University of Groningen - Faculty of Economics and Business
Job Search Behaviour in Social Assistance
A research about different types of job search behaviour and its relation withthe characteristics of the unemployment assistance recipients in Groningen.
Thesis supervisor: prof. dr. J.P. (Paul) Elhorst Co-assessor: prof. dr. J. (Jakob) de Haan
Name: Joëlle Soepenberg
Email address: joellesoepenberg@gmail.com Student number: s2564262
2
Preface
This thesis for the master International Economics and Business (IE&B) at the Faculty of Economics and Business (FEB) builds on a previous thesis I wrote for the master Economic Geography (EG) at the Faculty of Spatial Sciences (FSS) from the University of Groningen.
Both theses use the data from an experiment in Groningen involving social assistance recipients. However, they focus on different aspects of the primary data results. The thesis for EG focused on norm behaviour, which are actions according to certain norms and values of social assistance recipients. The relationship between norm behaviour and the characteristics of recipients were studied as to what extent norm behaviour influences the search for an occupation and the potential outflow to employment.
This thesis focusses on job search behaviour of social assistance recipients. In particular, different types of job search behaviour, which could be extracted from scientific literature, are examined. In addition, the influence of personal characteristics on job search behaviour is studied in more detail. Two new variables could be added to the existing database. These variables are the existing intervention groups within the experiment in Groningen and the neighbourhood the participant lives in. They are used as additional independent variables in the regression analyses.
So, several types of job search behaviour are distinguished, and characteristics of recipients are studied in the context of job search behaviour, including the neighbourhoods and intervention groups of the individuals. Focussing on job search behaviour of social assistance recipients provides the opportunity to examine the potential outflow to employment based on different types of job search behaviour. The chance of outflow to employment for those who are not motivated to search for an occupation is lower than the chance of finding a job for those who are motivated. However, current legislation related to social assistance does not distinguish between different types of job search behaviour. Therefore, it is proposed in this research to revise this legislation by incorporating differences in job search behaviour of recipients to provide for a more personalized approach in social assistance. This is a contribution to the scientific literature about social assistance.
Although this thesis builds on the master thesis for EG, the theses can be read independently of each other. However, because the sources of primary and secondary data contain overlap, several tables and figures are used for practical reasons in both theses. If that is the case, a reference to the first thesis is noted below the corresponding figure or table.
3
Abstract
The aim of this research is to examine the relation between job search behaviour and social assistance in Groningen. Three key findings can be deduced. First, there are five different types of job search behaviour in Groningen: autonomous motivation to search, controlled motivation to search, amotivation to search, autonomous motivation not-to-search and controlled motivation not-to-search. Second, the degree of job search behaviour of recipients is significantly higher for those who are high-educated, male, non-Dutch, participants in intervention groups 2-4, short-term unemployed and young. Third, the results indicate that there is no significant spatial pattern of job search behaviour in Groningen.
4
Index
1. Introduction ... 5
1.1 Societal Relevance ... 5
1.2 Academic Relevance ... 5
1.3 Empirical Data from Groningen ... 5
1.4 Main Questions and Structure ... 6
2. Theoretical Framework ... 8
2.1 Social Assistance ... 8
2.2 Job Search Behaviour ... 10
2.3 Independent Variables ... 12
2.4 Intervention Groups... 13
2.5 Conceptual Model ... 14
3. Research Design ... 15
3.1 Acquisition of the Data ... 15
3.2 Variables in the Data ... 18
3.3 Econometric Model ... 20
3.4 Constraints within the Methodology ... 21
4. Results of the Empirical Analysis ... 22
4.1 Data Characteristics... 22
4.2 Results of the Factor Analysis ... 25
4.3 Implementation of the Regression Analysis ... 28
4.4 Results of the Ordered Logit Regressions ... 29
5. Conclusion of this Research ... 34
5.1 Main Findings ... 34
5.2 Reflection ... 36
6. Literature ... 37
7. Appendices ... 40
7.1 Question from the Survey used in this Analysis ... 40
5
1. Introduction
1.1 Societal Relevance
Last October, the NOS (the Dutch Broadcasting Foundation, 2018a) announced that the city of Rotterdam cancels the obligation to apply for a job when an individual is older than 60 years. At the same time, de Volkskrant (a Dutch newspaper, 2019) noted that since 2015 the national government successfully provided better fiscal arrangements for single parents, if they would work part-time instead of remaining unemployed. This resulted in 15,000 single parents out of social assistance and thereby, 200 million euros each year for the Dutch municipalities. This fits the trend of several municipalities in the Netherlands, which experiment with the current system of social assistance. While some political parties on a national and local level advocate keeping current policies, others propose to radically change the way people receive benefits while being unemployed (NOS, 2018b).
1.2 Academic Relevance
This debate also occurs in the scientific literature, in which it is known as the ‘deservingness’ debate (Van Oorschot, 2000 in Van der Waal et al. 2010). In short, ‘who deserves what’ is discussed thoroughly, which makes it an interesting research topic. Building on that, should policy implications be based on carrot or stick methods or a combination of both? Van der Klaauw and Van Ours (2013) define carrot incentives as bonuses when an unemployed individual finds an occupation (again). They define stick incentives as punishments when someone does not comply with the rules and regulations related to social assistance. The suitable method is different for everyone receiving social assistance. Therefore, scientific research could point out which policy measures are useful to implement in social assistance and which are not useful or too expensive to execute on an individual level. This is discussed in chapters 4 and 5.
1.3 Empirical Data from Groningen
In Groningen, the municipal government is currently conducting a two-year experiment related to the current system of social assistance between 2017 and 2019. This name of this experiment is: “Bijstand op Maat”. The participants in this experiment are divided into six groups: four intervention groups, one control group and one reference group. The aim of the experiment is to provide the opportunity to social assistance recipients to experience more freedom and trust, rather than a reciprocal approach (Edzes et al., 2018). The data of this experiment are used in this thesis.
6 1.4 Main Questions and Structure
Both in practice and scientific research, social assistance and the outflow to employment are important themes, because they occur in various media items and because there is a lot of literature related to these topics. The aim of this thesis is to examine job search behaviour among different unemployment assistance recipients in Groningen. This research focusses on different types of job search behaviour, the characteristics of those receiving social assistance and on a potential spatial pattern of job search behaviour. This topic is chosen, because there is a gap in scientific literature about job search behaviour of social assistance recipients in the Netherlands at the moment. However, insight in this matter could aid the outflow to employment and the well-being of social assistance recipients by adjusting current legislation. Therefore, the following research question and sub-questions are defined:
What is the job search behaviour of social assistance recipients, based on data from the experiment related to social assistance in Groningen?
1. What different types of job search behaviour occur in Groningen and to what extent do these types overlap with types of job search behaviour defined by Vansteenkiste et al. (2004)? 2. What is the relation between characteristics of social assistance recipients and their job search behaviour?
3. To what extent is there a spatial pattern of job search behaviour in Groningen?
The structure of this research is as follows. Chapter 2 provides a theoretical framework for this research and chapter 3 contains the research design, including the acquisition of data. Access to secondary administrative microdata of Statistics Netherlands, hereafter abbreviated to CBS (Centraal Bureau voor de Statistiek) is required on an individual level to answer the research question and sub-questions. These data are collected by the Department of Research and Statistics in the municipality of Groningen and thereafter made available via CBS. In addition, primary survey data from the conducted experiment in Groningen are needed, to determine different types of job search behaviour. This primary survey data could be combined with the microdata from CBS on an individual level. That means that there is a lot of information available for each recipient of unemployment assistance, who is also a participant in the experiment of Groningen.
7 Chapter 4 provides the results of the empirical analysis. The answer to the first sub-question is that this research demonstrates that the types of job search behaviour in Groningen do overlap with those determined by Vansteenkiste et al. (2004). This means that there are five types of job search behaviour in Groningen: autonomous motivation to search (41% of the participants), controlled motivation to search (11% of the participants), amotivation to search (17% of the participants), autonomous motivation not to search (13% of the participants) and controlled motivation not to search (10% of the participants). The answer to the second sub-question is that a higher level of job search behaviour is common for those who are younger and short-term unemployed. Also, for those who have a higher level of received education, for individuals who are not Dutch, for males and for those who participate in intervention groups 2, 3 and 4. The answer to the third sub-question is that the neighbourhood in which recipients live does not play a significant role in the self-perceived job search behaviour.
8
2. Theoretical Framework
This chapter discusses the most important concepts in this research. In the first chapter it is stated that this thesis focusses on job search behaviour within social assistance. Therefore, paragraphs 2.1 and 2.2 define these two concepts. Thereafter, the focus shifts to what influences the job search behaviour of social assistance recipients. Because of that, sections 2.3 and 2.4 respectively discuss personal characteristics of participants considered in the analysis and the intervention groups of the experiment in Groningen. At last, paragraph 2.5 provides a conceptual framework summarizing the key concepts as discussed in this chapter.
2.1 Social Assistance
This section focusses on the main theme of this research, which is social assistance, also known as welfare benefits or unemployment assistance. An individual can receive social assistance when he or she meets four requirements (Het Juridisch Loket, 2019a):
1. This person is Dutch or has a valid residence permit; 2. This person lives in the Netherlands;
3. This person has insufficient income or capital to live on; 4. This person is not entitled to other unemployment benefits.
Social assistance benefits are part of the Participation Law (the Dutch rules and legislations related to the current system of unemployment assistance), which means that municipalities in the Netherlands need to pay these benefits. Before the Participation Law was implemented, the national government in the Netherlands was responsible for executing these payments. After decentralising this obligation, the municipalities became responsible. In addition, the municipalities need to ensure the well-being, social participation and if possible, reintegration of the recipient into the labour market. It is noteworthy to mention that this was the exact reason why municipalities asked to experiment with the Participation Law, given that the current legislation could not sufficiently improve the three aspects stated above, for which they are responsible (Edzes et al., 2018). According to Van Ryn & Vinokur (1992), individuals who find reemployment, experience that their level of well-being are restored to levels found among the individuals who are employed stably. So, this is an argument for interventions that promote effective and persistent job-search behaviour to restore their levels of well-being.
9 UAs and UBs are a solution for labour market failure in the sense that they provide a financial safety net in case someone becomes unemployed and does not have (enough) savings left. Furthermore, UAs and UBs cover against risks related to unstable incomes, whether that is due to an occupation with a varying wage or due to fluctuations in the economy (Boeri & Van Ours 2013, p. 21). At last, UAs and UBs could increase the efficiency of the matching process between the unemployed and available jobs in the labour market. If the unemployed could not use UAs or UBs, they would search for occupations which are easily accessible. However, that means that they would not put their energy in finding a job which is more high-productive and thereby, more difficult to obtain. To increase the productivity of firms, UAs and UBs could provide the opportunity for jobseekers to search for more difficult, but also more productive occupations (Acemoglu & Pischke, 1999; Acemoglu et al., 2000; Marimon and Zilibotti, 1999). In addition, there is an increasing divergence between old EU democracies and new EU democracies, also known as European transition countries. This divergence over time is visible in figure 1. Social assistance benefit levels in old EU democracies increased from 7000 US dollar PPP in 1995 to 9000 US dollar PPP in 2005. At the same time, they stagnated below 4000 US dollar PPP for European transition countries and new southern democracies.
10 Therefore, Nelson states that all EU member states should redevelop their policies regarding social assistance to provide benefits above the poverty threshold. This is in line with the third intervention in the experiment in Groningen, in which participants can earn up to €199,- more than their monthly amount of social assistance benefits. Second, he argues that the gap between old and new EU democracies should be closed.
2.2 Job Search Behaviour
This section discusses the dependent variable. This variable is job search behaviour. Van Ryn & Vinokur (1992) state that the major determinant of job-search behaviour is ‘intention’ or in other words motivation. They argue that 26% of job search behaviour is significantly caused by this level of intention.
Vansteenkiste et al. (2004) distinguish between internal and fully internalized external motivation. This means respectively whether someone opts for a job or is forced to by an external factor. Their research is based on the self-determination theory (SDT). This theory advocates that an unemployed man with internal or autonomous motivation rather than external or controlled motivation feels less obligated by others or by himself to find a job. Paradoxically, he performs better and more persistent with a higher level of well-being, while searching for an occupation (Deci & Ryan, 2000; Vallerand, 1997).
Vansteenkiste et al. (2004) argue that within the SDT there are three main reasons to search for an occupation: autonomous job search, controlled job search and amotivation to search. In addition, there are two main reasons not to search for an occupation: autonomous motivation not-to-search and controlled motivation not-to-search. This classification is in line with the classification of Deci & Ryan (2000). The five types of job search behaviour and examples of reasoning related to each type of search behaviour are provided in table 1.
Type of Job Search Behaviour Example of Reasoning
Autonomous Motivation to search “I’m searching because work is personally meaningful for me”
Controlled Motivation to search “I’m searching because I am in need of the money”
Amotivation to search “I’m not really looking for a job, because I do not
feel competent to find employment”
Autonomous Motivation not-to-search
“I’m not really searching for a job because I give priority to alternative activities such as volunteer work”
Controlled Motivation not-to-search
“I’m not searching because others expect me to do an alternative activity or because I would feel like a bad person if I did not attend to other tasks”
11 Autonomous and controlled motivation are two different types of motivation. Autonomous motivation is based on the internal motivation of the individual (not) to search for a job, while controlled motivation is based on the compliance from an individual to meet the demand from him or herself or from others (not) to find an occupation. Contrasting to autonomous and controlled motivation, there is amotivation. An individual is amotivated, when he or she has a lack of intention and motivation to search for a job. This person often feels helpless, because he or she thinks that getting employed is impossible.
The main results of the research by Vansteenkiste et al. (2004) are visualized in table 2. The first six rows indicate the influence of independent variables on motivation (not) to search. Individuals with an autonomous motivation to search have a significant positive job commitment, also their job aspirations are positive and therefore, they are optimistic in terms of job search behaviour often stating that they expect to find an occupation soon. Participants with controlled motivation to search also have a significant positive job commitment, but they also have financial concerns. This leads to a significant level of extrinsic job aspirations, while they are not significantly intrinsically motivated. Individuals with an amotivation to search also have financial concerns and at the same time, these people do not expect to find a job soon, which makes them pessimistic in terms of job search.
Participants with a motivation not to search do not expect to find a job soon as well. In addition, they are also pessimistic in terms of job search, however they are less pessimistic than those with amotivation to search. All in all, Vansteenkiste et al. (2004) describe that only autonomous motivation to search leads to significant positive job search behaviour. In this research, I examine whether these five types of job search motivation and their corresponding results could be applied to Groningen to see if the results are similar. This is done in section 4.2.
Motivation to search Motivation not-to-search
Autonomous Controlled Amotivation Autonomous Controlled
Financial concerns 0.02 0.37*** 0.24** -0.12 0.08 Job commitment 0.43*** 0.48*** -0.02 -0.40*** -0.05 Intrinsic job aspirations 0.23*** 0.11 -0.1 -0.03 -0.12 Extrinsic job aspirations 0.19** 0.33*** 0.09 -0.18 0.02 Expectation to find a job in the near future
0.24*** -0.08 -0.44*** -0.20*** -0.29***
Job search
optimism 0.33*** 0.1 -0.46*** -0.22*** -0.35***
12 In terms of policy implications, Card et al. (2017) argue that the short-term unemployed benefit more from sanction programs and job search assistance, while the long-term unemployed benefit more from programmes to improve their human capital, such as training. In the experiment in Groningen, most individuals are unemployed for a longer period. Therefore, it is important to stress that often their human capital needs to be improved, before proper job search assistance can be provided. However, the level of self-perceived job search behaviour estimated by the participants could indicate to what extent they are motivated to search for an occupation. This is examined in chapter 4.
2.3 Independent variables
In sections 2.1 and 2.2, social assistance and the dependent variable job search behaviour of these research are discussed. Now, the focus shifts towards the independent variables in the analysis. In this study, age, education level, ethnicity (whether someone is Dutch or non-Dutch), gender and the neighbourhood (in Dutch: wijk) of social assistance recipients are used in the analysis as independent variables. The effect of each of these independent variables is tested separately in the regressions on job search behaviour.
There are a several scholarly contributions about personal characteristics and how they affect job search behaviour. These are discussed in the following paragraphs. Autor (2015) argues that due to innovations in technology, skill-biased technological change occurs nowadays. That means that some jobs, mostly those which require middle-skilled workers will disappear or substantially change over the next few years. This could lead to a shortage of jobs for a group of people with a specific level of received education. Specifically, these education levels are the largest in the data sample of this research, as noted in table 6: 97 participants finished primary and lower secondary education (vmbo, havo-,vwo-onderbouw, mbo1) and 331 participants finished upper secondary and post-secondary non-tertiary education (havo, vwo, mbo). Autor states that growth in employment levels does occur within non-routine tasks in which workers have a comparative advantage as opposed to machines. Therefore, the schooling and job training possibilities need to adapt to this type of technological change. Only then, newcomers to the labour market and those who are unemployed can adapt to these changes as well. Autor concludes with the fact that distribution of economic resources because of automation, rather than scarcity will be an issue the upcoming decades. This is interesting to consider for policies related to unemployment assistance.
Kroft et al. (1989, p.18 & p.19) argue that personal characteristics influence the ability to cope with unemployment and they could determine if and if so, when a recipient could get back to work. For example, it is easier to find an occupation for younger recipients of social assistance than older recipients, because they are more flexible. In addition, it is more difficult to find a job for those who are older, because they are more expensive to hire.
13 unpaid domestic work could make it easier for females to cope with unemployment. At the same time, males might find it harder to adjust to unemployment, because they are often the primary earners in their families.
The recipients who are not Dutch also have a comparative disadvantage on the labour market, especially when they do not speak Dutch fluently. Also, their network might consist of other internationals who do not have a strong network in the Netherlands either. However, according to Fieldhouse (1999), who conducted a case study in London, the ethnicity of a social assistance recipient has a smaller influence on job search behaviour than the neighbourhood he or she lives in. He argues that this could be due to the negative characteristics of his or her neighbourhood, but also because the individuals who are already disadvantaged concentrate in these areas.
2.4 Intervention Groups
There are four intervention groups in the experiment in Groningen. The first group is free of obligations, the second group gets intensive help, the third group can earn up to €199,- more than their social assistance benefits per month. At last, the fourth group can choose between the three interventions mentioned above. These four intervention groups are designed to examine the effect on the level of unemployment in the experiment of the municipality of Groningen (Edzes et al., 2018).
Unfortunately, the effect of the intervention groups on job search behaviour could not be studied over time between 2017 and 2018, because there is no data about job search behaviour in 2018. Probably questions about job search behaviour will be asked again in the follow-up questionnaire, which will be conducted in autumn 2019 for the experiment in Groningen. However, this research can examine the initial effect of the intervention groups on the job search behaviour of the participants. Section 4.4 shows which intervention groups in the experiment contain high levels of job search behaviour. Taking the intervention groups into account is important, because it could have important implications for policies. Interventions that show a significant positive change in job search behaviour over time are more likely to be used in future policies regarding social assistance benefits and general labour market issues. For example, if the second group, which gets intensive help does not show an increase in job search behaviour over time, this intervention is not likely to be implemented in labour market policies in the future.
14 like in the previous paragraph, it would save a lot of money if this intervention would not be implemented.
All in all, it is important to note that the effects of the intervention groups should be verified against different types of social assistance recipients. In other words, it is crucial that every policy intervention is linked to the corresponding recipient correctly. For example, if someone has an autonomous motivation to search, he or she does benefit most from the first intervention: an exemption of all obligations, while being unemployed. Also, for some recipients, receiving intensive help might be helpful to improve their well-being or it might help them to find a job, while it is not a solution for others. However, this aspect raises two issues. First, should individuals choose themselves which interventions are the most applicable to them? And second, how to deal with those who are amotivated to search, those who have an autonomous motivation not-to-search or a controlled motivation not-to-search?
2.5 Conceptual Model
This section summarizes the main concepts, which are discussed in this chapter. These concepts are the conceptual framework for the data analysis. The personal characteristics, duration of social assistance and intervention group of the recipient determine the individual job search behaviour of the recipient. This is visualized in figure 2. The personal characteristics considered in this analysis are: age, education level, ethnicity (whether someone is Dutch or non-Dutch), gender and his or her neighbourhood. It is assumed that the duration of social assistance has a negative effect on job search behaviour, because it is difficult to find a job after being long-term unemployed due to a loss of human capital and skills and because they must cope with labour market and personal hindrances (Krueger et al., 2014).
15
3. Research Design
This chapter focusses on the methodology of this research. Section 3.1 explains the acquisition of data to proceed with the variables in the analysis in section 3.2. Section 3.3 provides an econometric model for the analysis and thereafter in section 3.4, the constraints of the data are discussed. Chapter 4 provides the results of the data analysis.
3.1 Acquisition of the Data
As explained in section 1.3, “Bijstand op Maat” (“Social Assistance customized”) is a project from the municipality, in which they aim to decrease the level of unemployment and improve the health and well-being of the citizens involved. Table 3 shows that there are approximately 11,000 individuals, who receive unemployment assistance in Groningen. This is 5.4% of the total population of Groningen, which is 202,961 in September 2017 (CBS, 2019). Furthermore, table 3 indicates the division of recipients into the target group, intervention groups, control group and reference group.
Description Number of individuals
Total number of UAs recipients in Groningen
11,000
Target group 8,744
Randomized allocation to groups 1,711
Group 1: Exemption 183
Group 2: Intensification 144
Group 3: Extra earnings 153
Group 4: Choice Exemption Intensification Extra earnings 73 58 58 Control group 222 Total 891 Reference group 146
16 1,711 individuals were randomly allocated to groups with their consent. The ‘entrance ticket’ to the experiment was filling in this survey. 891 participants filled in the survey and they were divided into five different groups. Four intervention groups and one control group receive the following:
- Intervention 1: exemption of all obligations while receiving social assistance;
- Intervention 2: an intensification of personal assistance through individual coaching; - Intervention 3: the opportunity to earn extra money while receiving social assistance
benefits, up to €199,- each month;
- Intervention 4: a choice between one of the three options mentioned above;
- Control group: the control group fills in the questionnaire but remains within the rules and legislation of the Participation Law. The individuals in this group receive unemployment assistance. As a reward for filling in the survey three times, each participant in this group receives €120,- in total.
In this research, two main sources of data are used. In the first place, there is primary data available from the experiment in Groningen. Data about the participants in this experiment are collected using a questionnaire (Edzes et al., 2018). It includes questions related to among other things health, job search behaviour and well-being. Some questions consist of a Likert scale, which contains five options to answer ranging from ‘completely agree/ always’ to ‘completely disagree/ not at all’. If (some of) the interventions turn out to be successful, there could be a change in national policy related to social assistance in the Netherlands.
The focus in this research is job search behaviour, and therefore the question in the survey related to this topic is the primary source of data. The question used from this survey is taken up in the first appendix of this thesis (in Dutch). The participants filled in the questionnaire in the autumn of 2017, which was the start of the experiment. The total duration of the experiment is from September 2017 until September 2019. In this period, three questionnaires are conducted on a yearly basis. Currently, two surveys have been executed.
The second source of data is microdata from CBS. These are data, collected at the individual level about citizens in the Netherlands. For example, the education level, ethnicity and the neighbourhood of citizens is included in the data, but also information about the history of labour market participation of individuals who take part in the experiment. The data of the experiment are directly linked to microdata from CBS, using a code with eight random symbols or characters instead of the citizen service number (BSN). Thereby, the data are pseudonymized, leading to the protection of the privacy of the participant.
18 Figure 3 visualises the relative amount of UA recipients, before the start of the experiment in 2017 (CBS, 2014). The map at the top of figure 3 shows the percentage of UA recipients per neighbourhood as a share of the total population. The map in the middle shows the location of Groningen in the Netherlands. At last, the map at the bottom indicates which neighbourhoods have a higher share of UA recipients than the average share of recipients in Groningen compared to the total population.
The average share of individuals receiving social assistance benefits is 6% in the municipality of Groningen. However, in some neighbourhoods this share of recipients is up to 18% of the total population. All in all, CBS data from 2014 show that the share of UA recipients contains an unequal spatial distribution.
3.2 Variables in the Data
The collection of data for the analysis is discussed in the previous section. Therefore, we can now proceed with the indicators that are used in the analyses in the next chapter. Table 4 shows the indicators used in the regression analysis, including their source and variable type. Table 5 shows the type of data and their transformation for the ordered logistic regressions, if needed.
Indicator Data source Survey Question/ Variable name in
Microdata
Variable type
Age Secondary
data (CBS)
LEEFTIJD2 (age at 1st November 2017, which was the start date of the
experiment) Independent Duration of Social Assistance Secondary data (CBS)
MAANDBIJSTAND (number of months receiving social assistance continuously until June 2017) Independent Education Level (finished) Secondary data (CBS) OPLNIVSOI2016AGG4HBMETNIRWO (highest level of education received in 18 categories) Independent Ethnicity Secondary data (CBS) GBAGEBOORTELAND (country of birth) Independent Gender Secondary data (CBS)
GBAGESLACHT (gender; male or female) Independent Intervention group Secondary data (CBS)
groep (number of intervention group) Independent
Neighbourhood Secondary data (CBS)
RINPERSOON (neighbourhood code) Independent
Job search behaviour Primary Data (experiment) Question 16 Dependent
19 All the independent variables are based on secondary data, the microdata of CBS. These are objective data, mostly collected on a regular basis. The education level and the intervention group were added separately to the dataset, which makes them available for the sample only. This contrasts with the data on for example age and gender, because these are available for the sample, population and the entire municipality of Groningen. This is visualized in figure 4 and discussed later. It is important to note that the duration of social assistance is taken into account up until June 2017. So, the influence of the experiment itself on the duration of social assistance is not measured in this research.
Table 5 shows that the categorial, independent variables need to be transformed. The different categories in these variables are used separately to test their isolated effect in the analysis. The values of the dichotomous variables need to be changed to 0 and 1. The ratio variables do not require transformation.
As opposed to the independent variables, the dependent variable job search behaviour is based on primary data, the survey of the experiment in Groningen. Job search behaviour in this questionnaire is self-perceived job search behaviour by the participants in the experiment rated on an ordinal scale. This variable does not require transformation, because the ordered logistic regressions used in this analysis require a dependent, ordinal variable (Hill et al. 2012, p.607). All sub-questions related to job search behaviour are tested separately in the analysis.
Indicator Type of data Transformation needed to use this variable in the regression
Age Ratio No transformation needed.
Duration of Social Assistance
Ratio No transformation needed.
Education Level (finished)
Categorical 18 categories are reduced to six categories of education, based on grouping of the CBS (2017a). These categories are:
1. Education unknown
2. (11) Less than primary and primary education (basisonderwijs);
3. (12) Primary and lower secondary education (vmbo, havo-,vwo-onderbouw, mbo1);
4. (21) Upper secondary and post-secondary non-tertiary education (havo, vwo, mbo);
5. (31) Short cycle tertiary, bachelor or equivalent (hbo-, wo-bachelor);
20
Indicator Type of data Transformation needed to use this variable in the regression
Ethnicity Binary Transformation to non-Dutch = 1 and Dutch = 0.
Gender Binary Transformation to male = 1 and female = 0.
Intervention group
Categorical The variable ‘groep’ is split in five different groups. Four intervention groups and one control group.
Neighbourhood Categorical Out of 16 different values for neighbourhoods, 4
are excluded because these neighbourhoods contained less than 10 people. All the other neighbourhoods are included as follows: 1. Centrum (Binnenstad)
2. Oud-Zuid (Schilders- en Zeeheldenwijk) 3. Oud-West (Oranjewijk)
4. Oud-Noord (Korrewegwijk) 5. Oosterparkwijk
6. Helpman en omgeving (Herewegwijk en Helpman) 7. Zuidwest (Stadsparkwijk) 8. Hoogkerk en omgeving 9. Nieuw-West (Noorddijk) 10. Noordwest 11. Noordoost 12. Noorddijk en omgeving Job search behaviour
Categorical No transformation needed.
Table 5: Transformation required to use the indicators in the regressions
The variable neighbourhood requires special attention. From almost all individuals in the sample, the neighbourhood they live in is known. However, some individuals live in a neighbourhood with less than ten UA recipients. Due to privacy issues, these neighbourhoods are left out of the analysis. Also, individuals with unknown neighbourhoods were left out of the analysis. In total, 27 individuals needed to be withdrawn from the analysis. The neighbourhoods which contain less than ten UA recipients were ‘Zuidoost’, ‘Meerdorpen’ and ‘Meerstad en anderen’.
3.3 Econometric Model
The econometric model for the ordered logistic regression analysis is equal to the econometric model for ordinal logistic regressions as formulated by Hill et al. (2012, p. 608). That is the following model:
21 in which Y = ordinal dependent variable job search behaviour, βn = the slope parameter also known as the effect of the independent variables Xn, Xn = independent variables and ε = error term. There are three assumptions of the error term, which are the same as the assumptions of the error term for the simple linear regression model. They are defined by Hill et al. (2012, p. 47) as:
1. ε is expected to be zero, which means that Y = β1* X1 + β2*X2 + … + βn*Xn;
2. The variance of ε is similar to the variance of Y, because the assumed effect of the independent variable on the dependent variable is zero;
3. There is no covariance between any combination of two errors e1 and e2. This model is further specified and applied to this research in the next chapter.
3.4 Constraints within the Methodology
The conducted experiment is a Randomized Controlled Trial (RCT), which means that the individuals were randomly divided into the intervention groups. If a RCT was not conducted, a self-selection bias could occur, which would lead to results that do not reflect the real situation regarding the average duration of social assistance and the outflow to employment. However, there could still be a small bias, because only those who saw the benefit of the experiment for them ended up participating.
There might be omitted variables in this analysis, which could explain the variance of job search behaviour. An example of this is the variable networking behaviour, studied by Van Hoye et al. (2009). Unfortunately, there are no data about networking behaviour of UA recipients in Groningen. Also, there might be reversed causality between the duration of social assistance and job search behaviour. Currently, it is assumed that the duration of social assistance influences job search behaviour. But job search behaviour could also influence the duration of social assistance. For example, when an individual is not motivated to search for an occupation, he or she probably does receive social assistance longer.
22
4. Results of the Empirical Analysis
This chapter focusses on the results from the data analysis. Descriptive statistics of the indicators in the analysis are provided in section 4.1. The first analysis of this research is a factor analysis. This is executed in section 4.2. Thereafter, the regression analyses are executed. The preparation of the variables is done in section 4.3 and section 4.4 provides the results of the regressions.
4.1 Data Characteristics
In section 3.2, the indicators in the analysis are noted and explained. This section continues by providing the descriptive statistics of all indicators except the intervention group. This is because the descriptive statistics of the intervention groups are already noted in table 3 and discussed in section 3.1. Table 6 shows the descriptive statistics for the other indicators, which are independent variables in the analysis. The numbers are given for the sample used in the analysis, the total population of UA recipients, who could participate in the experiment and the entire population of the municipality of Groningen. Unfortunately, the education level is not available for the UA recipients’ population and the entire population in Groningen. In addition, there is no information about duration of social assistance for the total population of Groningen. However, most UA recipients in Groningen are either located in the categories sample and/ or population.
Table 6 indicates that most individuals in the sample finished upper secondary and post-secondary non-tertiary education. Because only individuals between 27 and 64 could participate in the experiment, the average age of the participants is much higher than the average age within the municipality. However, the difference between the average age in the sample and population is small. There are relatively more male UA recipients and there are also more male participants in the experiment. But the difference with the entire municipality in terms of gender is small. This contrasts with the ethnicity of the population. While only one in four people is non-Dutch in Groningen, this is one in three for the sample and close to one in two for the population of UA recipients in Groningen. In other words, the share of non-Dutch citizens among UA recipients is higher than the share of non-Dutch citizens in the entire municipality. However, this share of non-Dutch individuals is smaller for the sample than for the population, so there are relatively more Dutch citizens involved in the experiment compared to the population of UA recipients in Groningen.
23
Sample Population Groningen
Variable F % F % F %
Age: 0-15 years 0 0 0 0 24,589 12
Age: 15-25 years 0 0 0 0 48,066 24
Age: 25-45 years 402 46 3,835 47 61,584 30
Age: 45-65 years 462 53 4,251 53 43,344 21
Age: 65 and older 0 0 0 0 25,053 12
Duration of Social Assistance: 1 month -
5 years 565 65 5,032 62
N.A. N.A.
Duration of Social Assistance: 5 years -
10 years 189 22 1,805 22
N.A. N.A.
Duration of Social Assistance: more than
10 years 109 13 1,249 15
N.A. N.A.
Education level unknown 50 6 N.A. N.A. N.A. N.A.
Less than primary and primary
education 104 12
N.A. N.A. N.A. N.A.
Primary and lower secondary education 97 11 N.A. N.A. N.A. N.A.
Upper secondary and post-secondary
non-tertiary education 331 38 N.A. N.A. N.A. N.A.
Short cycle tertiary, bachelor or
equivalent 175 20
N.A. N.A. N.A. N.A.
Master, doctoral or equivalent 107 13 N.A. N.A. N.A. N.A.
Ethnicity: Dutch 560 65 4,505 56 154,367 76 Ethnicity: non-Dutch 304 35 3,581 44 48,269 24 Gender: Male 446 52 4,285 53 101,315 50 Gender: Female 418 48 3,801 47 101,321 50 Centrum 66 8 512 6 22,635 11 Oud-Zuid 79 9 442 5 20,525 10 Oud-West 42 5 407 5 14,825 7 Oud-Noord 152 18 1,330 16 18,580 9 Oosterparkwijk 85 10 870 11 11,840 6 Helpman en omgeving 80 9 617 8 19,105 9 Zuidwest 39 5 309 4 10,950 5 Hoogkerk en omgeving 19 2 270 3 12,260 6 Nieuw-West 54 6 639 8 16,505 8 Noordwest 96 11 1,056 13 18,170 9 Noordoost 101 12 813 10 16,465 8 Noorddijk en omgeving 51 6 548 7 16,400 8 Total 864* 100 8,086* 100 202,636* 100
24 Figure 4 and table 7 show the distribution of self-perceived job search behaviour for all sub-questions from question 16 in the questionnaire related to this topic. The first and the six question have a normal distribution, but to the other questions most individuals tend to agree more than disagree. Specifically, more than 60% of the sample agrees to: “An occupation means more to me than money alone.” and/ or “I can make a good impression when I apply for a job.”
Figure 4: Primary Data about Job Search Behaviour (in %). Source: Soepenberg (2019)
Q16_a Q16_b Q16_c Q16_d Q16_e Q16_f F % F % F % F % F % F % Valid 0 5 0.6 10 1.1 9 1.0 10 1.1 10 1.1 9 1.0 1 108 12.2 76 8.6 59 6.6 16 1.8 25 2.8 153 17.2 2 269 30.3 149 16.8 122 13.7 36 4.1 58 6.5 285 32.1 3 294 33.1 251 28.3 245 27.6 104 11.7 231 26.0 312 35.1 4 176 19.8 280 31.5 323 36.4 415 46.7 436 49.1 102 11.5 5 36 4.1 122 13.7 130 14.6 307 34.6 128 14.4 27 3.0 Total Valid 888 100 888 100 888 100 888 100 888 100 888 100 M 7216 7216 7216 7216 7216 7216 Total 8104 8104 8104 8104 8104 8104 Me 2.72 3.22 3.36 4.05 3.62 2.48 SD 1.061 1.196 1.148 0.983 0.979 1.034 Table 7: Descriptive Statistics about Job Search Behaviour. Source: Soepenberg (2019). Note: F = Frequency, M = Missing, Me= Mean, SD = Standard Deviation
0 10 20 30 40 50 60 No answer given Fully disagree
Disagree Not agree/ not disagree
Agree Fully
Agree
I can find a paid job, if I really put effort in it.
I want to find an occupation in the upcoming four months.
I think that I will find a job in the future.
An occupation means more to me than money alone.
I can make a good impression when I apply for a job.
I can find an occupation which fits my education and
25 4.2 Results of the Factor Analysis
A factor analysis is conducted in SPSS to study underlying factors in the database (Yong & Pearce, 2013), using the answers to the questions related to job search behaviour. This analysis is used to answer the first sub-question of this research. This is done by examining whether the five motivations (not) to search for a job as formulated by Vansteenkiste et al. (2004) correspond to underlying factor types in the data. If they do, that means that the five motivations are also applicable to UA recipients in Groningen, who participate in the experiment.
According to Yong and Pearce (2013), there are three requirements that need to be met to interpret the results of the factor analysis. First, the database should be large, without missing values. Second, the Kaiser-Meyer-Olkin Measure of Sampling Adequacy should be larger than 0,5. Third, the Bartlett’s Test of Sphericity should be significant. The database used, is close to 900 individuals, the KMO measure is 0,739 in this sample and the Bartlett’s test is significant. So, the analysis does meet the requirements.
Table 8 indicates that a total of six factors could be extracted from the data. That means that six different types of job search behaviour can be distinguished based on the data in the analysis. The meaning of these factors is discussed at the end of this section.
Component Initial Eigenvalues Extraction Sums of
Squared Loadings Total % of Variance Cumulative % Total % of Variance Cumulative % 1 2.465 41.090 41.090 2.465 41.090 41.090 2 1.050 17.495 58.585 1.050 17.495 58.585 3 0.808 13.471 72.056 0.808 13.471 72.056 4 0.684 11.401 83.457 0.684 11.401 83.457 5 0.574 9.566 93.023 0.574 9.566 93.023 6 0.419 6.977 100
Table 8: Total variance in database explained by common factors. Note: Extraction Method: Principal Component Analysis
Table 9 visualizes the result of the conducted factor analysis. In this analysis is chosen to limit the results to five components or common factors, because together they explain more than 93% of the variation in the data. In addition, the goal was to link five types of job search behaviour to the types of job search behaviour as distinguished by Vansteenkiste et al. (2004). These five components were the most fitting, which is elaborately discussed at the end of this section.
26
Components
Labels 1 2 3 4 5
I can find a paid job, if I really put effort in
it. 0.649 -0.532 0.100 -0.074 -0.450
I want to find an occupation in the
upcoming four months. 0.604 0.282 -0.618 0.340 0.071
I think that I will find a job in the future. 0.811 -0.091 -0.218 -0.091 -0.127 An occupation means more to me than
money alone. 0.517 0.617 0.109 -0.567 -0.014
I can make a good impression when I apply
for a job. 0.553 0.364 0.573 0.477 -0.063
I can find an occupation which fits my
education and experience. 0.669 -0.407 0.168 -0.081 0.588
Table 9: Component Matrix. Note: Extraction Method: Principal Component Analysis
Based on the conducted factor analysis, there are five common factors which can be extracted from the data. These are most applicable to the five motivations (not) to search for an occupation according to Vansteenkiste et al. (2004). These five types of motivations correspond to: Type 1 = autonomous motivation to search - 41% of the participants;
Type 2 = amotivation to search - 17% of the participants;
Type 3 = autonomous motivation not-to-search - 13% of the participants; Type 4 = controlled motivation to search - 11% of the participants; Type 5 = controlled motivation not-to-search - 10% of the participants.
This paragraph provides a detailed answer to the first sub-question of this research, related to different types of job search behaviour in Groningen. Component 1/ autonomous motivation to search has a strong positive correlation with all sub-questions, so individuals in this group are willing to find a job and they also perceive themselves as able to find one. Component 2/ amotivation to search has a positive correlation with questions related to the willingness of finding a job, but individuals in this group do not estimate themselves as able to find one. In contrast, component 3/ autonomous motivation not-to-search has a positive correlation with questions related to the self-perceived ability of finding a job, but individuals in this group are not willing to find an occupation.
27 In table 10, a prediction is given about the correlation between the motivation (not) to search determined by Vansteenkiste et al. (2004), which is discussed in section 2.2 and the questions related to job search behaviour from the survey. This prediction is based on the content of each separate motivation. For example, someone with an autonomous motivation to search is both willing and able according to himself or herself to find a job. This corresponds with strong positive correlation values on questions about job search behaviour. If table 10 is compared to the results of the factor analysis, there is a lot of overlap between the results and the predictions in the table below. There is one, negative value in red, in the amotivation to search. This is a positive value in the factor analysis, but it was expected to be negative. However, except for this value, all assumed correlations correspond with the correlations found in the table 9. Therefore, it can be concluded that the UA recipients in Groningen can be categorized in the same way as the five motivations (not) to search in the research of Vansteenkiste et al. (2004).
Motivation to search Motivation not-to-search
Autonomous Controlled Amotivation Autonomous Controlled
I can find a paid job, if I really put effort in it.
Positive Neutral Negative Positive Negative
I want to find an occupation in the upcoming four months
Positive Positive Positive Negative Neutral
I think that I will find a job in the future.
Positive Neutral Neutral Negative Negative
An occupation means more to me than money alone.
Positive Negative Positive Positive Neutral
I can make a good
impression when I apply for a job.
Positive Positive Negative Positive Neutral
I can find an occupation which fits my education and experience.
Positive Neutral Negative Positive Positive
28 4.3 Implementation of the Regression Analysis
In the previous section, a factor analysis is conducted to study whether the five motivations (not) to search based on Vansteenkiste et al. (2004) could also be applied to this research. Now, this research proceeds with the regression analysis, to study the impact of individual independent variables on the dependent variable job search behaviour. This analysis is used to provide an answer to the second and third sub-question of this research.
In section 3.3, a preliminary framework was provided for the regression analysis. In this section, that framework is extended, using the variables discussed in section 3.2. So, the independent variables age, duration of social assistance, education level, ethnicity, gender, intervention group and neighbourhood in the analysis are tested on job search behaviour as dependent variable. The exact preparation of the variables for the analyses is noted in the do-file in the second appendix. The methodological framework can be specified as following:
Y = β1*Age + β2*DurationSocialAssistance + β3*EDU1 + β4*EDU2 + β5*EDU3 + β6*EDU4 + β7*EDU5 + β8*EDU6 + β9*Ethnicity + β10*Gender + β11*Group1 + β12*Group2 + β13*Group3 + β14*Group4 + β15*Group5 + β16*NBH1 + β17*NBH2 + β18*NBH3 + β19*NBH4 + β20*NBH5 + β21*NBH6 + β22*NBH7 + β23*NBH8 + β24*NBH9 +
β25*NBH10 + β26*NBH11 + β27*NBH12 + ε
In these models, Y = job search behaviour. β1 - β27 (this means β1 to β27) = the effect of the independent variable X on job search behaviour (Y), X = respectively age, duration of social assistance, education level, ethnicity, gender, intervention group & neighbourhood and ε = error term. Based on the model, a null hypothesis and an alternative hypothesis can be formulated according to the Wald principle for hypothesis testing (Hill et al. 2012, p.599):
H0= β1 - β27 = 0; H1= β1 - β27≠ 0.
In words, the null-hypothesis assumes that there is no significant effect of the independent variables on job search behaviour. If this is not true for at least one independent variable, the null hypothesis is rejected.
The dependent variable job search behaviour is split in six different sub-questions. The independent variables are regressed on each of the different questions separately. All sub-questions are formulated below:
1. “I can find a paid job, if I really put effort in it.”
2. “I want to find an occupation in the upcoming four months.” 3. “I think that I will find a job in the future.”
4. “An occupation means more to me than money alone.” 5. “I can make a good impression when I apply for a job.”
29 As mentioned earlier, in this research ordered logistic regressions, also known as ordered logit models are conducted. Because the variable job search behaviour contains six sub-questions, six analyses are executed.
The requirements for this type of regression are as follows, according to Laerd Statistics (2019). First, the dependent variable should be an ordinal variable. Furthermore, it is necessary to use one or multiple independent variable(s), which can be either categorical, continuous or ordinal. Also, there should not be multicollinearity in the data. Multicollinearity in the data can be examined by estimating the Variance Inflation Factor (VIF) of each regression. To do this, all sub-questions were estimated as linear regressions without loss of generality, because it does not matter for the calculation of the VIF whether the dependent variable is an ordinal or a ratio variable. VIF values should be under ten, but in this analysis, all VIF values were below eight. Therefore, this assumption is also met. Finally, it is assumed that all categories of the dependent, ordinal variable contain proportional odds. In other words, the effect of the independent variables on each of the categories of the dependent variable is the same. All these requirements are met for all regressions.
4.4 Results of the Ordered Logit Regressions
Table 11 shows the regression results based on the data from 2017. Six ordered logistic regressions are conducted, with six different aspects of job search behaviour as dependent variable. The corresponding do-file is provided in appendix 2. Column (1) of table 11 shows the results for job search behaviour ‘1’ as dependent variable, column (2) shows the results for job search behaviour ‘2’ as dependent variable, and so on.
Age, the duration of social assistance, ethnicity and the intervention group often have quite a significant effect on job search behaviour, as opposed to the effect of the education level, gender and neighbourhood on job search behaviour. The variables age and duration of social assistance are both significant for five out of six regressions, but the effect these variables have on job search behaviour is small. If the education level is significant, the effect on job search behaviour is negative. The variable ethnicity is significant in all regressions and the effect is quite large. The effect of gender is significant in three out of six regressions, but the effect is both positive and negative, depending on the sub-question related to job search behaviour. The intervention group has a significant effect on job search behaviour. However, this is different for each intervention group. Group 1 only has two significant results out of six regressions. Group 2 and 4 have four significant results out of six and for group 3, all regressions contain significant values. The variable neighbourhood is hardly significant at all.
30 (1) (2) (3) (4) (5) (6) Age 0.952*** (0.007) 0.982*** (0.007) 0.922*** (0.007) 0.978*** (0.007) 0.995 (0.007) 0.987* (0.007) Duration of Social Assistance 0.996*** (0.001) 0.998* (0.001) 0.996*** (0.001) 0.999 (0.001) 0.996*** (0.001) 0.998** (0.001) Education level unknown 1.158 (0.401) 0.644 (0.221) 1.219 (0.425) 0.547* (0.191) 0.779 (0.281) 0.736 (0.252) Less than primary and
primary education 0.725 (0.198) 0.645* (0.169) 1.000 (0.269) 0.335*** (0.095) 0.571** (0.156) 0.917 (0.249) Primary and lower
secondary education 1.047 (0.277) 0.586** (0.157) 0.856 (0.227) 0.368*** (0.103) 0.634* (0.171) 0.839 (0.224) Upper secondary and
post-secondary non-tertiary education 1.077 (0.225) 1.008 (0.208) 1.240 (0.256) 0.537*** (0.118) 1.010 (0.218) 1.005 (0.211)
31 Table 11: Explaining changes in job search behaviour in 2017. Source: CBS microdata and survey data. Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parenthesis. Column (n) shows the results for job search behaviour ‘n’ as dependent variable. Numbers are proportional odds ratios, obtained by exponentiating the ordered logit coefficients.
The interpretation of odds ratios for binary, independent variables is as follows. The variable gender is considered in this analysis, which means that there is a distinction between males and females in the regression. If the odds ratio is between 0 and 1 for the variable gender, the chance is lower that a male (X = 1) agrees to the sub-question than a female (X = 0). However, if these numbers are larger than 1, chances are higher that a male agrees than a female. If the odds ratio = 1, the chance is equal for both males and females. So, the baseline value of the transformed βn equals 1.
In table 5 in section 3.2, the transformation of the variables is noted. Categorical variables have multiple categories and therefore, several new variables are created which sum up to the total amount of categories - 1 to use these variables in the regression analysis. The last category in the original variable is the benchmark category for all other variables. For example, the education level is divided into six categories. Therefore, 6 - 1 = 5 new variables are created with X = 1 for one specific education level and X = 0 for the other education levels.
The interpretation of odds ratios for categorical, independent variables is explained from now on. If odds ratios are calculated, chances should be interpreted based on the specific education level of that variable and the benchmark category, which is ‘6. (32) Master, doctoral or equivalent (Hbo-, wo-master, doctor)’ in this analysis. So, if the variable gender in the example in the previous paragraph is replaced by the variable ‘less than primary and primary education’, the interpretation should thus be as following. If the odds ratio is between 0 and 1, the chance is lower that an individual, who finished less than primary or primary education (X = 1) agrees to the sub-question than an individual with a master, doctoral or equivalent (X = 0). However, if the odds ratio is larger than 1, this chance is higher for a lower educated individual.
32 Switching from the interpretation of odds ratios to the exact interpretation of the numbers is done by the following example. The coefficient for someone who is not Dutch (X = 1) is 0.724. This is noted in the first column, eighth row: 0.724 (the coefficient) - 1 (the baseline) = -27,6%. In words, the odds are 27.6% smaller that a participant, who is not Dutch agrees to: ‘I can find a paid job if I really put effort in it’ than that a Dutch recipient agrees to that statement. Another example is that the coefficient for someone who participates in intervention group 3 (X = 1) is 1.809. This is noted in the second column, twenty-third row: 1.809 - 1 = 80,9%. In words, the odds are 80.9% larger that an individual in intervention group 3 agrees to: ‘I want to find an occupation in the upcoming four months’ than that a recipient in the control group agrees to that statement.
This paragraph provides a detailed answer to the second sub-question of this research about the relation between the characteristics of social assistance recipients in Groningen and their job search behaviour. As mentioned before, the effect of age on job search behaviour is small. It is slightly negative. That means that the degree of job search behaviour decreases when an individual is getting older. This effect is similar for the duration of social assistance. That means that the level of job search behaviour decreases, with the increase of the duration of unemployment. The results for low education levels are all below 0.7. That means that these levels of education score 30% lower in terms of job search behaviour than their benchmark category ‘(32) Master, doctoral or equivalent (Hbo-, wo-master, doctor)’. Individuals who are non-Dutch score on average 40% higher than Dutch natives in terms of willingness to find an occupation. However, their self-perceived estimation of being able to find one is 27.6% smaller as mentioned. Males have 29% more chance to state that they want to find an occupation in the next four months, but they have 31% less chance to state that employment means more than money alone and 35% less chance they can make a good impression when applying for a job compared to females. The effect of the intervention groups on job search behaviour compared to the control group is positive for all interventions. However, groups 1,2 and 4 contain less significant results than group 3. In addition, the effect of group 3 on job search behaviour is larger as well. The neighbourhood of the individual is hardly significant at all.
All in all, individuals tend to have a higher degree of job search behaviour, when they are younger, short-term unemployed, when they have a higher degree of received education, when they are non-Dutch, male, and when they are participant in intervention groups 2, 3 and 4. Based on the results, it is remarkable that recipients, who are non-Dutch have a higher level of self-perceived job search behaviour. It could be that they estimate their chances of finding an occupation relatively higher compared to Dutch natives or that they have more motivation to search for a job.
33 an individual originates from does influence his or her job search behaviour more than his or her ethnicity. This non-significant result means that there is no need for a neighbourhood-specific policy related to social assistance in Groningen. Whether there is a significant difference between different regions in the Netherlands should be examined in further research.
At last, it is important to mention policy implications which can be deduced from this research. As mentioned, 41% of the participants has an autonomous motivation to search. This indicates that most of these people are willing and able to find a job themselves. However, there is also a substantial part of the recipients with an amotivation to search or an autonomous motivation not-to-search. An individual who is amotivated to search might need personal assistance (intervention 2) to improve their self-confidence and their well-being to find an occupation. This is a carrot incentive as defined in section 1.2. On the other hand, an individual with an autonomous motivation not-to-search might need to be motivated to search for a job by stick incentives, such as the obligation to apply for occupations. After the experiment in Groningen is completed, the results could indicate which method is most suited for each type of job search behaviour. In addition, existing schooling and job training possibilities need to be adapted to ongoing skill biased technological change to prevent a potential mismatch in demand and supply within the Dutch labour market, as elaborately discussed in section 2.3.
34
5. Conclusion of this Research
This chapter provides the conclusions of this research starting with the main findings in section 5.1. A reflection on this research including its limitations is given in section 5.2.
5.1 Main Findings
In section 1.4, the research questions of this research were formulated. This chapter returns to these questions to conclude and reflect on this research. The main question is: “What is the job search behaviour of social assistance recipients, based on data from the experiment related to social assistance in Groningen?”. To answer this question, three sub-questions were defined. The first sub-question is related to different types of job search behaviour in Groningen, in the context of job search behaviour as defined by Vansteenkiste et al. (2004). The second sub-question is concerned with the relation between recipients’ personal characteristics and job search behaviour. At last, the third question is associated with the spatial pattern of job search behaviour in Groningen.
In chapter 2, a conceptual framework is developed using scientific literature, in which several key concepts were defined. They are summarized in this paragraph. A distinction can be made between unemployment assistance and unemployment benefits. Individuals receive unemployment benefits up and until they are unemployed for two years. Benefits are related to their previous income. After two years, they are eligible to unemployment assistance. This research focusses specifically on the job search behaviour of those who are unemployed for more than two years. Vansteenkiste et al. (2004) distinguishes five different types of job search behaviour. These are: autonomous motivation to search, controlled motivation to search, amotivation to search, autonomous motivation search and controlled motivation not-to-search. Job search behaviour itself is determined by personal characteristics of the recipient, the duration of social assistance and the intervention group of the individual. This is visualized in figure 2. Personal characteristics can be further specified into: age, education level, ethnicity, gender and the neighbourhood of the recipient.
In this research, both primary and secondary data are used to provide an answer to the research questions. Primary data are collected by surveys as part of an experiment in Groningen. Secondary data are microdata, collected at an individual level from CBS. The variable job search behaviour was deduced from the primary data, while the independent variables were deduced from the secondary data.
