BA Thesis
For how many hours did you get signed?
-The difference in contract hours between native Dutch and second-generation Moroccan-Dutch and Turkish-Dutch employees-
Abstract
Research on ethnic differences in labour market outcomes is mostly focussed on wage differentials and employment. This thesis introduced contractual hours as another variable to measure difference in labour market outcome between ethnicities. The contractual hours of native Dutch employees were compared to second-generation Moroccan-Dutch and Turkish-Dutch employees whilst controlling for various variables. The results indicate that there the groups differ in minimum offered hours with the difference disadvantaging the second-generation Dutch employees.
U
N I V E R S I T Y
O
F
A
M S T E R D A M
L.O. VAN BRITSOM
10270183 LOUISEVANBRITSOM@LIVE.NL DR. B. LANCEE DRS. E.M. MILTENBURG SOCIOLOGY 15 / 08 / 20172
T |
able of Content
1 Introduction
3
2 Theoretical framework
5
2.1 Theories apprehending differences in labour market position
5
2.2 Selectivity bias
9
3 Methodology and operationalization
10
3.1 Data and sample
10
3.2 Operationalization
12
3.3 Methodology
14
3.4 Statistical hypothesis
17
3.5 Visualisation of the relationships between the variables
17
4 Descriptive statistics
18
5 Analysis of Variance
21
6 Oaxaca-Blinder decomposition
21
6.1 Model description
21
6.2 Exposition of the unexplained variance
22
7 Conclusion and Discussion
7.1 Conclusion
23
7.2 Limitations and recommendations
23
8 References
25
Appendix I : R script
28
3
1
|
Introduction
The integration of non-Western immigrants was a returning topic for all political parties during the Dutch electoral campaigns in 2017. According to the VVD (People’s Party for Freedom and Democracy), the party with the most seats in government, integration can best be achieved through integration into the labour market. VVD’s argument, suggests that the integration process starts when one obtains a job. In a grand experiment, Dolfing and Tubergen (2005) have shown that Dutch citizens with a non-Western background are less likely to be employed compared to native Dutch citizens. Academics provide several explanations for the discrepancy, such as language skills, educational attainment and social capital (Leutwiler and Kleiner, 2003). However, even after controlling for these factors, many Dutch citizens of non-Western backgrounds were still less likely to be employed compared to the native Dutch (Kee, 1995). The pool of (potential) non-Western Dutch employees is very diverse and contains several large and small ethnicities. Each of the large ethnic groups has its period of immigration and history of integration. These differences resulted into each group having various differences in positions on the Dutch labour market (van Gils, 2000). For instance, people of Surinam or Antillean descent have a higher likelihood of obtaining a job compared to those of Turkish and Moroccan descent (van Gils, 2000). Between the ages of 25 and 45, 54% Dutch-Moroccans have a job (Central Bureau for Statistics, 2014). The employment level increases from 51% to 63% between first and second-generation immigrants of Moroccan descent. The same proportions are found for second-generation Turkish immigrants. Nevertheless, their participation is significantly lower than Surinamese (73%) and Antilles (82%) and native Dutch (84%).
The above-mentioned figures indicate specific problems Moroccans and Turks face to access the labour market. A subsequent step is to look into ethnic inequality whilst participating on the labour market. Differences in outcome between ethnic groups on the labour market are markedly more under researched than inequality in obtaining a job. Moreover, the research on participatory inequality on the labour market is solely focused on discrepancies in hourly wages. Dagevos (1996) and Zorlu (2002) showed that racial wage gaps based on hourly wages scarcely occurs in the Netherlands. Veenman (2010) ascribes this to strict employment regulations on remuneration which safeguards against discrimination on the basis of ethnic origin.
Hourly wages, however, are not the only indicators of inequalities on the labour market. Signed contract hours are also rampant on the labour market and they show another facade of inequality. Whilst the law regulates hourly wages, the division of contracts allow for the manifestation of employers’ personal preference when distributing the hours. Starting out on the labour market, non-Western second-generation immigrants are more likely to get a temporary contract compared to native Dutch (Central Bureau for Statistics, 2014). Despite these figure, the academic community has up until now shown no interest in considering such contracts as a measurement for unequal treatment on the basis of ethnicity.
4
To fill the gap in the academic discourse, I will research the difference in contract hours between native Dutch and second-generation Dutch Moroccan and Dutch Turkish employees. The following research question will be leading the research: To what extent can the disparity in contract hours between employees be ascribed the differences in ethnic background? The thesis will start with theories explaining the differences in labour market outcomes. The theoretical framework will be followed by the methods, sample and operationalization of the statistical analysis. Whilst exploring the field of contracts, it is necessary to first identify the difference between the groups. The analysis of variance will be used to test for differences in contract hours between the groups of employees. The analysis will be followed by the Oaxaca-Blinder decomposition for linear regression models. As this research is one of the first attempts in researching contractual differences, suggestions for future research will be provided in the discussion.5
2
|
Theoretical Framework
This research builds on a broad framework of labour market inequality. Past research found relevant variables explaining difference in outcome on the labour market. Even though the literature touches various subjects, the analysis of contract hours is not one of them. Therefore, the aforementioned studies are only relevant in the confounding variables they adduce.
The theoretical framework starts with an exposition of labour market theories that examines the factors influencing the employee’s position on the labour market. Each section will start with the general influences of the variables and will be followed by relevant concepts in explaining differences between group outcomes. When applicable, a passage will conclude with a hypothesis related to the thesis’ subsequent analysis. The finale paragraph will discuss the selectivity bias.
2.1 | Theories apprehending differences in labour market position between groups Unequal treatment of groups
Discrimination is a common explanation for the disparity in labour market outcomes. It is accepted in academia that discrimination represents “the differential treatment of individuals or groups on grounds other than productivity, which is seen as a relevant criterion in decisions on hiring, wage setting, promotion and layoff” (Veenman, 2010:1806). It is the unjustified distinction in the treatment of people that results into the disadvantage or exclusion of a group. Ethnic discrimination is the unequal treatment of an individual based on their ethnicity. It is important to highlight that discrimination in employment is the result of the employer’s attitude and behaviour towards the individual and not based on the individuals skill set (van den Cruyce, 2000).
Van den Cruyce (2000) distinguishes two types of discrimination on the labour market: income discrimination and job discrimination. Income discrimination happens when the employer disburses different wages to the employees who are equal in expertise but differ in ethnicity. Job discrimination implies that candidates or employees’ ethnicity is taken into account in the job, promotion or redundancy process. It is also possible that job discrimination could lead to difference in wages due to job position of the candidates.
While discrimination is the general explanation for the discrepancies of group outcomes, it is hard to identify its existence on the labour market. Employers are not likely to admit the difference in treatment, as it is punishable by law. Economists use the well-known method of revealing discrimination is through counterfactual analysis (Veenman, 2010). A counterfactual analysis firstly identifies all characteristics the employer deems relevant for a position, promotion or layoff. Using these traits/factors, researchers then take individuals with the same skill sets and different ethnic backgrounds and examine whether they are equally likely get the position, promotion or be laid off. If the likelihood of the two groups differs, then the difference in outcome cannot be explained by differences in skills and is therefore attributed to discrimination (see e.g. Reimers 1983; Cancio et al., 1996; Montgomery & Wascher, 1987). One of the problems with this method is that the unexplained variance could be due to other unobserved variances or
6
causes other than discrimination, which makes the method less valid and trustworthy.
Hypothesis
An unexplained disparity is expected between the native Dutch employees and second-generation Dutch from Moroccan and Turkish descent. The cause will not be attributed to discrimination as other unobserved variables might be at play. Human capital The employee can acquire skills, experience and knowledge that are relevant for a job position. This form of surplus value of the labourer is called human capital and will strengthen their position during contract negotiations (Leutwiler & Kleiner, 2003). One method of gaining the skills and experience is through education. Human capital theory argues that education is an investment made with the expectation of producing a higher income. The individual’s educational trajectory gives him the skills that increases the company’s productivity and therefore raises his value and negotiation position of the individual (Becker, 1975). Even though educational attainment influences the individual’s labour market position, it does not seem to explain differences outcome between ethnicities. Friendman and Krackhardt (1997) researched the educational returns for Asian American employees and Caucasian Americans. They found that the occupational achievement is not equal to the educational achievement. Asian American employees tend to have lower wages and career positions compared to Caucasian Americans with equal levels of education. Thus, as the educational returns are lower for the Asian Americans, the human capital theory does not suffice in explaining the difference. Therefor other factors might be at work. Control Human capital does not seem to explain the difference in labour market out. Yet still educational attainment should be controlled for as it does have an influence on the labour market outcomes. Social capital theory In theories on labour market outcome social capital is an individual’s social resource that he can access and use to attain his goals (Lin, 1999). The individual’s social capital comes from the way his live is embedded in social networks around him. Social capital increases the individual’s chances in various ways. One of which is the information that flows through the networks. The information filled networks increasing the individual’s likelihood of hearing about job opportunities and learning from others behaviour. These fields of information entangled in the network increase the chances of an employee to get the job. Social capital can be analysed by the amount of variety in such characteristics or in the ties between individual and others in their network. Most quantitative research on social capital, however, focuses on the attainment of social capital through the parent-child relationship that allows the parents’ human capital to be transferred to the children (Kao, 2004).
7
O’Neill (1990) researched the accumulation of social capital through the transfer of parental human capital, and the earning differences between black and white American men. The results indicated a narrowing of the earning differentials between whites and blacks in the consecutive generations as result of increased human capital from the side of the blacks. As over the generations black American men become more part of the American educational system, they gain the same knowledge and skills as their white counterparts. This knowledge is once again past over to their children causing wage differential due to difference in social capital to decreases. The effect of education becomes less vocal in explaining the difference in wages. Control Social capital could be of influence on the contracted hours. For instance, having a good network could present an individual with information on jobs with more contract hours. In the case of ethnic differences on the labour market, part of the difference in contractual hours could be due to lack of information on the side non-native. On the other hand, there is also a possibility of the gap in social capital between native and non-native groups to be less significant as information has been transfer from parent to child. To be certain that the difference in contractual hours can be ascribed to difference in ethnicity, the analysis needs to control for social capital Age Age is another important factor to explain the employee’s labour market position (O’Neill, 1990). If income is set as an indicator of labour market position, the general trend is that income increases as the age of the workers increases (Weizsäcker, 1988). The increase in wages can be due to a multitude of factors. For instance, according to Dutch law, the minimum hourly wage increases every year until it is set at the age of 23, to a minimum of €8,96 per hour. Furthermore, age is intrinsically linked to the years an employee is active on the labour market. Experience, contacts and knowledge gained in the field as a result of the years they work increases once again the wage. Wages are likely to increase with a promotion, which is more likely to happen at a higher age. Additionally, age is related to the type of contract. Heyma and Werff (2013) have found the number of temporary and permanent contracts differ between age groups. Young and new employees are more likely to have temporary contracts compared to the older and more experienced workers on the labour market.
Even though no comparative research on disparities between ethnicities in labour market outcome has made a distinction between age groups, age might still have dissimilar effects for ethnic groups. An annual report on integration policy in the Netherlands alludes at a difference in unemployment between ages and across ethnicities (Central Bureau for Statistics, 2014). For instance, on average, 8.3% of the native Dutch citizens were unemployed 2014. The average unemployment rate for Dutch-Moroccans was 23%. If more attention is paid to different age groups one can see that unemployment is not evenly distributed amongst the ethnicities. Focussing on the ages of 15 through 25, 34% of Dutch-Moroccan youth is unemployed, compared to only 13% of native Dutch (Central Bureau for Statistics, 2014). The age group of 25 to 45 also holds a difference in unemployed rates: 6% for native Dutch and 20% for Dutch-Moroccans. These figures indicate that age affects the labour market outcome of native Dutch and Dutch-Moroccans differently. Age is not merely a variable that one simply needs to control for in a ‘group-difference’ analysis. Even though there are no theories explaining these differences, age will be included as an explanatory
8
variable. Control Age is expected to have different effects for two employee groups and will as could result partially explain the group difference in contracted hours. Life course paradigmThroughout their lives, people encounter various transitions in their educational, professional careers and in their family life (Hareven & Massaoka, 1988). The life course paradigm sees the changes as a multilevel phenomenon triggered by changes ranging from “structured pathways through social institutions and organizations to the social trajectories of individuals and their developmental pathways” (Elder, 1994: 5).
Life course paradigm takes changes in the employee’s life as an explanation for changes in their labour market positions. One effect of the shifts in pathways is the (conscious) alteration of the structures in the employee’s life. The two most common variables indicating a phase shift in the employees’ life are partnership and parenthood. Possible alterations by the employee could be working extra hours, changing to a more flexible job or applying for a different position. For instance Kaufman and Uhlenberg (2000) found a correlation between the hours worked and having children.
Little research has been done on the effects of pathway changes in the lives of different ethnicities on their labour market outcomes. One of the few researches is Kee’s (1995) decomposition in wage differentials between men of immigrant employees and native Dutch. He found that being unmarried has negative effects on the wages of native Dutch. Unmarried native Dutch men tend to have lower wages compared to married Dutch men. The marital status did not influence the wages of employees with an immigrant background. Control The statistical analysis will control for the effects of the life course paradigm as it influences the individuals choice to be more or less active on the labour market.
9
2.2 | Selectivity Bias
Leutwiler and Kleiner (2003) and Reimers (1983) raise awareness surrounding the selectivity bias. Firstly, occupational sex-typing is the increased the presence of a specific gender in a function as a result of gender biases. Teachers, secretaries and nurses are typical examples of female oriented jobs whilst jobs like lawyers and top executives are dominated by men. As a result of the societal placement men are placed in positions of higher payment compared to women (Leutwiler and Kleiner, 2003).
Research on the selectivity bias on the basis of ethnicity could be explained as an effect of the employees’ social- economic background. Employees are likely to hold jobs of comparable positions as their parents (Laband & Lentz, 1983). In explaining this effect, Breen and Johnson (2005) have found a correlation between the parent’s level of education and labour market position on the children. Employees whose parents enjoyed high levels of education ended in high position jobs. As of yet, however, there is no exhaustive research on selectivity bias on the basis of ethnicity in the Netherlands nor on the influences the parent’s job has on the career aspirations of the child. Nonetheless, it appears that people of Moroccan and Turkish descent tend to work on a lower occupational level (van Gils, 2000). Another form of selectivity bias lies between employees and contractors. The difference between the two lies in the contractual agreement with the firm. Employees receive a fixed regular payment for work or services. The contractor has their own company and undergoes a contract with another firm to provide a service. In contrast to the employee, the contractor can outsource the labour. The wage suggestions to a wage earner may differ from the rate offered to an independent contractor with identical observed characteristics. Treating the contracts of employees and contractors equal to one another could create a bias, particularly when observed personal characteristics are non- randomly selected (Reimers, 1983:303).
10
3
|
Operationalization and Methodology
This section will entail a description of the data and justification of the sample selection. Hereafter, relevant concepts that came out of the theoretical framework are transformed into matching variables. An extensive exposition of the methodology will follow. The chapter will close with a visualisation of the relationship between the variables and the statistical hypothesis. 3.1 | Data and Sample Dataset
The second wave dataset of Netherlands Longitudinal Life Course Study (NELLS) is used for the thesis’s statistical analysis (Tolsma, Kraaykamp, de Graaf, Kalmijn, Monden, 2014). The dataset collects information on central sociological themes through face-to-face interviews and online surveys. The NELLS dataset is designed for longitudinal studies. The first wave of surveys was published in 2011. This paper will use the second wave of surveys collected in 2014.
NELLS dataset is especially useful for the upcoming analysis as it actively oversampled people of Turkish and Moroccan descent. For the first wave, 35 Dutch municipalities were selected with a stratified sampling technique. The local authorities were asked to draw three random samples: [1] inhabitants aged 15-45 who were born in Morocco or whose father or mother was born in Morocco; [2] inhabitants aged 15-45 who were born in Turkey or whose father or mother was born in Turkey; [3] inhabitants aged 15-45 excluding those belonging to group 1 and 2. The age group of the second-wave ranged from 15 to 50 as it is a continuation of the first data collection. The response rate of the second wave was 75%. Among the sample of the native Dutch 83% responded. The response rate of Moroccans and Turks were lower, respectively 62% and 65%. Sample The sample consists of male employees between 18 and 50 years old born in the Netherlands and from full Dutch, partially Turkish or Moroccan descent, who work at least 12 hours a week. The following demonstration will expound on the motivations behind the sample selection per variable. Hours worked The Central Bureau for Statistics considers respondents to be part of the labour market, when they work a minimum of 12 hours per week. The same guideline will be used in the thesis. Labour market status The contractual hours are the working hours agreement between employer and employee. As stated in section 2.2, the agreement between self-employed and wage earner might be subjected to the selectivity bias. To lessen the chance of a higher unexplained variance, only wage earners under direct employment are selected.
11
EthnicityRespondents are categorised as native Dutch when they and both of their parents are born in the Netherlands. Respondents are classified as second-generation Moroccan-Dutch when they are born in the Netherlands and at least one of their parents is from Morocco. The same categorization applies for second-generation Turkish-Dutch. There is full awareness of the weaknesses in the operationalization of ethnicity. Various characteristics can serve as ethnic identifiers, whilst the relevance depends on the researcher’s choices the operationalization. The choice for the above-mentioned categorization lies in that it is in line with the research and data of the Central Bureau for Statistics.
Only Moroccan-Dutch and Turkish-Dutch have been selected instead of all ethnic groups. This has been done as the two ethnic groups show more similarities in their positions on the Dutch labour market compared to other ethnic groups (Crul & Doomernik, 2003). Furthermore, their period of immigration to the Netherlands, integration policies, cultural background and the views of other groups in society on Moroccan and Turks are factors that bind the groups together (Hansen, 2003). Whilst acknowledging there are differences between the Moroccan and Turkish citizens in the Netherlands, these are believed to be not significant enough to distinguish the two groups in the analysis. The selection of only second-generation Dutch employees is a method of factoring out variance in contract hours that is due to the devaluation of foreign diplomas and the lack of knowledge of Dutch language and customs (Leutwiler and Kleiner, 2003; Veenman, 2010; Kee, 1995; Støren & Wiers-Jenssen, 2010). Kee’s(1995) research shows that the second generation’s knowledge of the Dutch language no longer has a significant influence on their wages as it did for the first generation. Furthermore, as most second-generation employees gained their education in the Netherlands, the devaluation of their skills as a result of a foreign diploma is also no longer a relevant variable.
The selection of second generation Dutch-Moroccans and Dutch-Turks was for practical reasons. NELLS dataset oversampled people of Moroccan and Turkish descent. A great benefit of the oversampling is the lowered possibility of too small sample sized for the statistical analysis and will increase the validity. Gender The selection of merely men in the sample was also done for practical reasons. The draft reduces possible unexplained variance in the model that could have been the result of triple interaction effects with gender, ethnicity and the other variables. Furthermore, it limits the effects of the occupational typecasting put forward in section 2.2. Finally, selecting one occupational field would have reduced the effects of occupational ethnicity typecasting. Societal placement could have an effect on the placement of the respondents and, therefore, their contractual working hours. It has been a conscious decision to not limit the research to a specific field. The sample has already decreased as explained above. If a selection were made based on the occupational field, the sample size would become too small for a statistical analysis.
12
3.2 | Operationalization
The variables can be divided into three groups: explanatory and control variables and the response variable. The explanatory and control variables are the operationalized concepts put forward in the theoretical framework (see section 2). The explanatory variables help to test if these concepts influence the contracted hours, the response variable. Control variables are put in place to control for their effects in the sample on the response variable. They are not expected to be significant influencers on the dependent variable. However, due to the large sample size differences between the groups and in order to reduce possibility of ‘omitted variable bias’, these variables do need be included and can be kept as a constant. Response variable The dependent variable in the analysis is the ‘contractual hours’. NELLS’s survey contains an open-ended question on the number of hours the employee has signed for. In the Netherlands, the working time directive limits the hours of work to 60 hours a week. Answers in the survey exceeding the maximum of 60 hours are considered faulty and are removed from the sample. There was only one case of a native Dutch respondent being excluded for this reason. The contractual hours of the Dutch Moroccans and Dutch Turks follow the normal distribution. The contractual hours of native Dutch are left skewed and therefore violate the assumption of normality. An anti-log transformation of the contractual hours could have solved this problem. However, it has been a conscious decision to not apply the transformation. Firstly, the Kolmogorov-Smirnov test indicates a normal distribution of the contractual hours. Additionally, the central limit theorem states that with a sufficiently large sample size (n>100), the sample mean will be approximately equal to population mean.
Explanatory variable
The ethnicity of the employees is the explanatory variable. These ethnicities are native Dutch and the second-generation Dutch Moroccan and Dutch Turks. NELLS constructed the categories based on the definition of Central Bureau for Statistics. A respondent is considered native-Dutch when both parents are born in the Netherlands. To be considered a second-generation Dutch citizen one needs to be born in the Netherlands and have at least one parent born in Morocco or Turkey. The native Dutch are in the statistical analysis the reference category. Control variables Age, Level of education, function, partnership and parenthood are all control variables. The first control variable is the employee’s age. Age is a numerical variable measured in years. Educational attainment is measured in NELLS as highest level of education one has received. Educational attainment functions as an operationalization of human capital theory. It is originally an ordinal variable with thirteen categories. In the original data educational attainment contained the following category: ‘acquisition of a foreign diploma that is incompatible to the Dutch educational system’. Since the diplomas of such respondents are - to respondent’s own account - incomparable to the Dutch educational system, they form an anomaly in the statistical analysis. On that
13
account respondents belonging to this category were excluded from the sample. There was only one case of a native Dutch employee that befitted the exclusion. Educational attainment is re-categorised into educational level. Educational level consists of three categories. The first category (‘low level of education’) consists of the respondents who had no education, went to primary school or finished VMBO or MAVO. The second category (‘mid-level of education’) consists of people who completed HAVO, VWO or MBO 1 through 4. Finally, to be considered part of the high educational level, one needs to have completed HBO or higher. Educational level is an ordinal variable. The reference category will be group of the mid-level education.
Generally, the employee’s social capital is expressed through the accumulation of the resources he has gotten from his network (Lin, 1999). Nevertheless, what ought to be part of such resources is up for debate. Various researches use different variables as an expression of social capital. These variables could be the parents’ education, the family structure, resources in the household, parents’ educational aspirations for their children, parents’ contact with schools and parents’ interaction with their children, and any other resource that is tied to educational outcomes (Portes, 1998). These concepts do encompass the employees’ social capital, yet other forms transmitting social capital could still be at play.
Instead of using an accumulation of variables associated with social capital, Mcguire (2000) suggests using the expression (indirect measure) of social capital, such as the employees’ hierarchical location in the occupation. The executive function within the organisation is an expression of the hierarchical location. According to Mcquire (2000), requiring such a position is likely to be the effect of social capital when other variables such as education and age are controlled for. Furthermore, controlling for executive function lessens the variation in contract hours due to occupational typing. This thesis’s analysis will also use executive function as an expression of social capital. In an executive function one of the employee’s assignments is to manage and supervise others. The variable consists of two categories: ‘having executive functions at the company’ and ‘having no executive functions at the company’. The reference category will be ‘having no executive functions at the company’. The final two variables, partner and parenthood, are dummies. The question ‘do you have a partner in the household who receives an income?’ will be used. This variable is preferred over the marital status of the employee, as it is more open to alternative forms of relationships. The reference category is not having a partner who receives an income.
Unfortunately, NELLS does not have a direct question on whether the children still live at home. Respondents are asked whether the oldest child lives at home. It might be the case, however, that the oldest one has left and the younger siblings are still living at home. Due to a lack of a better measurement, the variable oldest child living at home was used for the test. This will, nevertheless, influence the 𝛽 − 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 of this variable. The reference category is not having children.
14
3.3 | Methodology
As put forward in the theoretical framework, variations in contractual hours could be the result of a multitude of factors. Several achieved and ascribed qualities could influence the employee’s contractual hours. The linear regression (Ordinary Least Square, OLS) seems an obvious method for determining whether these qualities in fact are significantly influencing the contract hours. A hypothetical OLS equation would look as follows:
𝑌!"#$%&!$ !!"#$= 𝛼 + 𝛽!"#$%&'()%* !""!#$%&$"+ 𝛽!"#+ 𝛽!"#$"%+ 𝛽!"#$%"&+ 𝛽!"#$%&#+ 𝜀 (𝟏)
The 𝛽 − 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 reveals the effect that ethnicity and the other variables have on the contract hours. If the 𝛽 − 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 for ethnicity is significant, one could say ethnicity attributes to the difference in contract hours.
In the OLS equation (1) ethnicity is part of the equation. However, this might not be a correct view on the role of ethnicity. O’Donnel (2008) has shown that variables such as education, age and gender, have different effects for the hourly wages for employees of different ethnicities. Lee (2011) has also found that educational attainment and age are not likely to affect the native-born and immigrants in the same manner, and therefore result into different outcomes between the groups. The OLS regression does not take this into account. It merely produces one regression equation and thus one effect for ethnicity. It would be possible to make all variables interact with ethnicity. The equation, however, would become very messy and less reliable.
The Oaxaca-Blinder (OB) decomposition provides a solution.1 The core idea of the OB decomposition is to explain the distribution of the outcome variable in question by a set of factors that vary systematically between ethnicities (O’Donnel et al., 2008:148). The OB decomposition abides to the same assumptions as a linear regression. It creates equations for both groups and compares these equations. As a result the problem of one long messy equation of the OLS is rectified. The relationship between the equations looks as follows:
The Regression lines
𝑁𝑎𝑡𝑖𝑣𝑒 𝐷𝑢𝑡𝑐ℎ: 𝑌!"#$%&!$ !!"#$= 𝛼 + 𝛽𝓍!"#$%&'()%* !""!#$%&$"+ 𝛽𝓍!"#+ 𝛽𝓍!"#$%&'#+ 𝛽𝓍!"#$%&#+ 𝜀 (𝟐. 𝟏)
𝑆𝑒𝑐𝑜𝑛𝑑 − 𝐺𝑒𝑛.: 𝑌!"#$%&!$ !!"#$= 𝛼 + 𝛽𝓍!"#$%&'()%* !""!#$%&$"+ 𝛽𝓍!"#+ 𝛽𝓍!"#$%&'#+ 𝛽𝓍!"#$%&#+ 𝜀 (𝟐. 𝟐)
Depending on the prefered decomposition, 𝓍 is the effect of the variable without the group specific effects. The 𝛽 − 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 is group’s special intercept for a specific variable.
The difference
Δ𝑌!"#$%&!$ !!"#$= Y!"#$%& !"#$!− Y!"#$%& !"#"$%&'(# !"#$! 𝟐. 𝟑
1 O’Donnel (2008) gives the most comprehensive explanation of the decomposition. His work has been used as the main reference in this piece. Hlavac (2014) article has been used as to complement O’Donnels work.
15
Thus, the gap in mean outcome of contract hours, Y!"#$%& !"#$! and Y!"#$%& !"#"$%&'(# !"#$! , is equal to
Y!"#$%& !"#$!− Y!"#$%& !"#"$%&'(# !"#$! = 𝛽!"#$%& !"#$! 𝓍!"#$%& !"#$!− 𝛽!"#$%& – !"#.𝓍!"#$%& – !"#. 𝟐. 𝟒
The full exposition of the equation is as follows: To see if there is a significant difference between the groups, the OB decomposition creates two models. It compares situation ‘contract hours without the influence of ethnicity’ to situation ‘contract hours with the influence of ethnicity’.
The first model is the base level in which is assumed that there is no effect of ethnicity on the contract hours. The regression coefficients produced in this model are the regressions coefficients that would emerge if ethnicities were treated equally (Hlavac, 2014). Thus, it tests how age, education, partnership, parenthood and function affect the contract hours. The equation of the base level is comparable to equation 1 without of course ethnicity. The base-level model functions as a references category and is captured in 𝛽!"#. The second model creates the equations 2.1 and 2.2. As can be seen in the equations, a special group intercept (𝛽) for each variable is formed and made to interact with each variable (𝓍). As demonstrated in equation 2.5, the gap in hourly wages is composed of two portions: explained and unexplained variance. Part 1 shows the portion where the dependent variables clarify the cross-group gap in hourly wages. This is the variation that can be seen as the effects of all the variables, except for ethnicity.
Crucial to the analysis of this research are portions 2a and 2b, which present the unexplained variance. As the reference coefficient (𝛽!"#) is the state in which both groups are equals, the sub-components (2ab) measure the part of the mean difference in outcomes that originates from favouring the native Dutch employees (𝛽!"#$%& !"#$! − 𝛽!"#) and the part that disfavours the second-generation Dutch employees (𝛽!"#− 𝛽!"#$%& – !"#.). The unexplained variance is the difference in contracted hours due to systematic different distribution of contracted hours between the two groups, which cannot be explained by a difference in the distribution of other characteristics (education, parenthood, etc.) among the workers. The 𝛽 − 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡𝑠 in equations 2.1 and 2.2 capture the unexplained variance and hold the benefits or disadvantages of being part of a specific group.
The unexplained variance will give three different results. Firstly, it discloses if ethnicity on its own has a significant
16
effect on the unexplained variance. Secondly, it controls for the interaction effect of ethnicity with the control variables. And finally, it will indicate if there is significant difference in the 𝓎-intercepts of the groups. The 𝓎-intercept is 𝛼 in equations 2.1 and 2.2. A significant difference in intercepts signifies a difference in minimum offered contract hours. This thesis will follow the above-described two-folded OB decomposition. An adjustment in weights the coefficients is needed as the sample sizes differ per ethnic group. Neumark (1988) correction uses the pooled variance to regulate for the sample size differences. As a similar method is used accounting for group differences in a two-sample T-test, this method was chosen for this research. The significance level (𝛼) has been set to 0.05. Besides the two-folded decomposition, the three-folded decomposition could have been chosen. The methods differ in decomposition formulation, but essentially follow the same equations and produce similar outcomes. The choice of decomposition is to the researcher’s taste (O’Donnel, 2008). The justification in choosing the two-folded decomposition lies in its convenience. The two-folded decomposition is easier to understand for those who are new in the field of OB decompositions. 2
Even though the Oaxaca-Blinder decomposition carries less noise compared to a linear regression model, there are some valid points of critique. Firstly, the decomposition shows the eventual inequality and not the process of how the disparity came about (Veenman, 2010). To fully understand the gap in contractual hours additional methods are needed.
Furthermore, the original decomposition compares the means of the groups whilst controlling for other attributes. The averages may not be reflective of the results of employees at other earning percentiles. To understand the earning dynamics of employees away from the mean is therefore not possible with the classical decomposition (Giaimio et al., 2010). Heinze (2010) and Machado and Mata (2005) suggest a decomposition of using quantile regression. It implements the decomposition across the whole distribution. Unfortunately, I am unable to apply this method due to a gap in knowledge and R studio program functions.
The classical Oaxaca-Blinder decomposition is currently most encompassing method for measuring inequalities whilst retaining the relevant group specific qualities and I am able to understand. Consequently, aware of the critical remarks, the Oaxaca-Blinder decomposition will be used for the upcoming analysis of contractual inequality.
2 The three-folded decomposition was applied in the previous version of the thesis.
17
3.4 |Statistical hypothesis Three statistical hypotheses are drafted from the theoretical framework. Hypothesis 1 H0 Ethnicity has no significant effects on the contracted hours. (𝛽!"#$%& !"#$!− 𝛽!"#$%& – !"#.= 𝛽!"#.) Ha Ethnicity has a significant effect on the contracted hours. (𝛽!"#$%& !"#$!− 𝛽!"#$%& – !"#.≠ 𝛽!"#.) 𝜶 0.05 Hypothesis 2 H0 Ethnicity has no significant effects on groups’ intercepts of the contracted hours. (𝛼!"#$%& !"#$!= 𝛼!"#$%& – !"#.) Ha Ethnicity has a significant effect on groups’ intercept of the contracted hours. (𝛼!"#$%& !"#$!≠ 𝛼!"#$%& – !"#.) 𝜶 0.05 3.5 | Visualisation of the relationships between the variablesFig 1. Relationship-tree of the variables
18
4
|
Descriptive Statistics
In order for the statistical analysis to be meaningfully interpreted, it is necessary to explore and visualise the data in boxplots. In the following section, a summary of he sample and the relevant variances of the variables will be laid out. The sample consists of 611 respondents. Despite NELLS’s active oversampling of respondents from Moroccan and Turkish descent, merely 63 Dutch Moroccan and Turkish employees remained after the selection procedure. There is a clear overrepresentation of native Dutch employees, with a sample size of 548 respondents.
Table 1 depicts all the variables information per group: mean, standard deviation, confidence interval and range. The confidence interval has been set to 95%. A confidence interval of 95% indicates that the true population mean is expected to lie within these intervals.
Even though the NELLS’s first wave of surveys has set the age limit to 45 years, the current sample does contain some respondents over the age of 45. This is due to its longitudinal qualities of the data collection. A respondent in the first panel, who was to turn 46 in 2011 after the survey, would turn 50 in the second panel. The Dutch Moroccan and Dutch Turkish employees are on average 32 years old and range from 19 through 47. The native Dutch employees’ ages are on average 39 with a minimum of 19 and a maximum of 50. Furthermore, in both groups the ages are relatively equally dispersed. There is a slight overrepresentation of 40 to 50 year olds in the sample of native Dutch.
The dispersion of educational attainment is equal across groups. All respondents have completed a form of education. Both groups have an overrepresentation of the midmost category: MBO 1, 2 and 4. Furthermore, it is worthy to note that the sample of Dutch-Moroccans and Dutch-Turks does not contain employees with a master degree or higher. However, the frequency of such graduates is also very low in the large sample of native Dutch. Furthermore, the distribution of educational attainment among the groups in the sample seems to be consistent with the distribution of previous research conducted by Central Bureau for statistics (2001). In comparison with their publication in 2001, the highly educated native Dutch employees in the NELLS sample seem be overrepresented by 7%. Table 1 shows an interesting difference in the functions when it comes to the percentages. Nearly 40% of the native Dutch employees in the sample practice a profession with executive aspects. This is in sharp contrast with the sample of second-generation employees in which merely 20% have executive power. The disparity might be due to the sample selection or it might be a correct representation of societal placement. Regardless, it is important to control for this variable. On average employees receive contracts of 33 hours with a standard deviation of 11.9 hours. As can be seen in table 1, the employees from the second generation have an average contract of 28.89 hours a week whilst native Dutch sign on average contracts for 34.06. Thus, there is a difference of 5.18 hours. Due to the large standard deviations of the means; it is more useful to look at the dispersion in a boxplot (see figure 2). The visualisation indicates proximity of medians; 36 and 38 hours. The boxplots’ shapes, however, are very different from one another.
19
The contractual dispersion between native Dutch seems left skewed and leptokurtic. The data appears leptokurtic since most data points are centred in a relatively small range around the mean. The negative skewness is visible in the small range of data being at the end of the scale. The boxplot having such a shape indicates a small range of offered contractual hours, namely between 30 and 40 hours a week.
The boxplots of the employee groups are very differently shaped. Whilst the native Dutch quartiles are more or less equal in size, the generation employees’ quartiles are not. For instance, the fourth quartile of the second-generation of employees is equal to the maximum. This means that 25% of Moroccan-Dutch and Turks sign for 40 hours a week. On the other hand, the top 25% of native Dutch employees have contracts of 40 to 45 hours. Even more interesting is the range of contracts belonging to the first quartile. The range of native Dutch employees lies between 30 and 35 hours. The contract of the bottom 25% of the second-generation employees lies much lower: 0 to 15 hours. The disparity becomes more evident when zooming in on the 0 hour contracts. Native Dutch receive so-called ‘0-hour contracts’, but it appears to be more common in the group of the second-generation Moroccan and Turkish-Dutch employees Figure 2 does indicate a disproportionate distribution of contract hours. It is, however, not clear whether the difference is significant and if it is the result of the ethnicity or due to the other influencers stated in section.
20
Native Dutch Second-generation Dutch 𝜇 or % 𝑠. 𝑒. 𝐶. 𝐼 𝑟𝑎𝑛𝑔𝑒 𝜇 or % 𝑠. 𝑒. 𝐶. 𝐼. 𝑟𝑎𝑛𝑔𝑒 Contract Hours per week 34.06 11.36 33.5 – 35.1 0 - 60 28.89 15.9 24.8 – 31.5 0 - 40 Age 38 8.2 37.3 – 38.8 19 - 50 32 7.2 30.9 – 33.9 19 - 47 Educational attainment MBO 3
and 4 - - Primary School - Doctorate
MBO 3
and 4 - - Primary school – University Bachelor Educational level reference mid-level
Low level High level 21.2% 17.7% - - - - - - 20% 8.3% - - - - - - Function Executive function 38.8% - - - 20.6% - - - Partner Yes 72.9% - - - 55.6% - - - Children Yes 49.7% - - - 41.3% - - - N 548 N 63 Table 1 Exposition of the variables of interest between native Dutch and second-generation Dutch employees. Fig. 2 Comparison of the contract hours between native Dutch and second-generation Dutch employees.
21
5
|
Analysis of Variation
The analysis reveals a significant difference in contracts between the two groups of employees, F(1) = 14.98, p = .000. Native Dutch seem to sign for higher hours than the employees from Moroccan and Turkish descent. Although this model explains 19.3% of the variance, it does not encompass the (interaction) effects of other relevant and confounding variables. Nevertheless, the ANOVA does provide a reason to delve more into the discrepancy in contracts.
6
|
Oaxaca-Blinder decomposition
The relevant findings will be discussed in the following section. Firstly, a coarse exposition will be done on the general model. Subsequently, the results of the unexplained variance will be explored and linked to the statistical hypotheses. 6.1 | Model description Table 2 contains both the mean and the accompanying standard errors of the explained and unexplained variances. The total variance is equal to the hourly mean difference of 5.18 hours put forward in the descriptive statistics. Four out of the five-hour gap is explained by a cross-group difference in dependent variables. Only one hour remains unexplained. Thus, most of the gap is the result of a difference in the distribution of the characteristics among the groups. Since the employee’s age is the only significant variable in the explained variance, this is the characteristic that is unequally distributed among the groups and explains the gap. Older employees at the company tend to be native Dutch and get as a result higher contracted hours compared to the young employees, whom most are from Moroccan or Turkish descent. Thus, most of the variation in contracted hours is explained by the age difference and not ethnicity.
The decomposition in R does not provide the p-value of the (un)explained variances. However, the difference between native Dutch and second-generation employees seems to be insignificant, considering the standard deviation. The standard deviation is higher than the mean of the unexplained variance. A high standard deviation is an indication of the data being more spread apart and, thus, showing less of a relationship between variables. The standard error of the unexplained variance indicates a 64% likelihood that the population mean difference lies between -0.63 and 2.74 hours. In this range one standard deviation away from the mean surpasses the 0 – hour baseline. The range containing Effects (in hours) 𝝁 𝒔. 𝒆. Explained 4.15 0.89 Unexplained 1.03 1.71 Total 5.18 - Table 2 The contribution of the explained and unexplained variance of the average mean difference in contract hours
22
values under the 0 – hour difference and the standard deviation being higher than the sample mean make the probability of an insignificant unexplained variance more likely.
6.2 | Exposition of the unexplained variance
𝑌!"#$%&'$ !!"#$!"#$%& !"#$!
= 8.34 + 0.62!"#+ 1.11!"# !"#$%&'()− 0.37!!"! !"#$%&'()+ 1.37!"#$%&'#+ 2.58!"#$%&# − 0.61!"#$%& (𝟑)
𝑌!"#$%&'$ !!"#$!"#$%&!!"#.
= −9.56 + 1.33!"#− 6.50!"# !"#$%&'()− 11.14!!"! !"#$%&'()+ 7.62!"#$%&'#+ 1.46!"#$%&# − 4.98!"#$%& (𝟒)
Following the theoretical framework, ethnicity is expected to partially explain the gap in contractual hours. This appears to be the case to a certain degree. Ethnicity does not have a significant effect on the unexplained variance in contract hours, T(1) = - 0.76, p = .45 . The first null hypothesis (𝛽!"#$%& !"#$!− 𝛽!"#$%& – !"#.= 𝛽!"#.) in section 3.4 still stands. There is, on the other hand, a significant difference in 𝓎-intercepts of between the employees, T(1)= 2.5, p = .01. The gap in minimum contract hours is 17.9. As becomes apparent from regression equation 3, native Dutch employees receive 8 hours more than the minimum of base-level model (𝛽!"#). The minimum contracted hours of the second generation employees are about 10 hours under the base-level model. The native Dutch employees have a favourable position when it comes to minimum contract hours compared to Moroccan-Dutch and Turkish Dutch employees. The second null hypothesis is rejected, 𝛼!"#$%& !"#$!≠ 𝛼!"#$%& – !"#..
Unexplained variance
𝛽
𝑠𝑒
𝑃
a
17.90
9.59
.01**
Age
-23.46
8.56
.00***
Education
reference mid-levelLow level
High level
1.43
0.87
0.85
0.79
.54
.46
Function
executive1.86
0.97
.54
Partner
Yes0.84
1.87
.00**
Parent
Yes0.84
1.57
.20
Ethnicity
-1.09
1.43
.45
N 611
Table 3 Decomposition of the unexplained variance
23
7
|
Conclusion and Discussion
7.1 | Conclusion
Academic research on ethnic differences in labour market outcomes mostly focuses on wage differentials and employment. This thesis introduced another labour market outcome that has been studied much less so far: the division of contract hours among employees. This thesis investigated the extent to which the disparity in contract hours among employees is affected by ethnic background. The contractual hours of native Dutch employees were compared to second-generation Moroccan-Dutch and Turkish-Dutch employees whilst controlling for various variables. The decomposition showed a five-hour difference between the groups of employees of which one hour remained ‘unexplained’.
At first sight, ethnicity does not seem to be related to the one-hour difference between the groups. There is no significant difference in the means of the contractual hours of the different ethnic groups. There is, however, a significant difference of 17.90 hours found in the 𝓎-intercepts of the groups. The native Dutch employees have a favourable position when it comes to minimum set contract hours compared to Moroccan-Dutch and Turkish Dutch employees. This finding may indicate that an employee’s ethnic background influences his or her minimum contractual hours.
This finding, however, comes with a note of caution. We should be wary of the results of the unexplained variance. The standard deviation is higher than the mean of the unexplained variance. A high standard deviation is an indication of the data being highly spread and, thus, showing less of a relationship between variables. This increases the likelihood that the unexplained variance is insignificant and its interpretations are not relevant.
7.2 | Limitations and recommendations
The results of the decomposition are hard to embed into the existing literature, as the unexplained variance is likely to not have effects on the gap in contractual hours. Further research is needed to provide conclusive insight into the relationship between ethnicity and contractual hours. Additionally, several critical remarks can be made on the research that was conducted for this research. The following section will discuss the limitations of the analysis and suggestions for future research.
The foremost limitation of the thesis is the classical OB decomposition. The classical decomposition uses the mean in order to make a comparison between groups. This became a problem in the conducted research: the ANOVA showed a significant difference in means between the groups, but in the decomposition the groups’ ethnicities effects no longer explained this gap. This could have been expected, when you would take a closer look at the boxplot in figure two. The medians of both groups are quite similar to each other, whilst the dispersion of contractual hours is extremely different. The boxplot even shows the difference in minimum contractual hours.
24
(2005) suggest a decomposition by using quantile regression. This implements the decomposition across the whole distribution. Unfortunately, this method could not be applied in this thesis due to limited time and because the R studio program functions were unavailable, but this would be highly recommended for future research on contractual hours.
The second limitation of the research concerns the sample. Firstly, the underrepresentation of second-generation Moroccan-Dutch and Turkish-Dutch employees in the sample increases the chance of a type II error. This group consisted only of 63 respondents. This small sample size makes it difficult to provide a proper estimation and modelling of the population group. The sample could not be a good representation of the entire population, reducing the validity of the statistical analysis. It is therefore possible that the null hypothesis has been falsely accepted. Secondly, the analysis has not taken the type of occupation into account. To get a more comprehensive view with less influence of confounding variables on the dispersion of contracts hours between ethnicities, it is recommendable to do more exhaustive research on only one or a small number of occupations in future research
