Introduction
Ever since the introduction of the minimum wage policy by New Zealand, there has been a vivid debate about its economic consequences. The policy aims to protect employees at the bottom end of the labour market from wage exploitation, yet it is simultaneously linked to increased unemployment. In the Netherlands the debate’s focus recently took a sharp turn towards youth minimum rates in response to a social initiative of unions and political parties called ‘Young and United’. The Dutch minimum wage policy provides a very interesting setting since the gap between the youth rates and the adult rates is among the largest of all countries . Young and United considers the subminimum rates for the youth as age-discrimination since older employees earn more for the exact same work . Another recurrent argument is the crowding out of people just above each age threshold as their labour costs are artificially higher due to the legislation . Proponents of the segregation of minimum wages claim that equalling the rates will have a disproportional negative effect on youth unemployment .
This study aims to contribute to the debate by empirically exploring the effects of the current Dutch youth minimum wage policy on youth employment using a regression discontinuity design. More specifically, the impact of the statutory increases in minimum wage rates between the ages of 20 to 24 are examined . Employment is thereby assessed as both the chance of acquiring a job and actual hours worked per week. The use of a discontinuity regressions is relatively novel in this field of study and resolves estimation difficulties encountered in the conventional time-series and cross-sectional regressions. Dickens et al. (2010) and Olssen (2011) applied discontinuity regressions to investigate minimum wage effects in respectively England and Australia, finding evidence of positive and zero employment elasticity of minimum wages. These ramifications have challenged the traditional consensus of negative employment elasticity of minimum wages and encourage new research. Though the Dutch system maintains one of the world’s largest segregation in minimum rates recent studies to the Dutch situation are absent and hence the value of this study illuminates.
The outcome of the discontinuity regressions only shows a significant negative effect for the employment chances related to the youth minimum wage increase on the age of 22. For the other two age categories no significant effect was found, just as hours worked per week turned out not be effected by the different subminimum rates maintained in the Netherlands. The evidence thus mainly rejects the old consensus of negative employment elasticity of youth minimum wages. This is in line with recent studies such as Olssen (2011) and Hyslop & Stillman (2007). The results are unique however for the Netherlands. Most of the existing studies examining the Dutch system have found evidence for negative employment effects of youth minimum wages.
The next section of this paper provides an overview of the related literature. Section 3 explains the theoretical consequences of minimum wages, followed by an outline of the current Dutch legislation of minimum wages in section 4. Section 5 present the applied methodology, whereas section 6 summarizes the data sources. In Section 7 the results are discussed. Finally section 8 concludes.
‘
Review of Literature
Youth minimum wage studies are part of a much broader literature on general minimum wages. Neumark and Wascher (2006) conducted a meta-analysis to more than a hundred of these studies inferring a considerable consensus that prevailed over the majority of the 20th century. More than half, and over three-quarters of the papers judged most credible, provided evidence of a negative employment elasticity of minimum wages. Moreover, the empirical evidence was backed-up by standard economic theory. The meta-analysis includes a wide variety of papers regarding both geographical settings as well as methods of analysis. Yet time-series regression and cross-regional comparisons in the United States are recognized as dominant in the early literature (Neumark & Wascher, 2006).
The established consensus of negative employment effects of minimum wages has been challenged by a series of studies from Card (1992a;1992b) and Card & Krueger (1994), who found neglectable to positive effects on employment as a result of increased minimum wages in the United States. It was not merely the contradictive results that revived the debate, but more importantly the innovatory approaches used in these studies to investigate minimum wage effects. Problematic for the time-series and cross-sectional regression dominant in the early literature is the typical lack of variation in minimum wages over time or regions. Large revisions of minimum wages occur at irregular intervals and are relatively rare, whereas yearly adjustments for inflation does not augment the real cost or benefits of the minimum wages. Discrepancies between state minimums have remained small for a long time as well (Pereira, 2003), while a multi-country comparison leads to severe cofounding complications . With such a small base of variations any general conclusion is heavily dependent on extra-polluting, thereby quashing the validity (Moore, McCabe, Alwan, Craig, & Duckworth, 2011). Card tackled the issue this issue in his 1992(a) study by combining time and regional variations in a panel-data study to evaluate federal minimum wage changes.
Card and Krueger (1994) tackled the issue by focussing on a single policy reform. Policy reforms are typically associated with significant and unanticipated variations in the minimum wage level. This creates a natural-experimental setting where the reform acts as ‘treatment’. The effects of the treatment on the labour market can subsequently be determined by applying a difference-in-differences analysis. Other advantages of the method are its simplicity in use and its detachment from any functional form of the relationship between variables, allowing to detect a variety of different relationships. This simultaneously holds the drawback that it is not possible to construction a relational equation for the phenomenon of interest. Still the Card and Krueger study proves that the conclusion can be very influential. In their application of the difference-in-differences approach they compared the employment in fast food restaurant in New Jersey -where the statutory minimum had risen 19%- with Pennsylvania, where the rate remained constant. The results showed that the employment in New Jersey had increased with 13% relative to Pennsylvania as a consequence of the policy reform. It is one of the most cited studies in this field of study and caused a new upswing of research to minimum wages (Neumark & Wascher, 2006).
In the new upswing of academic research to minimum wages enlarged emphasis was put on the specific characteristics of different policies to establish a more sophisticated picture. This included studies to youth minimum wage regimes, which is the subfield this paper qualifies for. Mixed results have been found so far in estimating the effects of a segregated minimum wage system. Hyslop and Stillman (2007) investigated the large amendments in the youth minimum wage in New Zealand of 2001, again exploiting the natural-experimental setting presented by such a reform. In line with the trend the authors utilised the difference-in-differences approach to compare employment among teenagers (16-17 years) and young adults (18-19 years) to a control group (20-24). The results provided weak evidence of employment loss in the medium run. At the same time total earnings for the youngest wage groups did increase following the increased minimum wage, partly due to this higher wage and partly by an increase in the hours worked per person . Importantly, Hyslop and Stillman acknowledge that the use of the difference-indifference encompasses two strong assumptions. First it must be assumed that the policy reform does not influence the employment rate of the control group in any way. Secondly the employment rates of the different age groups are assumed to react in an exactly similar way to any other economic event. If for example during the time of analysis an economic crisis kicks-in and youths would be more likely to be fired than the older people in the control group, the results would be biased.
Pereira (2003) performed a research very similar to Hyslop and Stillman’s to the abolishment of subminimum wages in Portugal. This time decisive negative effects on youth employment were present. Estimates of the minimum wage elasticity for youth employment range between -0.2 and -0.4. At the same time the conclusion are in sharp contrast with findings of Portugal and Cardoso (2006) who inferred a positive effect of minimum wage increases on youth employment examining the exact same Portuguese reform .
In resolving the puzzle at hand most recently a new research method has emerged in minimum wage economics: the regression discontinuity design. Dickens et al. (2010) and Olssen (2011) applied the discontinuity design to investigate the labour market in respectively England and Australia, estimating a positive effect on employment in the former case and no significant effect in the latter. The discontinuity regression exploits the vast increase in the statutory minimum associated with ageing one year, by comparing those aged just below and above the threshold. It lingers on the assumption that these persons will have very similar characteristics except for their entitlement of the minimum wage. Any difference in employment between the groups can therefore be linked to the mandatory wage increase. Since the regression centres around a single and significant change in the minimum rate, enough variation is captured to make meaningful inferences. In addition the discontinuity design also deals with the two assumptions concomitant with the difference-in-differences approach. First of all the regression technique makes the use of a control group obsolete, discarding the first assumption. The second assumption is relaxed since a fundamental aspect of the discontinuity regression is that all examined individuals are in near proximity of the age threshold and therefore very similar in characteristics. Along this reasoning there is thus little suspect of differentiated reactions between the groups due to exogenous shocks. As the discontinuity design has many advantages over conventional methods applied to estimate minimum wage effects on employment, and the design is highly suitable for examining increments in minimum rates occurring at youth ages, the results of Dickens (2010) and Olssen (2011) prove that the old established consensus of negative employment elasticity of minimum wages is not as evident as it once seemed.
With the global evolution of studies to minimum wages presented, reviewing both the implications of the results as well as the applied research methods, our focus narrows down to studies set in the Netherlands. Van Soest (1994) assessed the Dutch situation using both a macro-economic and a micro-economic point of view. The macro analysis consisted of an ordinary least squared regression on the relative employment of youths to the whole population, using aggregate data of the Dutch labour market from 1974 until 1988. The outcome was ambiguous however. In line with the complications encountered in similar time-series regressions, the lack of independent variations in the minimum wage was recognized as main fallacy. For the micro analysis Van Soest estimated minimum wage effects on potential earnings and chance of employment of individuals in an econometric model. Using this econometric model he did find a negative impact of youth minimum wages on their employment chances, in accordance with conventional theories. Van de Berg & Ridder (1997) also concluded a negative labour elasticity of minimum wages in the Netherlands. Their conclusion reached even further by stating that youths are more hurt by the legislation than adults. Though both studies reach similar conclusion for the Dutch labour market, they also share the ambiguous conventional research methods. Moreover the studies are executed two decades ago, making them somewhat out-dated. Van Soest’s use of data over the years 1974 to 1988 is for example very questionable as the minimum wage policy changed drastically in 1983 . Re-examining the Dutch situation using the discontinuity regression design can therefore shed new light on the consequences of the Dutch minimum wage legislation.
‘
Theoretical Framework
As mentioned before minimum wage policies are originally designed for the benefit of the least prosperous employees in the economy, yet their consequences are diverse. This paragraph provides an overview of the economic theory surrounding minimum wages, supporting multiple views on the matter. First the most basic minimum wage model is discussed, predicting negative employment effects. Thereafter the model is extended by focussing on distributional concerns and subminimum rates, whereas the last section considers non-competitive labour markets.
The most basic model of minimum wages assumes a perfectly competitive labour market, consisting of a downward sloping demand curve and an upward sloping supply curve. Without any intervention the market reaches its equilibrium at the intersection of both curves, indicating that the economy is at its natural level of employment paying the market-clearing wage (Stigler, 1946). When a minimum rate is installed, the wage level is artificially raised above its equilibrium value creating an excess supply of which the implications are twofold. On one hand the labour supply increases as the higher wages lures more people into the labour market. On the other hand firms are hiring less employees since the labour cost increased, hence a decreased labour demand (Stigler, 1946). This story is well presented in the graph below. Originally the wage level is equal to w0 where supply and demand meet. With a minimum wage set at wmin unemployment is created equal to Sm ‘ Dm.
Graph 1: Labour market and the minimum wage intervention
The negative impact on employment depends on the height of the minimum wage in conjunction with the elasticities of labour supply and demand. The elasticities measure the response in labour supply and demand to a change in the wage level. More elastic supply and demand are therefore associated with more unemployment keeping the minimum wage constant (Borjas, 2013). At the same time a higher minimum wage level widens the gap between supply and demand, therefore also enlarging unemployment. For this reason policy makers face a trade-off between higher incomes for the poor and more unemployment in the economy.
Another dilemma appears when distributional concerns are taken into account. Minimum wages are implemented to increase the income of those on the bottom-end of the labour market, which are typically low-skilled and young inexperienced workers. This exact same group however, is also most likely to be among the least productive employees of an economy (Frank, 2010). When firms are confronted with mandatory wage increases they will start by laying-off their least productive workers, creating the paradoxical problem that the policy puts at risk the jobs of those people it intentionally tries to help (Burkhauser, Couch, & Wittenburg, 1996) . The complications intensify when we recognize the fact that the real economy is not captured in one consolidated labour market, but rather a collection of many smaller and distinct labour markets. As mentioned before young employees are likely to be less productive than more experienced workers and as a direct consequence the equilibrium wage on the youth labour market must be below the market-clearing wage of elder more productive workers (Borjas, 2013). Keeping in mind that larger differences between the equilibrium wage and the statutory minimum induce more unemployment it becomes clear that a single minimum wage rate hits juvenile groups exceptionally hard. To counter the distributional problems of a single minimum wage, youth ‘subminimum’ rates can therefore be used (Dolado, et al., 1996). Subminimum rates are below the general rate to alleviate the negative employment effects. Additionally the reduce labour costs restores some of the incentives of employers to hire and train inexperienced workers. By doing so these workers are given an opportunity to increase their productivity and reach higher ends of the labour market in the future (OECD, 2008).
Though subminimum wages can alleviate the negative employment effects of the minimum wage legislation, or ideally make them obsolete, they are still unable to provide a theoretical foundation for positive employment effects found in some empirical studies. Therefore we have to look at the monopsonistic model. Monopsonistic firms face an upward sloping labour supply curve, implying that in order to hire an extra employee the company must increase the wage offer (Borjas, 2013). This is unlike firms operating in a competitive labour market , which face a perfectly horizontal labour supply curve. Assume that the monopsonist is unable to discriminate between its workers, so that every additional employee hired increases the salary of all employees. As a consequence the marginal cost curve is even steeper upward sloping than the labour supply curve. The marginal cost curve is presented by the red line in the graph below. Any profit maximizing firms should hire until the marginal costs of labour are equal to the labour demand (Frank, 2010). Initially this is indicated by point A, leading to a wage level w0 and employment of E0. The situation changes when a minimum wage is implemented, leading to a partly flat marginal cost curves in the beginning at the height of the minimum wage. This is presented by the dashed line in the graph. As a consequence the new profit-maximizing number of employees is pushed from E0 to Dm. The wage level is equal to wmin and hence we see that in the case of a monopsonist the minimum wage legislation can increase both wage and employment.
Graph 2: Monopolistic labour market
Conclusively we can see that the effect of a minimum wage intervention on the employment in an economy depends on many factors, such as the height of the minimum rate and the competitiveness in the labour market. The actual outcome in a realistic complex economy therefore remains largely an empirical question.
The Dutch Minimum Wage Legislation
Before elaborating on the specifications of the discontinuity design it is convenient to get a better understand of the current minimum wage system in the Netherlands. In the current Dutch legislation the statutory minimum wage is dependent on age. The law states that employees over the age of 23 must earn at least 1500 euros a month for a full-time job. Younger employees are only entitled to a fixed percentage of this general minimum rate, gradually increasing from 30% when 15 year old to 85% for people aged 22 . This system has been installed in 1983 and has changed little since (Altes, 1983). The rates are adjusted for inflation semi-annually, but no changes have been made to the relative entitlement of each age group since 1983. In appendix table 1 a full overview of the 2015 rates is provided.
Though the Dutch system encompasses an increase in statutory minimum for every age until the age of 23, which is very unique in the world, not all of these threshold satisfy the conditions necessary for the analysis. The discontinuity regression relies upon the assumption that individuals just above and below each age threshold are very similar except for their entitlement to the minimum wage (Thistlethwaite & Campbell, 1960). However in the age category 15 to 18 the Dutch legislation also constitute changes in work permissions and authorized work hours. Other factors that might co-found with the increase of the minimum wage are the average age of graduation and the entitlement to study benefits. Forasmuch as labour restrictions gradually expire until the age of 18 and changes in educational status are also more common in the younger ages, we only focuses on the following thresholds in our analysis:
23rd birthday 15% increase in legal minimum wage
22nd birthday 12,5% increase in legal minimum wage
21st birthday 11% increase in legal minimum wage
‘
Methodology
A discontinuity regression will be used to estimate the effects of the designated increments of the minimum wage on the youth employment. Aforesaid the discontinuity design exploits the fact that persons aged just below and above the threshold age should be very similar in characteristics except for their entitlement to the statutory minimum. Any statistical difference in employment between the two groups can therefore be linked to the increased minimum (Thistlethwaite & Campbell, 1960). The closer the sample centres around the threshold age, the more accurate the results will be (Lee & Card, Regression Discontinuity Inference with Specification Error, 2008). Individuals aged above the cut-off are referred to as the ‘treatment group’. It is important that the treatment group does not differ on any aspect rather than the entitlement of the minimum wage to prevent cofounding concerns. As explained in the last section there are serious cofounding complication for teenage workers, hence our focus on youth employees aged 21, 22 and 23. To evaluate employment among these youths two different measures are used. First a binary variable is introduced to indicate someone’s employment status. The variable takes value 1 if someone is employed and 0 for the unemployed. Secondly we assess employment by the average hours worked per week, allowing to establish a more sophisticated relationship.
Before the actual model is estimated, the collected data will be visualized. Dot plots are a very convenient way to detect possible discontinuities in the data (Imbens & Lemieux, 2008). The mean employment rate and the average hours worked are plotted against age, where age is defined as months from threshold age. A dashed line splits the graph in two indicating the cut-off point. Additionally a simple linear regression is performed at both sides of the cut-off. Any possible discontinuity is easy to detect as this would mean a clear jump in the data at the cut-off point. Finally the graphs are also a good first indication of the functional form (linear, quadratic etc.) between age and employment, on which information is required for the remainder of the analysis.
Following the construction of the plots the regression model is calculated. There are globally two different approaches for building the model. One option is to estimate a single regression over the whole age interval and include a dummy variable to indicate if a person is above the age threshold and thus receives the ‘treatment’, i.e. higher minimum wage (Olssen, 2011). The other option is to estimate a local regression at either side of the discontinuity point (Imbens & Lemieux, 2008). The average treatment effect is then calculated by taking the difference of the constant terms of both regressions . The use of two local regressions makes it is easier to allow for different parameters on both side of the threshold, such as different slope coefficient for age. At the same time however, allowing for different parameters could phantom a discontinuity when the functional relationship is actual constant over the whole interval . Additionally local linear regression are nonparametric functions, which typically require larger sample sizes to get significant results and they are very dependent on the chosen bandwidth (Lee & Lemieux, 2009). Given our dataset there is little space for adopting the bandwidth. Conclusively we apply the discontinuity regression over one interval, given by the following formula:
Y_i=c+f(‘age’_i )+ ��D_i+ ��_i
With Yi Employment variable
c Constant term
f(agei) Function defining the general relationship between age and employment
Di Dummy variable: 0 if agei < threshold 1 if age ‘ threshold
��_i Error term
Age is again specified as months from threshold, so that the constant term indicates the average value of employment at the threshold age. Since we use two different variables for employment we run two regression for each age threshold. A logistic regression is applied to estimate the chance of employment and a normal regression estimates the average hours worked per week. To determine the functional form between employment hours and age we use the graphs plotted before as a first indication. In addition we compare the F-significance of a linear, a quadratic and a cubic relationship between age and employment to back-up the choice. Pivotal in the results is the size and significance of the dummy coefficient Beta. Beta captures the average difference in employment between the discontinuity between the younger group and the older ‘treatment group’. Hence it functions as proxy for the discontinuity. Finally ��_i is the error term.
The correctness of the model will be checked in several ways. The discontinuity regression lingers on the assumption of homogeneous characteristics of the sample groups, except for the entitlement to the statutory minimum (Thistlethwaite & Campbell, 1960). To check the homogeneity of the groups we add control variables to the regression and investigate if the beta coefficient changes value. In case of perfect homogeneity, Beta does not change in value but only becomes more significant (Lee & Lemieux, 2009). We will control for educational status, gender and country of origin. The new regression is formulated as follows:
Y_i=c+f(‘age’_i )+ ��D_i+��X_i+ ��_i
Where Xi is the set of control variables.
In extend we try to falsify the model by checking if a discontinuity is present in the education variable. If a discontinuity is present in education this would again violate the homogeneity assumption and put ambiguity on the results.
‘
Data
All the data required for the analysis is collected from ‘het arbeidsaanbodpanel’ (AAP), a Dutch Labour force survey performed by the Netherlands Institute for Social Research. The AAP is a longitudinal panel survey with a biennial frequency. The survey draws from a set of over 2000 Dutch households, including around 5000 individual observations. Stratification of the panel is based on age, household situation, primary income source and region to assure a valid representation of the whole Dutch population. Around 400 records per survey are in the age group of interest. To derive a sufficient sample size all records from the surveys between 2004-2012 are pooled together . During this whole period the minimum wages have only been adjusted for inflation, leaving the real value of minimum wage benefits approximately constant. Yet it must be recognized that by pooling the records from multiple years some individuals will be present twice in the dataset. As the panel survey draws from a designated set of household and is held every two years, individuals aged 20 will also appear as 22 in the dataset. At the same time, the sample replaces around 25% of its households every survey round reducing the bias. Moreover there is no abiding relationship between having work today and in two years.
In the dataset the age of each individual is calculated using the variables on their month of birth, the year of birth and the date of completing the survey. It allows age to be specified up till months which is important for the analysis. Some records have missing values for date of completion, for which the average completion date is used to estimate the age. Since the survey is executed within the last quarter of each designated year, indicating that there are no large outliers, the impact of the possible errors is neglectable .
Employment is deducted from the ‘status of employment’ variable. The status of employment is a categorical variable with 7 groups, conveying info on the type of contract or reason of unemployment. We are only interested in the distinction of having a paid job or not and therefore merge the appropriate categories into employed and unemployed. For the measurement of hours worked per week, the actual hours worked per week are taken as proxy instead of the contractual amount. Besides the actual hours mirroring reality better, the contractual hours are problematic due to the existence of flexible contracts, which typically do not specify a fixed amount of hours per week.
What remains are the control variables. Gender is already in the appropriate binominal form, but country of origin and educational status have to be transformed. We again construct a binominal variable by pooling appropriate categories. For country of origin all foreign born respondents construct one group, while the other group consist of all Dutch natives. The new education variable takes value 0 when no education has been taken in the last two years, while it take value 1 if education has been followed in the last two years or is still being followed.
Appendix Table 2 summarizes the statistics of the sample. The columns in the table represent the different age categories necessary for the analysis of the three statutory minimum increases under investigation. The table reveals that our dataset for the most part fits to the assumptions encompassed in the discontinuity design. The difference in percentages of females, foreigners and students between the appropriate groups are all less than 5 per cent, except for the female rate of the 23 threshold sample. The sample group aged 23-23.5 has 9 per cent more women than the 22.5-23 group, indication that the results could be a little biased due to a gender difference in the groups. Another aspects that become clear in the summary statistic is that the percentage of foreigners in the sample is very low. Only around 2% of the respondents are foreigner. The conclusions are therefore mainly applicable to the native Dutch population. These two complications will be discussed in the conclusion section as well.
Results
With all data acquired and transformed into the appropriate form, the analysis has been performed in the way described in the methodology. This section presents the results. First we consider the dot plots constructed for the three age threshold, on both employment rates and hours worked per week. All graphs are printed in the appendix. Only one out of the six plots provides a clear indication of a discontinuity, namely the employment rate around the age of 22. The graph clearly shows a discontinuity at the cut-off point since the drop in the data is considerably larger than the general variability in the whole plot. Moreover the regression on both side of the cut-off seem to have a decent fit. Since the discontinuity is negative it implies a negative effect of the minimum wage increase at the age of 22 on employment chances. However, when we look at the hours worked per week for the same age group in Appendix graph 4, there is remarkably no clear discontinuity visible. Although there is also a small drop at the cut-off line, the disconnection is clearly smaller than the overall variability of the graph. In this case the disconnection is thus more likely to be an ordinary variation between the age intervals then a true discontinuity. The same holds for the other four graphs where the disconnection at the cut-off point is even smaller and in some cases neglectable. Relying on the graphs there is thus little indication of employment effects for the youth related to the minimum wage. Yet we continuity the analysis to acquire more statistical evidence for our conclusion.
The regression model we constructed is linear, since this is indicated by the plots and moreover backed-up by the line estimation, presented in appendix table 3. The first round of regressions consist of the logistic model to estimate the effects of the age related increments of minimum wage on the chance of being employed. The results of these regression are summarized in Appendix table 4. The outcome of a logistic regression normally present the coefficients as log odds, which are hard to interpret directly. For convenience percentage effects have therefore been calculated from these log odds and added to the table. The outcome reveals that the increases of minimum wage rates at the ages of 21, 22 and 23 lead to respectively a 21%, 39% and 23% decrease in employment chances. However, the results are only significant to a 10% level for the age of 22, while highly insignificant for the other two thresholds. Even when we add the control variables the results for the 21 and 23 threshold remain highly insignificant. Inclusion of the control variables further reveals that the homogeneity assumption for the 23 age threshold is questionable. The deviation in discontinuity coefficient before and after the inclusion of the control variables is 18% , indicating that the control group just younger than 23 is not completely similar to the treatment group who are just aged above 23. This confirm the concerns expressed in the data section; that the group 23-23.5 consist of more women and is thus a source of bias. For the 21 and 22 age groups the homogeneity assumption does seem to hold, since the value of the discontinuity itself has not changed more than 10%.
Although youth employment chances do not change much as results of the increases in subminimum rates, it is still possible that the more expensive employee is used less often. This possibility is tested in the next set of regressions, where we estimate the hours worked per week based on age and entitlement to the minimum wage. The results are presented in Appendix table 5. This time the discontinuity coefficient indicates the nominal difference in employment hours for persons just above and below the threshold. The results are striking. For none of the age threshold a significant discontinuity is established. Even in the 22-age threshold, which did show a significant negative effects on employment chances, the discontinuity is highly insignificant with a P-value of 0.359. Adding the control variables does not change the implications of the results, while it does again show that the 23 threshold suffers from severe heterogeneity issues.
As we only found 1 significant discontinuity, a negative impact on the chance of employment at the age 22, we only run the falsification on the age of 22. We again construct a single regression discontinuity, where educational status is the dependent variable and age and the treatment dummy are the explanatory variables. The outcome is given in Appendix table 6 and shows that educational status has no discontinuity at the age of 22. It is very remarkable that we infer one significant discontinuity in the data, while the other five regression show a highly insignificant discontinuity. Explanations for this results could either be in a lurking variable/process that is related to turning 22, or a coincidence in the data.
The majority of the results is however in mutual congruence and suggest that there are no effects on youth employment caused by the increases in subminimum wages. Part of the results can be explained by the theory that youth minimum wages alleviate the negative employment effects. Since the youth minimum rates better mirror the productivity of the youth than would be the case if they would be entitled to the full rate, there is less adverse selection for hiring youths. The subminimum rates thereby increases gradually from year to year, preventing one single large jump in the minimum wage. The increments associated with ageing are therefore smaller than would be the case with one single increase. This again reduces negative employment effects. Other possible explanation for the lack of evidence of negative employment effects could be an inelastic labour demand or a lack of competition in the youth labour market. Any inference about one of these two possibilities is however beyond the scope of the research and should be examined in a separate research.
Conclusion and Discussion
In this thesis the empirical effects of the Dutch youth minimum wage regime on youth employment have been explored using a regression discontinuity design. The regression discontinuity design was utilised as it has recently been introduced to minimum wage studies to overcome estimation complications encountered in conventional regressions and the difference-in-differences approach. With the discontinuity design we exploited the fact that youths just above and below a certain age are very similar except for their entitlement to the minimum wage. This allowed the estimation of employment effects due to differences in subminimum wage rates. The majority of the results indicate that there are no employment effects for Dutch youths as a results of minimum wage increases. Both the chance of acquiring a job and the actual hours worked per week are not influenced by the increased minimum wage. The only exception found in this study is the chance of acquiring a job when turning 22. For this age threshold the chance of employment remarkably drops with 40%, significant to the 10% level. It is hard to find an explanation for this deviance in the setting of this analysis. New research would be required to search for a possible causation of this result. The main conclusion is however congruent with recent studies to minimum wages, such as the study from Olssen (2011) and from Hyslop & Stillman (2007). The findings imply that the subminimum rate regime works well in alleviating the negative employment effects for youths. The youths still get hired as their subminimum wage better mirror their productivity than a full minimum wage. In extend the increases associated to ageing are not of such size that they cause substitution by younger persons and hence we detect no significant effects .
Although the study draw from a professional labour market survey, the sample encompassed some unsolvable complications. It must be recognized that the amount of foreigners in the sample is lower than in the real Dutch population. Our results are therefore applicable to the native population. Additional research should therefore include a larger share of foreigners to extend the implications. Still this study has proven that the old consensus of negative employment effects of minimum wages is not as evident in the Netherlands as once alleged.