-related productivity differences accurately. However, Bound (1989) argues that self-reported measures of health cause different kinds of biases, which are working in opposite directions. Therefore, there is some chance they cancel out. On the contrary, objective measures of health also contain bias as long as these health measurements do not perfectly correlate with those health aspects influencing economic outcomes. Since the objective health measurements only cause a one-way bias, Bound (1989) conclude that using self-reported measures of health gives a more accurate picture of the impact of health on economic outcomes in most cases.

(12)

Pagán (2012) separate all respondents in three different samples. They define all respondents who answered affirmative to the question "Do you have any chronic, physical or mental health problem, illness or disability?" as disabled. Then, they compare non-disabled men and women with disabled men and women not hampered in daily activities, assuming both groups are equally productive. Subsequently, they compare the non-disabled workers with disabled workers hampered in daily activities. Using this approach, these authors relax the assumption of a constant rate of discrimination among all disabled individuals. Among other things, Malo and Pagán (2012) used experience, level of schooling, self-measured health status, occupation and profession to measure productivity.

3 Data Description

As aforementioned, data from the LISS panel is used. The LISS panel contains data from 2008 till 2018, where observations in 2014 are missing.1 _{The panel is a repeated cross-sectional study}

that yearly reports all kinds of characteristics of household members in the Netherlands. Note that the panel dimension of the data is not exploited. First of all, this is done because of the fact that the panel is highly unbalanced. An other, more important reason, will be explained in Section 4. In total 68,683 observations are available,2 _{of which some are repeated observations}

for the same individual. A substantial fraction of these observations can not be used, since important variables are missing. First of all, individuals who did not know their net income from work or did not want to report it were dropped, which results in an attrition of 5.5 % of the observations. Individuals who did not indicate whether or not they had paid work were removed as well. The attrition is approximately 12 %. Furthermore, only men and women between 25 and 65 years old are considered. The probability that someone of age 25 or older is still studying is low, such that (almost all) students are excluded. Doing this it is implicitly assumed that people of age 25 and older are not enrolled in school anymore. People of age 66 and older have reached the statutory retirement age and have no obligation to (search for) work

1_{Observations in 2014 are lacking, since the survey was not done regularly. Observations from December 2013}

and from July 2015 are available.

2

(13)

anymore. Excluding those who are not between 25 and 65 years old results in an attrition of 26 %.

Men and women working less than 12 hours per week are excluded, since different definitions of unemployment are used in the Netherlands. According to the ’national definition’, an individual is not attached to the labour market whenever he or she works less than 12 hours a week. On the other side, the ’international definition’, which is in line with the guidelines of the International Labour Organization (ILO), defines an individual who has a paid job for at least one hour per week as employed (CBS, 2014). The Central Bureau for Statistics uses both definitions. Those who work less than 12 hours a week are dropped to refrain from any misunderstandings. Note that the attrition is only 1 %, since most of these ’side jobs’ are occupied by men and women younger than 25. Employees that did not report the number of hours worked per week are excluded as well, which costs 4 % of the original 68,683 observations.

The focus of this study is on the labour force, i.e. on people that are either working or looking for a job, since these are the only people that can get wage offers. Therefore, individuals who only take care of the household, live off private means or only perform voluntary work are not considered in the analysis. This results in an attrition of 11 %. Lastly, the self-employed are dropped (only 0.18 % of the data3), since an hourly wage paid by an employer is needed to determine whether or not discrimination is present. Self-employed however ’determine’ their own wage. In the end, the sample consists of 21,525 observations. This is approximately 31 % of the original sample. Men and women are analysed separately, since it is very well possible that there is gender discrimination on the Dutch labor market (Fransen et al., 2012). By observing men and women separately, gender discrimination and discrimination of disabled can be distinguished. It appears that 51 % of the remaining observations is female and 49 % is male.

As mentioned before, I will measure both the disability itself and its impact on productivity. As Loprest, Rupp, and Sandell (1995) already stressed upon, it is important to distinguish certain limitations related to disabilities. They discussed functional limitations, specific health impairments and work limitations. Functional limitations focus on the disabilities that exclude or make it more difficult for people to perform day-to-day, i.e. normal activities. Specific health impairments influence daily time availability: they include for instance taking medications, lost

3

(14)

time due to medical appointments, hospitalization etc. Since work limitations are endogenous, they are not easy to analyze. For example, someone who is blind may be able to work as a telephonist but not as a forklift operator. Furthermore, two people with the same disability might report different work limitations, even though they have similar jobs. Hence, the ex-tent to which a disability actually makes someone disabled depends on the type of job and on personal experiences. I will use self-reported work limitations to categorize non-disabled and disabled people, since these self-reported work limitations are correlated with the type of job the individual executes. Individuals who have physical and/or mental problems which hinder their work activities, household activities or other daily tasks, are classified as disabled. Among the disabled, individuals who experience ’some’ hinder are distinguished from those experienc-ing significant hinder, such that two different groups of disabled are considered. As Longhi et al. (2012) already noted, it might be that (especially unemployed) individuals are inclined to overstate their work limitations, based on social and financial incentives. I therefore assume that individuals correctly report their work limitations. By separating all disabled individuals into groups for which there is some or significant hinder (due to their handicap or long-standing disease) in executing their working tasks, I will be able to relax the assumption of a constant rate of discrimination across all disabled people. This will be discussed in Section 4 as well.

Furthermore, productivity will be measured by experience squared4, tenure, level of education, profession, sickness absence, marital status, proficiency in Dutch and occupation. Unfortunately, there is no variable present which represents the region an individual lives. Since participation rates differ among different regions in the Netherlands, this would have been an important factor to control for.

3.1 Summary statistics

Tables 5 till 10 represent a wide range of summary statistics, in order to give an overview of the data. These summary statistics will be discussed here. For ease of reference, from now on the term ’non-disabled’ is used for those that do not experience any work limitations. The term ’disabled’ covers all individuals experiencing at least some work limitations.

4_{experience itself is not included, because of perfect multicollinearity with the level of education, which will}

(15)

Table 5 reveals that approximately 17 % of all men experience at least some work limitations. For women this share is slightly higher than 23 %. The sample of all working individuals is decomposed based on their occupation as shown in Table 6. There are no large differences between the occupational distribution of non-disabled and disabled men. A relatively high share of the non-disabled men has a job in the business sector, whereas a relatively high share of the disabled men works in the industry sector. The latter might be partly due to reverse causation. Men working in the industry sector might be more exposed to situations in which they can get disabled or injured. Table 6 shows that the occupational distribution is quite similar for non-disabled and disabled women. However, comparing men and women some differences arise. Only 8-9 % of the women has a job in the industrial sector, whereas 35-39 % of the men are employed in that sector. On the other hand, relatively more women work in the social services sector, compared to men.

The distribution of all respondents over the different types of profession is given in Table 7. Men experiencing no work limitations are on average more often executing a higher academic or supervisory function than men with significant work limitations. This pattern is less clearly present among women. Also, men and women with significant work limitations are executing unskilled, semi-skilled or agrarian work more often than non-disabled men and women. How-ever, the distribution is more widespread among men. Men are relatively more often executing higher academic or supervisory functions than women. On the other hand, relatively more men are performing unskilled, semi-skilled or agrarian work as well. Note that there might exist reverse causality between having significant work limitations and having unskilled, semi-skilled or agrarian work. The probability to get injured while executing job tasks is probably highest in this type of profession. This might also be a reason why relatively many people suffering from significant work limitations are doing unskilled, semi-skilled or agrarian work. Table 8 represents the distribution of working men and women over the private and the public sector. Two out of three male employees work in the private sector, whereas this share is approximately 50 % among women. Disabled men and women work relatively more often in the public sector compared to non-disabled men and women.

(16)

’6’ means that the respondent is university graduated. Table 9 shows that the level of education is higher for non-disabled men compared to disabled men. The same pattern can be seen among women, although the differences are smaller. The measure for experience is based on age and the years of schooling to finish the highest level of education. In this way, the regulatory years of education rather than the actual years of education followed by the respondent are measured. Education of a higher grade is measured by (an) extra year(s) of schooling. If the highest level of education obtained is primary school, it is assumed that the respondent at least followed some type of education till the age of 16. In the Netherlands, it is compulsory to follow education at least till the age of 16. Then, experience is measured by the difference between the current age of a respondent and a combination of the number of years of schooling he or she followed and the age at which he or she started following education5. Note that this only measures potential experience, e.g. it does not take into account that the respondent may have been unemployed for a while. Potential experience is assumed to be a good estimator for actual experience. Table 9 reveals that the average number of years of experience is 27 for men without significant work limitations while men with significant work limitations have 30 years of experience, on average. This may be due to the lower average education level as well as to the higher average age of the latter. The relation between work limitations and experience is less clear among women.

The proficiency in Dutch is reported as well. It is either ’very well’ or ’not very well’. More than 90 % of all men and women without work limitations speak Dutch fluently. This percentage is somewhat lower among the disabled men and women. Since this is a self-reported measure, it may contain some self-reporting bias. Sickness absence is used as an objective measure of productivity to explain wages. It is a categorical variable with five categories, where the fifth category indicates a sickness absence from work, household activities and daily activities of more than ten days in the past month. The first category represents no absence in the past month. Men without work limitations were less than 1 day absent due to sickness, on average. However, men with significant work limitations were 3 to 5 days absent due to sickness, on average. This difference must be due to the difference in work limitations, assuming that ’normal’ sickness absence, e.g. absence due to a fever, is similar among men with and without work limitations. The relationship between work limitations and sickness absence among women is comparable to what is observed in the male sample.

Less than 10 % of all men and women without work limitations have spent some time in a

5

(17)

hospital in the 12 months before they filled in the questionnaire. As expected, this percentage is substantially higher among men and women with significant work limitations. A self-reported health measure is included to explain labour force participation. It can take on three different values, ranging from zero till two. A zero implies that the respondent has a bad health status and a two indicates an excellent health status. Table 9 indicates that non-disabled men and women report a much better health status than disabled men and women. Furthermore, it can be seen that men and women are on average equally healthy according to their own reports.

The participation rate among men and women without work limitations is 94-96 %, compared to 86-87 % for those with significant work limitations. Hence, there is a substantial difference in participation rates between non-disabled and disabled people. A share of this difference will be explained by human capital characteristics and productivity limitations due to long-standing diseases and/or handicaps. However, a share of this difference might be unexplained, which is either due to unobserved/omitted factors or to discrimination of the disabled. In Section 4 this participation rate differential will be decomposed. Table 9 contains summary statistics with respect to demographic variables as well. Men with significant work limitations are on average older, which seems to support the association of disability with age. However, this association is not reflected by the female sample. The number of children is highest among those without any work limitations. It might be that those with significant work limitations are less certain about their job opportunities or do not have the financial resources to raise children, and therefore choose to have few or no children. Lastly, non-disabled men and women are on average more often married than disabled men and women. Also, relatively more working men than working women are married. Married women might choose to be a housewife more often than unmarried women.

Table 10 represents summary statistics for those variables which are only observed for working men and women. The variable tenure indicates the number of years an employee works for his or her current employer. Men with significant work limitations work on average longer at their current employer than non-disabled men. For women, the relationship is the other way around: women without work limitations work longest at their current employer. Also, men on average work two to three years longer at their current employer than women.

(18)

using the Consumer Price Index. In the empirical analysis the logarithm of the real hourly wage is used in order to measure percentage changes in real wages rather than absolute changes. However, in Table 10 the real hourly wages are used, since it only presents averages and the real hourly wages are more easy to interpret. The real hourly wage for men without work limitations is on averagee0.43 higher than for men who suffer from significant work limitations. Surprisingly, men with some work limitations, on average, hourly earned e0.35 less than men with significant work limitations. This last wage differential might (among other things) be caused by government grants for employers hiring disabled employees. The Dutch government provides a wage cost subsidy to compensate the employer for the loss of productivity of a disabled employee (Rijksoverheid, 2018). Furthermore, Table 9 indicates that the average number of years of experience is 2.4 years higher among men with significant work limitations compared to men with some work limitations. More experienced employees might have a higher hourly wage, which could also (partly) explain the wage differential between men with some work limitations and those with significant work limitations. The summary statistics on real hourly wage for women showed more unexpected results. On average, women with significant work limitations hourly earned e0.20 more than women without work limitations. Again, this is probably due to government grants.

In the next section the wage equation with the determinants of the logarithm of the real hourly wage is explicitly modeled. In this way, the wage differentials observed in Table 10 can be decomposed into explained and unexplained parts.

4 Econometric Model

(19)

an Oaxaca-Blinder decomposition (Oaxaca, 1973; Blinder, 1973). The former is due to differences in observed human capital characteristics and the latter may be caused by omitted/unobserved factors or labour market discrimination towards the disabled. Lastly, the observed participation rate differential will be decomposed in explained and unexplained components as well.

The wage equation is estimated for three different samples, since the entire sample is divided into three different subsamples based on the severity of work limitations. People without any work limitations (the non-disabled) are compared with people experiencing ’some’ and people experiencing ’significant’ work limitations (the disabled). The real hourly wage is measured using human capital variables such as the level of education, tenure and experience squared. I assume that a high level of education obtained and many years of tenure and experience indicate the capability to exercise job tasks at a certain level, which will be rewarded with a higher hourly wage. The proficiency to speak Dutch fluently is also used to explain the real hourly wage. In the majority of jobs, possessing this characteristic may ease communication with colleagues and customers and might therefore have an impact on the hourly wage received. Next to that, I will include the type of sector in which an individual works, the type of profession he or she exercises and a dummy which indicates whether or not the individual works in the public sector. Sickness absence is included in the wage equation as well, since it is a measure of productivity. An employee is probably less likely to receive a wage increase if he or she has a high sickness absence. Furthermore, some demographics such as age, the number of children and marital status are included. These covariates are assumed to have an impact on the wage negotiations. Lastly, a time trend variable will be used as well. Matrix XXXj, with dimensions i x k, contains

all these covariates. Here, i is the number of individuals in group j and the k columns represent the k characteristics included. Note that j can take on three different values: A is the set of individuals without work limitations, B corresponds to those with some work limitations and C indicates the individuals with significant work limitations. The wage equation will then look like

log(wwwj) = XXXjβββj+ j, j ∈ {A, B, C}, (1)

where log(wwwj) is a i x 1 vector containing the logarithms of the hourly wages of all individuals

(20)

belonging to the characteristics matrix XXXj. All error terms in the i x 1 vector j are assumed

to have mean zero.

The wage equation in (1) can be estimated consistently conditional on the assumption of strict exogeneity: E(j|XXXj) = 000 for each j. Since the hourly wage is only observed for employed

individuals, the strict exogeneity assumption might well be violated. The estimated coefficients of (1) suffer from selectivity bias if participation on the labour market is nonrandom. This is most likely the case, as is clarified by the following example. If wage discrimination is already severe against disabled, those people might withdraw from the labour market because of that. Consequently, these individuals will not be taken into account when comparing real hourly wages among non-disabled and disabled workers, which will probably lead to downward biased esti-mates of the wage discrimination against the disabled. Therefore, to prevent possible selectivity bias the estimated wage equation has to be adapted.

Heckman (1979) proposes a two-stage procedure in order to make this correction. First, a probit specification of labour force participation is estimated. It is necessary to use a probit specification in this two-stage procedure, which induces the second reason for not exploiting the panel dimension of the data. As is described in Section 3, the panel being heavily unbalanced was the first reason. Next to that, a panel probit model can only be estimated by random effects. However, random effects assumes no correlation between the individual effect and the explanatory variables (Clarke, Crawford, Steele, and Vignoles, 2010). This assumption is likely to fail in this context, since for example the (unobserved) motivation to work will be correlated with the type of profession. This is the second reason for not exploiting the panel structure of the data.

(21)

Also, compared to (1), exclusion restrictions are added to ensure identification. The necessity of these exclusion restrictions will be elaborated on hereafter. If employment propensity πij

exceeds a threshold of 0, the individual will be employed. If π_ij is nonpositive, the individual will not be employed in the labour market. This is represented by an indicator variable Ii:

I_i =      1, if πij > 0 and 0, if πij ≤ 0 for i = 1, ..., n_j, j ∈ {A, B, C, } (2)

Now the employment propensity πij, which is the difference between the wage an individual is

offered and his or her reservation wage, is specified. The reservation wage is a point at which the individual is indifferent between working and not working, which is unobserved. However, the reservation wage and the offered wage can be compared by observing whether an individual is employed or not. If an individual is currently working, his or her offered wage was apparently higher than his or her reservation wage. Before both types of wages can be compared, they need to be modelled as a function of certain characteristics. These are almost the same as the characteristics contained in XXXj in (1), except for the variables which are only observed for the

subsample of working individuals. The offered wage equation is specified as

log(w_ijO) = qqqijζζζj+ νij, for i = 1, ..., nj, j ∈ {A, B, C}, (3)

where qqqij is a 1 x l vector containing covariates and ζζζj a l x 1 coefficient vector. Furthermore,

νij is a scalar error term assumed to have mean zero. The reservation wage is determined as

log(wR_ij) = zzzijαααj + υij, for i = 1, ..., nj, j ∈ {A, B, C}, (4)

where zzzij and αααj are 1 x m and m x 1 vectors, respectively. Next to that, υij is a scalar error

term assumed to have mean zero. Furthermore, vector zzzij is (partly) different from qqqij. These

(22)

This characteristic will therefore, in most jobs, be rewarded with a higher wage offer compared to the wage offered to someone who is not capable of speaking Dutch fluently, ceteris paribus. Jobs which do not require the employee to speak Dutch at all might be an exception to this. However, vectors zzzij and qqqij will have much overlap. The reservation wage will also be influenced

by human capital variables such as experience squared, tenure and education. For example, a university graduated person will probably demand a higher wage than someone who only finished primary school, ceteris paribus. Next to that, the number of children, marital status and the age of an individual will have an impact on the reservation wage as well. Furthermore, it is assumed that the reservation wage and thereby the participation decision is influenced by a self-reported health status and whether or not individuals spend time in a hospital in the past 12 months.

The employment propensity π_ij can be defined as π_ij = log(wO_ij)−log(wR_ij). Then, the probability that a specific individual i is participating in the labour market is

P[πij > 0] = P[log(wOij) − log(wijR) > 0] = P[qqqijζζζj− zzzijαααj > υij− νij], (5)

where (3) and (4) are used. Now, assuming that both υ_ij and ν_ij follow a normal distribution, the combined error term τij = υij − νij also follows a normal distribution with variance σ2τ j.

This can be used to rewrite the above into

P[π_ij > 0] = P ñ τij στ j <qqqijζζζj− zzzijαααj στ j ô = Φ Ç qqqijζζζj− zzzijαααj στ j å = Φ Ç_y_y_y0 ij,•γγγj,• στ j å , (6)

where Φ is the cdf of a normal distribution, γγγj,• is the combined vector of ζζζj and αααj and yyyij,•

is the associated combined row vector of qqqij and zzzij. Note that both γγγj,• and yyyij,• are stacked

vectors. Using a probit specification the labour force participation decision can be modelled. The coefficients can be estimated by maximizing the likelihood function

L = Y i∈E ñ φ φφ Ç_y_y_y0 ij,•γγγj,• στ j åô Y i∈ ¯E ñ 1 − φφφ Ç_y_y_y0 ij,•γγγj,• στ j åô , (7)

(23)

problem of selectivity when estimating the wage equation. As already mentioned, Heckman (1979) proposed to include a selectivity variable, indicated by λ, in order to correct for the selectivity that may exist. This λ, the Inverse Mills Ratio (Heckman, 1979), is found as follows:

λij =

φ(yyy0_ij,•γγγˆ_j,•)

Φ(yyy0_ij,•γγγˆ_j,•), (8)

where φ(·) denotes the probability density function of a normal distribution and Φ(·) the cu-mulative density function of a normal distribution. The updated wage equation, which can be estimated consistently by OLS, is

log(wwwj) = XXXjβββj + ρjλλλj+ j, j ∈ {A, B, C}, (9)

where λλλj is the i x 1 vector containing all i individual Inverse Mills Ratios within group j. To be

able to draw the correct inference regarding selection, an exclusion restriction must be included. This means that one or more predictors which are included in the probit participation model must be excluded in the wage equation. This prevents high collinearity between the Inverse Mills Ratio, which is based on the predictors used in the probit participation model, and the predictors used in the wage equation. I used two exclusion restriction variables: self-reported health status and a dummy indicating whether or not someone spend any time in hospital over the past 12 months. First of all, both are likely to affect the participation decision of an individual. Self-reported health status and hospitalization will not influence the real hourly wage, as long as these variables are only related to invisible issues. However, if these exclusion restriction variables relate to a visible issue or disability, they still might have an impact on the real hourly wage. I assume that these exclusion restrictions are valid, i.e. self-reported health status and hospitalization are assumed to not be related to a visible issue or disability and therefore also not to the real hourly wage.

(24)

which is due to differences in human capital characteristics such as education, tenure, profession, occupation and experience. The unexplained part is the part of the wage differential which is not attributable to differences in human capital related characteristics. It actually represents the difference in rewards both groups receive for a certain characteristic. Using for example j = A and j = C, where A indicates the set of individuals without work limitations and C the set of individuals with significant work limitations, the OB decomposition would look like:

log(wwwA) − log(wwwC) − ( ˆρAλλλ¯A− ˆρCλλλ¯C) = explained part z }| { (XXXA− XXXC)ˆβββA+ unexplained part z }| { X X XC(ˆβββA− ˆβββC), (10)

where ¯(.) indicates the average over all individuals in a set. Note that bootstrapped standard errors have to be used (Clougherty, Duso, and Muck, 2016). The OB decomposition explains the wage differential between abled workers and a group of disabled workers. The way the wage gap is decomposed is indicated in (10). Here it is assumed that the majority group, men and women experiencing no work limitations, face a perfect wage structure and no labour market discrimination. Having performed the decomposition, it can be tested whether there possibly exists wage discrimination against different groups of disabled in the Dutch labour market. However, the unexplained part needs to be interpreted with caution. The unexplained part of the wage differential is likely to reflect uncaptured differences in tastes and productivity, in addition to the possibility of employment discrimination (Kidd et al., 2000). If the model lacks some (unobserved) variables which are influencing productivity or tastes, e.g. the (unobserved) motivation to work, then their impact will be reflected in the unexplained component of the wage differential. The unexplained component might also reflect labour market discrimination. However, since the impact of omitted factors on this wage differential is unknown, prudence is needed interpreting it.

(25)

difference between for example ˆPAand ˆPC can be decomposed as follows: ˆ PA− ˆPC = " 1 nA nA X i=1

Φ(yyy0_iA,•γγγˆ_A,•)

# − " 1 nC nC X i=1 Φ(yyy0_iC,•γγˆγ_C,•) # , (11)

which can be rewritten as

ˆ PA− ˆPC = explained part z }| { " 1 nA nA X i=1

Φ(yyy0_iA,•γγˆγ_A,•)

# − " 1 nC nC X i=1

Φ(yyy0_iC,•γγγˆ_A,•)

#! + unexplained part z }| { " 1 nC nA X i=1 Φ(yyy0_iC,•ˆ γγγ A,•) # − " 1 nC nC X i=1 Φ(yyy0_iC,•γγγˆ_C,•) #! (12)

Again, the unexplained component is likely to reflect both uncaptured differences in tastes and productivity as well as labour market discrimination (Kidd et al., 2000).

5 Empirical Findings

(26)

5.1 First stage probit specification of labour force participation

The first stage probit specification of labour force participation is estimated in order to be able to form the Inverse Mills Ratio, which is included in the second stage wage equation to correct for possible selectivity. Tables 11 and 12 present the estimates of the probit model for men and women, respectively. These estimates show that an increasing age has a positive effect on the probability of employment. However, this effect is decreasing as age increases, since the variable age2 has a negative coefficient. Note that these coefficients are insignificant and have a different sign for men with significant work limitations and disabled women. The probability of employment was at a lower level in the period 2010 to 2015 compared to the reference year 2008, ceteris paribus. This might be a direct consequence of the financial crisis. The unemployment rates from the Central Bureau of Statistics support this finding (CBS, 2018b). Those who obtained a high level of education have a higher probability of participation compared to those who only finished primary school. This result holds true for all women, able-bodied men and men with some work limitations.

(27)

probability of employment. Being married increases the chances of being in employment among men, which is in accordance with what Kidd et al. (2000) found. This effect is ambiguous for women.

5.2 Second stage wage specification

The selectivity corrected wage estimates for male and female workers are given in Table 13 and 14, respectively. Contrary to what was found by Kidd et al. (2000), I observe a positive significant Inverse Mills Ratio (IMR) for two out of three male subsamples. This implies that there has been made a correction for selectivity bias for these two subsamples. There exists positive selection, since the IMR is positive. This means that, without the correction made by including the IMR, the impacts of some covariates would be overestimated (Setzler, 2014). For men with significant work limitations an insignificant IMR is observed. This may imply that no selection exists within this sample. Since this is unlikely, it can also be that the exclusion restrictions used, self-reported health status and hospitalization, are weak. Weak exclusion restrictions cause Heckman (1979) to suffer from multicollinearity and failure of the correction for selectivity bias. This also might be the cause of the insignificant IMR coefficient. Unfortunately, the validity of the exclusion restrictions can not be checked. It might also be that the insignificant IMR is in this case due to the small sample size. The Inverse Mills Ratio is insignificant for all subsamples containing women as well. As explained above, this may imply that no selection exists or that the exclusion restrictions used are weak. The results for women need to be interpreted with prudence, since the Inverse Mills Ratio does not correct for selectivity bias. It is checked that the second stage wage equation without correcting for selectivity gives similar results for women.

(28)

higher wages than those who only finished primary school. For example, observing an employee without work limitations who graduated only from primary school and comparing him with an employee without work limitations who is university graduated an increase of 34 % in the real hourly wage is expected6, ceteris paribus. Female workers without work limitations receive a higher real hourly wage when they are university graduated compared to similar women who only finished primary school. On the other hand, for both women with some work limitations and women with significant work limitations the educational effects are almost all negative. However, most of these coefficients are insignificant. Still, it is peculiar that these coefficients have a negative sign.

Men and women who are working one year longer at their current employer are rewarded with a 0.2-0.6 % higher real hourly wage. Note that this effect is strongest among disabled men and women. Employees working in the business sector and experiencing no/some work limitations have a significantly higher wage than similar employees working in the industry sector, which is the reference group. On the other hand, male employees working in the retail have significantly lower wages (up to 17 % lower for those with significant work limitations) compared to men working in the industry sector. Women working in the retail sector receive up to 20 % lower real hourly wages than women working in the industry sector. The type of profession an employee exercises clearly has an impact on the real hourly wage. This effect is strongest for men without work limitations and women with some work limitations. As expected, men and women with a higher academic or supervisory profession (e.g. managers, directors or academic teachers) receive a higher real hourly wage than men and women performing semi-skilled or unskilled manual work, which is the reference group. A woman with a higher academic or supervisory profession experiencing some work limitations is expected to have a 43 % higher real hourly wage compared to a woman exercising semi-skilled or unskilled manual work who also experiences some work limitations.

For non-disabled male employees a high sickness absence has a negative impact on the real hourly wage. The impact of a high sickness absence on the real hourly wage is varying between the different subsamples containing women. As expected, the proficiency to speak Dutch fluently is positively affecting real hourly wages among men. Similar to what Malo and Pagán (2012) found, being married has a positive influence on the real hourly wage for all men. However, being married has a negative impact on the real hourly wage for all female workers.

6

(29)

5.3 OB decomposition of real hourly wage differentials

In Table 15, the differentials of the logarithm of the real hourly wage among the three subsamples containing men are decomposed into explained and an unexplained parts. Table 16 does the

same for the subsamples containing women. The key findings are summarized in Tables 1

and 2. The Oaxaca-Blinder decomposition (Oaxaca, 1973; Blinder, 1973) described in (10) is applied. First of all, the average wages of workers without work limitations and workers with some work limitations are decomposed. Next to that, the average wages of employees without work limitations and employees with significant work limitations are compared as well. Both groups of disabled men have a lower average real hourly wage compared to the non-disabled men. The wage differential between non-disabled and disabled male workers with some work limitations is 0.062 log points. A large part of this differential, 92 % (0.057 log points), is explained by differences in observed characteristics. This is a higher fraction than Kidd et al. (2000) found. They discovered that approximately half of the wage differential between non-disabled and non-disabled workers can be explained by observed human capital characteristics. The real hourly wage differential between men without work limitations and men with significant work limitations is 0.037 log points. More than 200 % (0.089 log points) of this differential is explained by observed human capital characteristics. This causes the unexplained part to be significantly negative (-0.053 log points), i.e. the unexplained wage differential is in favour of men with significant work limitations. Note that it is assumed that workers without work limitations have a perfect wage structure and face no discrimination. The only way discrimination against disabled with significant work limitations could still exist is if unobserved and omitted factors cover more than 100 % of the unexplained part of the wage differential.

(30)

higher the real hourly wage, since the type of profession is on average higher among non-disabled men (Table 7). The sickness absence, which is lowest among the non-disabled men (Table 9), causes 0.009 and 0.028 log points of the wage differential in favour of men without work limita-tions relative to men with some and significant work limitalimita-tions, respectively. Hence, the lower the sickness absence, the higher the real hourly wage. Being married and being able to speak Dutch fluently, both characteristics which are more frequently observed among non-disabled men compared to those with some work limitations (Table 9), contribute in favour of the non-disabled men to the observed wage differential.

Decomposition wage gap Decomposition participation rate gap No vs. Some No vs. Significant No vs. Some No vs. Significant work limitations work limitations work limitations work limitations Group 1 2.580*** (649.50) 2.580*** (758.58) 0.957*** (445.22) 0.957*** (435.72) Group 2 2.518*** (257.59) 2.543*** (134.51) 0.932*** (130.17) 0.866*** (58.91) Differential 0.062*** (5.36) 0.037* (1.89) 0.025*** (3.32) 0.091*** (6.12) Explained 0.057*** (7.29) 0.089*** (5.85) 0.006** (2.07) 0.012 (1.50) Unexplained 0.005 (0.54) -0.053*** (-2.77) 0.020** (2.26) 0.080*** (5.01)

Number of observations group 1 8,261 8,261 8,825 8,825

Number of observations group 2 1,143 494 1,243 582

t statistics in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01 Group 1 = individuals without work limitations

Group 2 = individuals with some (1st and 3rd column) or significant work limitations (2nd and 4th column)

Table 1: Main results OB decomposition and participation rate decomposition men

(31)

It is important to realize that the contribution of nominal and ordinal variables to the unex-plained part of the OB decomposition is very sensitive to the chosen reference group (Sen, 2014). Therefore, the impact of nominal and ordinal variables on this part of the wage differential has to be interpreted with prudence. However, the total unexplained part of the wage differential is robust to changes in the reference groups of nominal and ordinal variables.

The real hourly wage differential between women without work limitations and women with some work limitations is 0.019 log points (Table 2). Slightly more than 100 % of this wage differential (0.022 log points) is explained by human capital characteristics. The other part (-0.003 log points) is unexplained. The unexplained part is caused by omitted factors, unobserved factors or labour market discrimination. The real hourly wage differential between non-disabled women and women with significant work limitations is, surprisingly, negative. This implies that on average, women with significant work limitations earn more than non-disabled women. However, the wage differential is small and insignificant. Next to that, as outlined in Section 5.2, these results probably suffer from selectivity bias. Therefore, it will be tough to interpret this decomposition. Note however that the explained part of the wage decomposition is positive. In other words, based on the average human capital characteristics a higher real hourly wage is expected for non-disabled women. Therefore, either there exist omitted or unobserved factors which cause the unexplained part of the wage differential to be in favour of women with significant work limitations, or there exists labour market discrimination against non-disabled women. Since the latter is assumed to be not the case, I conclude that the former is causing the negative unexplained part of the real hourly wage differential.

(32)

The unexplained parts of the wage differential reveal that non-disabled women are on average more rewarded for a certain level of experience or education. On the contrary, women with some work limitations are on average more rewarded for exercising a specific profession. However, as already mentioned, the contributions of nominal and ordinal variables to the unexplained part of the wage differential have to be interpreted with prudence. These contributions are very sensitive to a change in the reference group.

Decomposition wage gap Decomposition participation rate gap No vs. Some No vs. Significant No vs. Some No vs. Significant work limitations work limitations work limitations work limitations Group 1 2.496*** (875.12) 2.496*** (641.71) 0.943*** (351.49) 0.943*** (335.63) Group 2 2.477*** (304.77) 2.502*** (210.07) 0.910*** (131.88) 0.860*** (68.37) Differential 0.019** (2.16) -0.006 (-0.50) 0.033*** (4.71) 0.082*** (6.51) Explained 0.022*** (2.71) 0.052** (2.26) 0.007** (2.15) 0.015 (1.57) Unexplained -0.003 (-0.26) -0.058** (-2.49) 0.025*** (3.19) 0.068*** (4.45)

Number of observations group 1 7,718 7,718 8,335 8,335

Number of observations group 2 1,466 761 1,638 902

t statistics in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01 Group 1 = individuals without work limitations

Group 2 = individuals with some (1st and 3rd column) or significant work limitations (2nd and 4th column)

Table 2: Main results OB decomposition and participation rate decomposition women

5.4 Decomposition of participation rate differentials

(33)

is present. The difference in participation rates between non-disabled men and men experienc-ing significant work limitations is 9.1 percentage points, of which only 1.2 percentage points (13 %) is explained. Again, the vast majority of the participation rate differential is unex-plained (8 percentage points, i.e. 87 % of the differential) and might be due to labour market discrimination.

The decomposition of the gaps in participation rates between women with different levels of work limitations is given in Table 2. The difference in participation rates is 3.3 percentage points between non-disabled women and women with some work limitations. A small fraction, 21 %, of this differential is explained by human capital characteristics. The remaining 79 % is unexplained and is caused by omitted/unobserved factors and labour market discrimination. The participation rate gap between non-disabled women and women with significant work lim-itations is 8.2 percentage points. Again, only a small fraction is explained by differences in for example education level, occupation, profession and proficiency in Dutch. The majority of the participation rate differential is again unexplained. These results are similar to the results found among men.

6 Conclusion

This study is one of the first using Dutch data to investigate labour market discrimination against disabled. Disabled are divided into subgroups based on the severity of their work limitations. Both men and women are subject of study. Repeated cross-sectional LISS data, collected between 2008 and 2018, is used to perform the statistical analysis. Differences in participation rates as well as in hourly wages are decomposed into explained and unexplained parts. For men, there have not been found substantial, unexplained wage differences between non-disabled and disabled workers. However, a large share of the difference in participation rates between non-disabled and disabled men is unexplained and might be due to discrimination. Similar results have been found for women. The unexplained parts of the wage differentials are not substantial. On the other hand, more than 75 % of the gap in participation rates between non-disabled and disabled women is unexplained and may therefore be due to labour market discrimination.

(34)

Dutch government should not pay too much attention to equality of wages. Ensuring disabled people to have the same job opportunities as non-disabled people is a much higher priority. Based on the policy evaluation in Section 8.1.2, it seems that the first step has been taken by the introduction of the Participation Law. However, the issue is far from being solved yet by the Dutch government.

7 Discussion

There are a number of caveats in my research, which on the other hand also provide interesting opportunities for further research in this area. First of all, a region variable is lacking. If such a variable is included in the analysis, as Kidd et al. (2000) did, one can control for unemployment differences among regions. For example, it is well known that the unemployment rate is relatively high in the North of the Netherlands (CBS, 2018a). Whether individuals live in the North of the Netherlands or in another region will therefore have a substantial impact on their job opportunities. Next to that, the use of potential experience instead of actual experience may cause some bias. Furthermore, a number of self-reported measures are used, which may suffer from self-reporting bias. First of all, self-reported work limitations may be overstated due to financial and social incentives, especially for unemployed people. Next to that, proficiency in Dutch is also self-reported and will probably contain some bias.

The OB decomposition has a few caveats as well. First of all, the unexplained part of the OB decomposition of the wage differential is very sensitive to the chosen reference group for nominal and ordinal variables. Secondly, the share of the unexplained part actually consisting of labour market discrimination is unknown. Furthermore, I assumed that those without work limitations face a perfect wage structure. This assumption is rather strict. Lastly, the exclusion restriction might not be valid. Especially the results obtained in the analysis of women point in this direction. Since self-selection for employment probably exist, this might cause the results to be less reliable. Unfortunately, there is no way to check the validity of the exclusion restriction.

(35)

Bibliography

Andriessen, I., E. Nievers, J. Dagevos, and L. Faulk (2012). Ethnic discrimination in the dutch labor market: Its relationship with job characteristics and multiple group membership. Work and Occupations 39 (3), 237–269.

Baldwin, M. and W. G. Johnson (1994). Labor market discrimination against men with disabil-ities. The Journal of Human Resources 29 (1), 1–19.

Blinder, A. S. (1973). Wage discrimination: Reduced form and structural estimates. The Journal of Human Resources 8 (4), 436–455.

Bound, J. (1989). Self-reported vs. objective measures of health in retirement models. CBS (2014). Cbs kiest voor internationale definities werkloosheid en inflatie.

CBS (2018a). Werkloosheid daalt in alle provincies. CBS (2018b). Werkloosheid lager dan voor de crisis.

Clarke, P., C. Crawford, F. Steele, and A. Vignoles (2010). The choice between fixed and random effects models: Some considerations for educational research. IZA Discussion Paper (5287). Clougherty, J. A., T. Duso, and J. Muck (2016). Correcting for self-selection based endogeneity

in management research: Review, recommendations and simulations. Organizational Research Methods 19 (2), 286–347.

Daalder, R. (2019). Hoeveel vakantiedagen hebben je werknemers recht op?

DeLeire, T. (2000). The wage and employment effects of the americans with disabilities act. The Journal of Human Resources 35 (4), 693–715.

Even, W. E. and D. A. Macpherson (1990). The gender gap in pensions and wages. The Review of Economics and Statistics 72 (2), 259–265.

Fransen, E., J. Plantenga, and J. D. Vlasblom (2012). Why do women still earn less than men? decomposing the dutch gender pay gap, 1996–2006. Applied Economics 44, 4343–4354. Heckman, J. J. (1979). Sample selection bias as a specification error. Econometrica 47 (1),

(36)

Johnson, W. G. and J. Lambrinos (1985). Wage discrimination against handicapped men and women. The Journal of Human Resources 20 (2), 264–277.

Jones, M. K., P. L. Latreille, and P. J. Sloane (2006). Disability, gender, and the british labour market. Oxford Economic Papers 58, 407–449.

Kidd, M. P., P. J. Sloane, and I. Ferko (2000). Disability and the labour market: an analysis of british males. Journal of Health Economics 19 (6), 961–981.

Lautenbach, H. and J. Sanders (2010). Nearly 1.7 million disabled people in the dutch labour force.

Longhi, S., C. Nicoletti, and L. Platt (2012). Interpreting wage gaps of disabled men: The roles of productivity and discrimination. Southern Economic Journal 78 (3), 931–953.

Loprest, P., K. Rupp, and S. H. Sandell (1995). Gender, disabilities and employment in the health and retirement study. The Journal of Human Resources 30, 293–318.

Malo, M. A. and R. Pagán (2012). Wage differentials and disability across europe: Discrimination and/or lower productivity? International Labour Review 151 (1-2), 43–60.

Ministerie van Sociale Zaken en Werkgelegenheid (2015). Participatiewet.

Ministerie van Volksgezondheid, Welzijn en Sport (2003). Wet gelijke behandeling handicap of chronische ziekte (wgbh/cz).

Oaxaca, R. (1973). Male-female wage differentials in urban labor markets. International Eco-nomic Review 14 (3), 693–709.

Oerlemans, W. G. M. (2009). Ethnic Diversity At Work: About interpersonal relations, well-being and performance in ethnically diverse organizations. Ph. D. thesis, Utrecht University. Phelps, E. S. (1972). The statistical theory of racism and sexism. The American Economic

Review 62 (4), 659–661.

Rijksoverheid (2018). Stimulering werkgevers aannemen mensen met arbeidsbeperking.

Sen, B. (2014). Using the oaxaca–blinder decomposition as an empirical tool to analyze racial disparities in obesity. Obesity 22 (7), 1750–1755.

(37)

8 Appendix

8.1 Theoretical Appendix

8.1.1 Creation of variables

This section clarifies the way in which some variables are created. The variable tenure is found by subtracting the year in which an individual started his or her current job from the year in which the survey was held. In this way, it represents the number of years a working individual works at his or her current employer. The variable experience is artificially created using the variable length of study. The latter represents the length (in years) of the study an individual has finished, including the years spent to graduate from high school. For example, an individual who is university graduated spent 4 + 6 = 10 years studying: 6 years to graduate from preparatory university education and 4 years to graduate from university. Note that this variable will be biased, since it represents the regulatory number of years to finish a type of education. The real number of years it took an individual to finish a type of education is not observed. Using the length of study and the age of the respondent, the number of years of potential experience on the labour market can be found: experience = age − (12 + length of study ), where it is assumed that people finish primary school at the age of 12. Note that potential experience will not necessarily coincide with experience, since potential periods of unemployment are not taken into account.

The nominal hourly wage is created by multiplying the monthly wage by 12 and dividing the result by the number of hours worked per week multiplied by 48, i.e.

hourly wage = monthly wage x 12

number of hours worked per week x 48. (13)

(38)

8.1.2 Policy evaluation

In order to identify the impact of the Participation Law, which was introduced in the beginning of 2015, I will compare decompositions of the participation rates of non-disabled and disabled men in 2013 and in 2018. I will not compare the decompositions of real hourly wage gaps in 2013 and 2018, since the Participation Law primarily focuses on the participation of disabled individuals. However, as already noted, it might be that the impact of the Participation Law is not yet visible two or three years after its introduction. The decomposition is similar to the one described in (12). Obviously, the time trend variable is disregarded, since individuals are observed at a specific moment in time. In Tables 3 and 4 the situations in 2013 and 2018 are described, respectively. It can be seen that the participation rate differential between non-disabled men and non-disabled men with some work limitations disappeared completely between 2013 and 2018. However, this differential was only 1.7 percentage points in 2013.

Table 3: Participation rate differentials men in 2013

No vs some No vs significant

work limitations work limitations

Group 1 0.945*** (133.92) 0.945*** (114.64)

Group 2 0.929*** (43.02) 0.825*** (14.56)

Differential 0.017 (0.72) 0.121** (2.11)

Explained 0.004 (0.46) -0.004 (-0.18)

Unexplained 0.013 (0.53) 0.124** (2.20)

Number of observations group 1 838 838

t statistics in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01

Group 1 = individuals without work limitations

Group 2 = individuals with some (1st column) or significant work limitations (2nd column)

(39)

relatively highest among the disabled men with significant work limitations. However, there still existed a participation rate gap of 8.4 percentage points in 2018. A large fraction of this gap is unexplained, and might therefore be due to labour market discrimination. All in all, it can be concluded that the labour market position of disabled men has improved since the introduction of the Participation Law. Whether this is a direct consequence of the Participation Law can not be stated with certainty.

Table 4: Participation rate differentials men in 2018

No vs some No vs significant

work limitations work limitations

Group 1 0.962*** (130.56) 0.962*** (127.67)

Group 2 0.964*** (65.60) 0.878*** (17.35)

Differential -0.002 (-0.14) 0.084* (1.66)

Explained 0.009 (1.11) 0.008 (0.31)

Unexplained -0.012 (-0.70) 0.076 (1.27)

t statistics in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01

Group 1 = individuals without work limitations

(40)

8.2 Tables

Table 5: Sample composition

Men Women

No/hardly any Some work Significant No/hardly any Some work Significant

work limitations limitations work limitations work limitations limitations work limitations

Percentage 82.8 % 11.7 % 5.5 % 76.6 % 15.1 % 8.3 %

Number of observations 8,825 1,243 582 8,335 1,638 902

Table 6: Sample composition working population by occupation

Men Women

(41)

Table 7: Sample composition working population by profession

Men Women

Higher academic/supervisory 24.2 % 16.1 % 18.4 % 10.7 % 10.0 % 9.5 %

Intermediate academic/supervisory 35.5 % 34.3 % 30.7 % 45.0 % 47.9 % 46.2 %

Mental work/skilled manual work 28.6 % 33.3 % 32.5 % 36.3 % 31.4 % 33.0 %

Un/semi-skilled manual work/agrarian work 11.8 % 16.3 % 18.4 % 8.0 % 10.8 % 11.3%

Total 100 % 100 % 100 % 100 % 100 % 100 %

Table 8: Sample composition working population public vs. private sector

Men Women

Public sector 31.2 % 31.5 % 33.4 % 46.5 % 51.2 % 51.2 %

Private sector 68.8 % 68.5 % 66.6 % 53.5 % 48.8 % 48.8 %

Total 100 % 100 % 100 % 100 % 100 % 100 %

Employment discrimination against the disabled: an analysis of men and women on the Dutch labour market

Employment discrimination against the

disabled: an analysis of men and women on

the Dutch labour market

Master’s Thesis Econometrics, Operations Research and Actuarial Studies

Supervisor: Dr. A. Postepska

Employment discrimination against the

disabled: an analysis of men and women on

the Dutch labour market

Contents

Acknowledgements

1

Introduction

2

Literature Review

3

Data Description

4

Econometric Model

5

Empirical Findings

6

Conclusion

7

Discussion

Bibliography

8

Appendix