The economic consequences of
partner choice
The effect of educational assortative mating
and labour participation of couples on income
inequality in the Netherlands.
Sam Verbraak
Supervisor: dr. A. S. Booij
Department of Microeconomics
University of Amsterdam
Master Thesis
August 28, 2015
AbstractTheory states there could be economic consequences to choices made by couples. This because assortative mating, both on education as on labour choices, could cause income inequality to rise. Research so far however is ambiguous in its conclusions and has been not able to find convincing results. Since the Dutch economy provides a mixture of the earlier studied countries, this research will take a closer look at it. First of all educational assorta-tive mating did rise and so did labour participation of women, creating more evenly educated couples and more dual-earner households. Yet even with these promising results, income inequality did not rise because of it. Thereby being once again unable to prove the theoretical link in practice.
Keywords: Education, Assortative mating, Labour participation, Couples, Income inequality
Contents
1 Introduction 3 2 Literature review 4 3 Data 5 4 Educational sorting 7 5 Inequality analyses 10 6 Labour participation 127 Discussion and Conclusion 16
References 18
8 Appendix 19
List of Tables
1 General statistics . . . 6
2 Kendall’s tau & Fisher statistics . . . 8
3 Educational homogamy . . . 9
4 Theil decomposition I . . . 12
5 Couples labour participation decisions . . . 15
6 Theil decomposition II . . . 15
7 Regression of education and income . . . 19
8 Average income ¯xj . . . 20
9 Group specific Theil indices Tj . . . 21
List of Figures
1 Hours worked per week . . . 131
Introduction
Choosing ones partner is a choice many people experience to be difficult. Not only does one need to actually find someone of ones liking, but even if one finds such a person, how does one know he or she is the right choice? Most of the time, people will try to reduce this uncertainty by finding someone with similar traits, like shared interests, background or age. This is referred to as assortative mating, the phenomenon of similarity between partners which occurs more often than what is to be expect by chance (Huber & Fieder, 2011).
Educational assortative mating is choosing a partner with the same educational level as yourself. In recent history there has been a growing trend of educational assortative mating (Schwartz & Mare, 2005). It is argued that this growing trend could have consequences for the income inequality within a country, because there are multiple channels through which educational assortative mating could influence inequality (Blossfeld & Buchholz, 2009; Greenwood, Guner, Kocharkov, & Santos, 2014).
While from a theoretical viewpoint this relationship between educational ho-mogamy and inequality is plausible, previous research so far has been unable to prove it. Richard Breen has done three studies in different countries trying to verify it, but he remains empty handed (Breen & Salazar, 2010, 2011; Breen & Andersen, 2012).
Next to the assortative mating itself, there are many other choices influenced by being in a relationship, therefore Druker and Stier (2014) also look at the effect of labour choices by couples. At the same time with the rise in assortative mating, there has been a rise in female labour participation. This changed the way couples divide their workload and could also influence income inequality (Druker & Stier, 2014; Blossfeld & Buchholz, 2009). This paper therefore looks at the influence of the choices of couples on income inequality in the Netherlands, specifically on the effect of the rise in educational assortative mating and the change in labour participation. In his article Breen already states there are country specific qualities that could influence the relationship between educational assortative mating and inequality (Breen & Andersen, 2012). The Netherlands can be considered to be a combination of the countries studied by Breen and it also has a unique labour market with a large number of people working part-time. Making the Netherlands a promising choice in order to research the effects of assortative mating. The present research is conducted with the use of data collected by Statistics Netherlands and the Theil index is used in order to measure inequality.
My expectation is that the unique characteristics of the Dutch economy could facilitate educational assortative mating to have an influence on inequality. However also the Dutch case fails to show that the increased educational assortative mating has led to higher inequality. Assortative mating did increase significantly but this does not lead to higher inequality. Secondly the labour choices by couples have a more ambiguous effect on inequality. Depending on what measure for income is used, the results change from a negative relationship to no relationship. So overall there seem to be no direct link between mating choices and inequality in the Netherlands. The rest of this research will be presented in the following order: First the link between education and income will be examined after which previous research will be discussed. In the third section the used data will be described. Which will be
followed by the analyses of the effect of assortative mating in section four and the analyses on the effect of labour choices in section five. Finally the discussion and conclusion finish up the paper.
2
Literature review
What is the theoretical framework by which assortative mating influences income inequality? As stated in the introduction, theory predicts multiple ways by which assortative mating can have an influence(Blossfeld & Buchholz, 2009; Greenwood et al., 2014). First of all if there arise more educationally homogenous couples, this creates a different income potential for these couples. Higher educated couples will be at the advantage if this also leads to higher income. While there is no precise return of an extra year of schooling, most studies do agree on the fact that more education has a positive effect on income (Card, 1999; Blundell, Dearden, & Sianesi, 2001; Harmon, Oosterbeek, & Walker, 2003; Leigh & Ryan, 2008). For instance Harmon, Oosterbeek and Walker (2003) claim in their paper that there is no doubt about the positive effect of schooling on income. Although their precise return falls in the range from 6% to 15% depending on the method used, each year of extra schooling has an unambiguous positive effect (Harmon et al., 2003). A similar conclusion can be drawn from the work of Leigh (2008) in his paper about returns to schooling in Australia. He shows that obtaining a high school diploma increases income mostly by increasing the number of hours people will work, while higher education increases income through higher productivity (Leigh, 2008).
So if education has a positive effect on income, this would mean that if highly educated people cluster together their household income would be above average. The same holds the other way around, if lower educated people form a pair they will earn less than the average. This would result in higher inequality between these groups, but as stated in the introduction, previous research does not find this effect of educational assortative mating on income inequality. Richard Breen and Leire Salazar have done two studies in order to capture the suggested link, however they are unable to find it in case studies in the United States of America (USA) and the United Kingdom (UK) (Breen & Salazar, 2010, 2011). They claim they are unable to find a link because education actually does not provide a valid indicator for future earnings in their sample. Therefore the educational sorting did not affect the income inequality in their studies. Both countries did experience a rise in inequality but this was caused by large income heterogeneity within the educational groups themselves rather than between the different groups (Breen & Salazar, 2010, 2011).
Another research, conducted in Denmark, did find a link between the change in assortative mating and a rise in income inequality (Breen & Andersen, 2012). However it was not a change in preference of people to start choosing a partner who matched their educational level. The change in educational sorting arose because of a change in the level of education of the entire population. This made different combinations of relationships possible (Breen & Andersen, 2012). So for example it was not the case that men actively started to choose to a partner with the same educational level as themselves. Rather it was the case that there were more women with a higher educational level available. Breen continuous that he now is able to find a link between educational sorting and income because there are differences in the labour markets between Denmark and the USA and UK. Within a regulated labour
market, as the Danish is, education will be a better indicator for income compared to the unregulated labour markets of the USA and UK (Breen & Andersen, 2012). He states that the social structure of the Danish welfare system, which makes it a regulated labour market, allows women to work, even when rearing a child. This way educational sorting leads to lower variation in income within the educational groups, since both the man and woman are working. However this does create larger differences between the educational groups, ergo more overall income inequality (Breen & Andersen, 2012).
Next to the different income potential of couples it could also be the case that couples make different choices regarding labour participation or hours worked. Es-pecially if higher educated workers choose to work more than their less educated colleagues this would worsen the inequality (Blossfeld & Buchholz, 2009). In Is-rael researchers confirmed this relationship between the number of hours worked by households and income inequality. While their primary goal was to find effects of ed-ucational assortative mating, they emphasise the importance of taking into account the actual hours worked by the different groups, because it also explains a part of the increased inequality (Druker & Stier, 2014). Next to increasing inequality, the vari-ation in hours worked could possibly explain some of the large heterogeneity within the educational groups, as mentioned by Breen and Salazar (2010, 2011). Worner (2006) has drawn similar conclusions from his research in Australia. He stated that changing labour supply patterns accounted up to one third of the increase in income inequality. Next to an increase of the inequality caused by educational sorting.
So current research is ambiguous regarding the effect of assortative mating on income inequality. In the case of assortative mating some researchers do find a small relationship, but most of the time they do not find a link altogether. In the case of labour participation there seems to be some consensus on the link between a change in hours worked and a rise in inequality, but overall convincing work is still absent. This gives reason to examine the Netherlands, because it has some interesting features. First of all it is a regulated labour market like the Danish, which creates possibilities for women to work. Although instead of all those women working full-time, most of them work part-time. Thereby creating a large variety in hours worked by couples. These factors together could provide a basis for finding a significant effect of assortative mating on income inequality.
3
Data
The data that I use for this research stems for the ”Arbeidsaanbodpanel” (labour panel). It is collected by Statistics Netherlands, the Dutch bureau responsible for collecting national statistics. They started collecting in 1985 and afterwards con-tinued by collecting data every even year. They collected this data first of all to research labour participation of the Dutch population. Therefore there are data available on wages, education and hours worked by the participants. Since the re-searchers were also interested in family patterns they tried to collect their data on a household level. Therefore there are also data on spouses education and labour choices. For my comparisons I use their first collection of data in 1985 and their most recent one in 2010. As can be seen in table 1 the samples are about the same size and the number of men and women remains roughly the same.
Table 1: General statistics
1985 2010
No. Obs. 3774 3727
Men Women Men Women Percentage (Std. error) 50% (0.008) 50% (0.008) 48% (0.008) 52% (0.008) Age 39.4 (0.254) 37.8 (0.247) 47.4 (0.285) 46.1 (0.264) Schooling 2.77 (0.022) 2.46 (0.022) 3.09 (0.020) 3.04 (0.019) Wage e1058 (17.877) e617 (31.895) e2211 (27.517) e1331.38 (20.150) Job 82% (0.009) 40% (0.011) 83% (0.009) 72% (0.010) Hours 40.5 (0.207) 26.7 (0.468) 36.7 (0.227) 24.4 (0.267) Differences Men Women Age 8*** 8.3*** Schooling 0.32*** 0.58*** Hours -3.8*** -2.3***
In brackets are the standard errors of the statistics. The top of the table shows the general statistics for men and women in both 1985 and 2010. Wage is average monthly after-tax wage. The number of hours worked is the average of those who are actually working, so not of the total number of men or women. The bottom part shows the differences between the years for men and women. All differences are tested for significance with a Wilcoxon rank sum test, since none of them are normally distributed. *** significant at the 1% level.
38 to 46 between the two samples. This could be caused by the fact that the average age of people increased between 1985 and 2010 in general. In 1985 it was 35 and in 2010 it was 40 (CBS, 2015). Since the collectors of the data state they sample at a representable level of that time, it could cause them to sample at a higher age in 2010 than in 1985. Hence I assume the difference in age between the samples should not lead to any biases. Also the age of the participants will most likely not affect how people choose their partner, thereby not effecting assortative mating.
In order to make comparisons on education, a participant is pooled on the edu-cational level of his or her highest diploma. They then fall in one of four possible categories: No education/missing or only preschool (1), lower high school (2), higher high school (3) and higher education (4). In the Netherlands there are different lev-els of high school, leading to the two different high school groups. Higher education refers to both higher vocational education and university. It can be seen that the educational level rose for both men and women, and both increases are significant. Women increased their educational level more than men, thereby almost catching up to an identical level. However, in 1985 as well as in 2010 men are significantly higher educated than women.
Average wages differ substantially between men and women in both years, part of this can be explained by the fact there are more men working and they work more hours. Wages out of labour are used as income for households since these possess the strongest link to education. Income generated out of other sources like rents, benefits or others, do not have such a direct link to education. I also took teenagers and seniors out of the sample since they are not yet, or no longer, working and therefore would have no income1. The wage used in the comparison is after tax wage. While
pre-tax wage would provide a better measure for inequality, especially because of the levelling income-tax policy in the Netherlands, I have no such data available. The levelling policy would not be a problem if it remained the same between 1985 and 2010, but since tax policies most likely did change I use a comparison method in which the absolute values of income do not matter. This also takes care of the problem that could have been caused by the fact that wages rise with age. Since the average age has risen between the samples this could have otherwise biased the average income of households.
Lastly the Netherlands is known for its high number of people working part time, which can also be found in the data. While the number of women working rises with 30%, the number of hours worked by those who are working actually decreases. Men also started working less and both decreases are significant.
4
Educational sorting
The first question which needs to be answered is whether there actually is an in-crease in educational sorting in the Netherlands. As stated before I use four levels of education someone can acquire. This creates twenty possible combinations of households, of which the distribution is presented in table 3. On the left side there is the education of the head of the household and in the middle the possible levels of education of his or her spouse. On the right hand side of table 3 there is the percentage of the people who remain single.
Table 2: Kendall’s tau & Fisher statistics
1985 2010 Difference No. Obs 1592 1347 Z’10-Z’85 0.091 Kendall’s tau 0.287 0.368 SE 0.037 Fisher’s Z 0.295 0.386 Fishers score 2.460**
The left side shows the Kendall’s tau b for each year and on the right side the Fisher Z statistic is calculated in order to check significance. ** significant at the 5% level.
As can be seen the percentage of households on the diagonal (which indicates the inhabitants have the same level of education) increases between 1985 and 2010. Also the number of singles and the general level of education increases. In order to assess whether the increase in couples on the diagonal has been significant I have conducted two different tests. Firstly a t-test on the number of people on the diagonal versus those off the diagonal. This is reported in the bottom of table 3, and this difference is significant. So there are more couples with the same level of education in 2010 than there were in 1985.
The second method I used to test educational homogamy is Kendalls tau. Kendalls tau is a measurement that calculates whether two ordinally ranked variables have a correlation. Its interpretation is straightforward in that it can take on a value between -1 and 1. A negative value of tau indicates that if the rank of X increases the rank of Y will decrease. If tau is positive an increase of X will be accompanied by an increase of Y. This allows to test the hypothesis, with as null hypothesis that there is no association between the two variables (Worner, 2006).
In table 2 the Kendall’s tau values for 1985 and 2010 are presented2. In 1985
and in 2010 the value of tau is positive, suggesting that if I were to randomly draw a couple out of the sample, there is a positive chance their educational level is similar rather than dissimilar. But in order to be able to calculate whether the value of 2010 is significantly higher, the taus need to be transformed to Fisher z statistics. After this is done the two z-values can be compared and this shows that the increase of the tau value is significant3. So the chance increased that the education of the
spouses is similar.
2These are tau b values, since there are quite a lot of ties in my data and Kendall’s tau a does
not correct for that.
3Following Worner (2006) I calculated the Fisher Z values by Z = 1
2 ln(1 + τ ) − ln(1 − τ ).
These are then divided by the standard error σ1− σ2=
q
1 n1−3 +
1
n2−3. Since this score is larger
T able 3: Educational homogam y 1985 Sp ouses Ed ucation Education head Pre sc ho ol Lo w er high sc ho ol Higher high sc ho ol Higher education No partner Pre sc ho ol 5% 2% 2% 0.4% 4% Lo w er high sc ho ol 3% 5% 4% 1% 5% Higher high sc ho ol 5% 10% 14% 3% 11% Higher education 2% 4% 8% 4% 7% T otal 2201 Diagonal p ercen tage 38% 2010 Pre sc ho ol Lo w er high sc ho ol Higher hig h sc ho ol Higher education No partner Pre sc ho ol 0.3% 0.5% 0.7% 0.2% 1 % Lo w er high sc ho ol 0.7% 7% 5% 1% 6% Higher high sc ho ol 0.6% 6% 11% 5% 16% Higher education 0.1% 4% 7% 12% 1 5% T otal 2202 Diagonal p ercen tage 49% Difference on the diagonal 0.108*** The division of households o v er the p ossible educational lev els. The diagonal p ercen tage is not of the to tal amoun t of households but only of the households w ho are a couple. The difference b et w e en the n um b er of households on the diagonal b et w een 1985 and 2010 is tested with a t-test. *** significan t at the 1% lev el.
5
Inequality analyses
So there is increased assortative mating in the Netherlands, but does it affect the inequality? To answer this question inequality will need to be determined, which can be done via a large variety of measures. The most known and used one is the Gini coefficient, which calculates the surface between Lorenz curve and the line of equality (45 degree line). However the Gini coefficient does not distinguish between group specific inequality and inequality between different groups. With the use of ratios this would be possible, for instance the 90:10 ratio, which compares the income of the 90th percentile with that of the 10th. This ratio on the other hand does not take into account the entire population, therefore not measuring overall inequality. The Theil index does take account of these properties and is therefore chosen in this analysis to measure inequality. The basic Theil index can be written as
T = 1 N n X i=1 xi ¯ xln xi ¯ x (1)
where N is population size, xi income of household i and ¯x average income of the
entire population. So the ratio of household income to average income is multiplied by the log of that ratio. In order to distinguish within group and between group differences of inequality it can be rewritten as
T =X j pj ¯ xj ¯ xln ¯xj ¯ x +X j pj ¯ xj ¯ xTj (2)
here the left hand side of equation 2 is the between group component and the right hand side the within group component. pj is the fraction of the population in group
j, ¯xj is the mean income of group j and Tj is a group specific Theil index of group
j. The j’s represent the different educational groups, in my case the 20 possible combinations of households. The group specific Theil is
Tj = 1 nj nj X i=1 xi|j ¯ xj ln xi|j ¯ xj (3)
where nj is the number of households in group j and xi|j is income of household i in
group j. The Theil index as represented by equation 2 has several nice properties. First of all it is defined as a general entropy measure, meaning it calculates the deviation from perfect equality. It can take on a value between 0 and 1. With 0 as perfect equality where every household has the same income and 1 as perfect inequality with one household having all the income. Secondly it also treats all the incomes the same, creating symmetry. It has a homogeneity of degree zero, meaning that if all incomes are multiplied by the same number, the index does not change. Lastly it also follows the Pigou-Dalton principle by which a transfer from a richer household to a poorer one, but which leaves the order of households on income the same, lowers the index value.
To make the comparison between 1985 and 2010 one final adjustment must be made to the formula:
T = X j pj ¯ xj P jx¯jpj ln ¯ xj P jx¯jpj +X j pj ¯ xj P jx¯jpj Tj (4)
Instead of ¯x, the average income is represented by P
jx¯jpj. This way there are
three variables which can take on year specific values: ¯xj, Tj and pj. If I change one
variable to its 2010 value while keeping the others at their 1985 level, I can isolate what changes that variable has had on the inequality. For instance if I change Tj to
its 2010 value the formula becomes:
T =X j p1j ¯ x1j P jx¯1jp1j ln ¯ x1j P jx¯1jp1j +X j p1j ¯ x1j P jx¯1jp1j T2j (5)
This way the specific effect of the within group inequality on total inequality can be measured. The same can be done with changing the average income of the groups.
T =X j p1j ¯ x2j P jx¯2jp1j ln ¯ x2j P jx¯2jp1j +X j p1j ¯ x2j P jx¯2jp1j T1j (6)
But the variation of the formula which gives answer to the effect of assortative mating is equation 7. T =X j p2j ¯ x1j P jx¯1jp2j ln ¯ x1j P jx¯1jp2j +X j p2j ¯ x1j P jx¯1jp2j T1j (7)
Here the pj takes on its 2010 value, while ¯xj and Tj remain on their 1985 value.
This way it can be seen what the effect would be if income had remained the same, but the group composition would be allowed to change. This isolates the effect assortative mating has had on income inequality in the Netherlands.
In the data section it is stated that wage will be used as the measure for income. Since I’m researching households this means that in the case of couples where both partners earn wages, these wages are added to calculate their total household income. This could create biased inequality measures between couples and singles. In order to correct for this I divide income by the square root of 2 for households which are a couple. This because costs will not double when living together compared to living alone, it could be said there are some economies of scale.
The respective Theil indices can be found on the left hand side in table 44. When comparing the inequality between 1985 and 2010 for Income 1, it shows that inequality has decreased. This is contrary to most of the literature (Breen & Salazar, 2010, 2011; Worner, 2006). This decrease is caused mostly by the drop of within group inequality. This effect is also reflected by the Theil in line 3, where Tj takes
on its 2010 value. So the incomes of the households within the specific groups came closer to each other over time. Not only does the within group inequality change the most, it is also far larger than between group inequality. Between group inequality accounts to less than 10% of total inequality. So the income difference between the different educational groups is actually small. When turning to the effect of assortative mating, line 4 shows that if pj takes on its 2010 value there
is no increase in inequality compared to the 1985 value, there is even a decrease. So it seems that the increased sorting on education did not cause inequality in the Netherlands to rise.
There is a another factor that could influence the inequality. It arises from the fact that the formula of the Theil index does not allow for incomes to be zero.
4The p
Table 4: Theil decomposition I
Income 1 Income 2
Theil Between Within Theil Between Within 1985 0.1997 0.0159 0.1838 0.4210 0.0272 0.3938 change ¯xj 0.1908 0.0150 0.1758 0.4169 0.0483 0.3686
change Tj 0.1315 0.0159 0.1156 0.3934 0.0273 0.3661
change pj 0.1901 0.0121 0.1780 0.3906 0.0161 0.3745
2010 0.1234 0.0102 0.1132 0.3465 0.0300 0.3165
The total Theil index is the sum of the between and within inequality of each line. The index is calculated with two different measures for income: Income 1 is regular income and for Income 2 all incomes of zero are set to one. The groups are defined by the educational level of the couples.
Since there are households who earn no wage and therefore have zero income in the sample, this should be accounted for. To do this I run a second calculation with the income for these households set to 1. This way they can be taken account for in the Theil index. This new Theil index is presented on the right hand side of table 4. The inclusion of households with no income causes inequality to double. This is somewhat predictable as there now are households included who earn a lot less than the earlier average, thereby increasing inequality. Most of the increase therefore stems from an increase in within group inequality. However since the inequality between the groups also doubles, this suggests that the households with zero income are not evenly spread over the different educational groups. With most of them in the groups of single lower educated households. Still, between group inequality only explains 10% of total income inequality. Also there is again no effect of educational assortative mating. In line three the inequality decreases compared to line one, similar to the left side of table 4.
So for both calculations of income there is no increase in income inequality caused by the increase in educational assortative mating. This is in line with the work of Breen and Salazar (2010, 2011). It could be the case that this lack of relationship between educational assortative mating and income inequality is caused by the other finding of Breen and Salazar in the USA and the UK, that there actually is no link between education and income. However, as can be seen in the appendix, when income is regressed on education, higher education does lead to higher income. So there is no reason to assume education does not influence future earnings in a positive manner.
6
Labour participation
Next to the educational assortative mating, Worner and Druker mention in their paper that it is also important to look at the number of hours worked by couples in order to grasp the full effect assortative mating could have on income inequality (Worner, 2006; Druker & Stier, 2014). Part of this influence will be captured in ¯
xj, since the income of the households will change with a change in the number of
hours worked. But couples of the same educational level might make different choices regarding labour participation, which can not be seen in the previous calculation of the Theil index. Therefore a closer look into these labour decisions could provide
1985 1990 1994 1998 2002 2006 2010 20 25 30 35 40 Year Hours Men: relationship Men: single Women: relationship Women: single
Figure 1: Hours worked per week
different results regarding income inequality.
When looking at the number of hours worked by both men and women, there has been a significant decline between 1985 with 2010. This can be seen in figure 1, where the average number of hours worked by men and women is displayed. Their averages are split between being in a relationship or being single. In general it can be seen that men work more than women and that for men the workweek decreased from 40 hours per week in the 80’s to one of 36 hours from the 90’s onward. Women work less than this, with single women working more than those in a relationship. On one hand this could suggest a more traditional family pattern, in which the woman starts to work less when she enters in a relationship in order to take care of the children. But, as stated in the previous section, it could also be the case that the costs for someone living on his or her own are higher than that of a single individual in a relationship. Thereby giving the couple the possibility to work less individually than those who remain single. If the second reasoning is true it would be expected that for men one would observe the same result, with single men working more than those with a partner. This is however not the case, since men in a relationship work significantly more than those who remain single. Thereby suggesting men still take up the role as provider in a relationship and women take care of the household/children and start to work less.
There could be another reason that would cause single women to work more than women in a relationship. If these single women have children living with them, this would give them cause to work more because they have only one income to feed everybody. This could also explain why single men work less than their partnered counterparts, because if the children are living with their mother, the costs for single men will be lower. Thereby giving them the possibility to have more spare-time at the expense of foregone income. However this line of inquiry is not pursued in this paper but could be interesting for further research.
This view on the division of work might be a too simplistic one, since there are far more women working in 2010 than there were in 1985. This is represented in figure 2, where the percentage of people working is displayed per year. A distinction is made between working full-time and part-time. This because in the Netherlands a
1985 1990 1994 1998 2002 2006 2010 0 20 40 60 80 100 Year Women, Relationship
Unemployed Part-time Full-time
1985 1990 1994 1998 2002 2006 2010 0 20 40 60 80 100 Year Women, Single
Unemployed Part-time Full-time
1985 1990 1994 1998 2002 2006 2010 0 20 40 60 80 100 Year Men, Relationship
Unemployed Part-time Full-time
1985 1990 1994 1998 2002 2006 2010 0 20 40 60 80 100 Year Men, Single
Unemployed Part-time Full-time
Table 5: Couples labour participation decisions
1985 Spouses’ choice
Heads’ choice Full-time Part-time Not working Total Full-time 200 327 779
Part-time - 20 86
Not working - - 180
1592
2010
Full-time Part-time Not working Total Full-time 92 587 284
Part-time - 111 151
Not working - - 122
1347
The different combinations of households regarding labour participation. The lower right columns remain empty since they are the mirror image of the upper left columns.
Table 6: Theil decomposition II
Income 1 Income 2
Theil Between Within Theil Between Within 1985 0.1738 0.0372 0.1366 0.3472 0.1263 0.2209 change ¯xj 0.1731 0.0354 0.1377 0.3513 0.1531 0.1982
change Tj 0.1078 0.0372 0.0706 0.2798 0.1263 0.1535
change pj 0.2166 0.0415 0.1751 0.3450 0.1128 0.2322
2010 0.1115 0.0344 0.0771 0.2376 0.1303 0.1073
The total Theil index is the sum of the between and within inequality of each line. The index is calculated with two different measures for income: Income 1 is regular income and for Income 2 all incomes of zero are set to one. The groups are defined by labour participation of the couples.
large degree of the workforce works part-time. In order for someone to be considered to work full-time someone needs to work more than 36 hours per week. In 1985 more than 60% of the women in a relationship were not working at all and almost 40% of the single women, compared to 16% and 25% respectively for men. These numbers decreased over time for women, settling around 27% in 2010, while they remained roughly the same for men. In the data section it was already mentioned that this large increase in the number of women working did not lead to more hours worked by them. This can be explained by the large number of women working part-time. Almost 50% of the women who are single and more then 60% of the women in a relationship work part-time in 2010. Since there are no major changes in the number of men who become unemployed, this created a big shift in the number of dual-earner households. This rise in female employment will increase the income of couples and this could have consequences for inequality in the Netherlands.
To access whether the changes indeed lead to more inequality I run an analyses similar to the one of assortative mating in the previous section. However this time couples are grouped on whether they work part-time, full-time or not, instead of on
their educational level. This results in six possible combinations, which are displayed in table 5.
Between 1985 and 2010 the biggest change occurred in the group in which one of the spouses works full-time and the other does not, a classical division of labour in which the man worked while the woman remained at home. This group more than halves, mostly in favour of the group in which one of the spouses works full-time and the other now works part-time. A more unlikely change is the decrease in the group of couples who both work full-time. This group also halves, so either most of the women who are working in 2010 do not work full-time or if they do, fewer of them enter in a relationship. These new groups allow for a similar comparison of the Theil index as the one made with the educational groups in the previous section.
The results for this calculation of the Theil index can be found in table 6. I follow the same principle as with the educational sorting in that I use two different calculations for income. Once again for income 2 the incomes of zero are set to one, in order to take them along in the Theil index. First of all it would be expected that although the groups now have changed, the Theil index in 1985 and 2010 should be the same as in the previous calculation. This is however not the case and it can be explained by the fact that they are calculated using a different sample. For this calculation of the Theil index the people who are single are not taken into account, for the first calculation they are.
Without the adjustment of income, so on the left of table 6, the changing of the groups over time does indeed increase inequality. With the pj set to its 2010
value, the Theil index increases from 0.1738 to 0.2166 on the left side of table 6, an increase of almost 25%. Since in the end the inequality is lower in 2010 than it was in 1985, this suggests that the rise in female employment counteracted the decrease of income inequality. However the right side of table 6 contradicts this result, since here the change in pj does not lead to higher inequality. Therefore
there seems to be no clear effect of the change in labour participation decisions by couples on income inequality in the Netherlands. With the labour participation to define the different groups there is more between group inequality. In the case of income 1 it now accounts to 20% of inequality and for income 2 it accounts to 30%. This rise compared to the previous analyses is understandable, since these groups by definition, work a different number of hours. While the effect of education on income may be ambiguous, working more hours will definitely increase income. So couples who both work full-time will on average earn more than a couple who both work part-time, ergo higher between group income inequality.
7
Discussion and Conclusion
Theoretically there are reasons to believe educational assortative mating could lead to an increase in income inequality (Blossfeld & Buchholz, 2009; Breen & Salazar, 2010). Previous research was ambiguous in this result, although most of the studies concluded there was no effective relationship (Breen & Salazar, 2011; Breen & An-dersen, 2012). Despite this lack of proof, the assumption was that the Netherlands could prove to have qualities suitable for a significant result. With its regulated labour market and high part-time employment it could possess the necessary in-gredients. Also educational sorting did take place among couples and it increased over the years. However the Netherlands is no exception and assortative mating
did not lead to higher inequality. The second relationship about the change in the number of hours worked did start of promising. There was a large change in female employment, leading to more dual earner households. This shift even increased in-equality without the incomes of zero, but when these households were included there remained no convincing result.
Possibly the fit of the data could be improved, since the data for this research was not collected with this specific research in mind. Also in 2010 the data was partly collected online, these self-reported values could therefore be biased towards socially preferred answers. For future research it would therefore be interesting to work with data specifically for this type of research. The problems could be circumvented for instance by using income data from the income-tax department of a national government. This would also take care of possible problems caused by having to use after-tax incomes.
Lastly the role of children was not taken into account in this research. They could have a large influence, because they could affect the labour participation de-cisions of both singles and couples. Secondly the assortative mating could also have consequences for the decisions of couples to have children in the first place. If for instance higher educated individuals choose to have less children and later on in life, the fact that there now are more homogenous couples could create differences between the educational groups. Because two higher educated partners could there-fore make a different choice than two lower educated partners. This problem could even have further consequences if parents of different educational backgrounds raise their children differently. These issues could grow problematic but are outside the scope of this research and data.
Concluding, also in the Netherlands there seems to have been no effect of the increased educational assortative mating and increased female employment. If the Dutch government where to try and reduce income inequality even further, the focus could best be at reducing the inequality within the different groups since this is the biggest cause of income inequality. The inequality between the different groups only causes a small part of total inequality. So the most efficient policy in the Netherlands would be to give benefits to single households in order to get them to their group average. This will reduce the within group inequality and consequently also reduce the national income inequality.
References
Blossfeld, H.-P., & Buchholz, S. (2009). Increasing resource inequality among fam-ilies in modern societies: The mechanisms of growing educational homogamy, changes in the division of work in the family and the decline of the male breadwinner model. Journal of comparative famlilie studies, 603-616.
Blundell, R., Dearden, L., & Sianesi, B. (2001). Estimating the returns to education: Models, methods and results. Centre for the economics of Education.
Breen, R., & Andersen, S. H. (2012). Educational assortative mating and income inequality in denmark. Demography, 49 , 867-887.
Breen, R., & Salazar, L. (2010). Has increased womens educational attainment led to greater earnings inequality in the united kingdom? a multivariate decom-position analysis. European Sociological Review , 26 (2), 143-157.
Breen, R., & Salazar, L. (2011, November). Educational assortative mating and earnings inequality in the united states. American Journal of Sociology, 117 (3), 808-843.
Card, D. (1999). Handbook of labor economics (Vol. 3). Elsevier Science. (Chapter 30)
CBS. (2015). Bevolking: Kerncijfers. Retrieved from http://statline.cbs.nl/ Statweb/publication/?DM=SLNL&PA=37296NED&D1=22-24&D2=35,60&VW=T (Retrieved on 26-08-2015)
Druker, E. H., & Stier, H. (2014). Educational assortative mating, womens labor force participation and inequality in israel. (Prepared for presentation at the 2014 Population Association of America annual meeting, Boston.)
Greenwood, J., Guner, N., Kocharkov, G., & Santos, C. (2014). Marry your like: Assortative mating and income inequality. National Bureau of Economic Re-search.
Harmon, C., Oosterbeek, H., & Walker, I. (2003). The returns to education: Mi-croeconomics. Journal of Economic Surveys, 17 (2).
Huber, S., & Fieder, M. (2011, July). Educational homogamy lowers the odds of reproductive failure. Plos one, 6 (7). (e22330)
Leigh, A. (2008). Returns to education in australia. Economic Papers, 27 (3), 233-249.
Leigh, A., & Ryan, C. (2008). Estimating returns to education using different natural experiment techniques. Economics of Education Review , 27 , 149-160.
Schwartz, C. R., & Mare, R. D. (2005). Trends in educational assortative marriage from 1940 to 2003. Demography, 42 (4), 621-646.
Worner, S. M. (2006, November). The effects of assortative mating on income inequality: A decompositional analysis. (DISCUSSION PAPER NO. 538)
8
Appendix
Table 7: Regression of education and income
1985 2010
Lower high school 0.268*** (0.047)
0.276*** (0.092) Higher high school 0.262***
(0.042) 0.392*** (0.089) Higher education 0.516*** (0.045) 0.585*** (0.089) Constant 6.445*** (0.037) 7.084*** (0.087) Num Obs. 1761 1760 R2 0.075 0.062
Regression of the different educational levels on log income. Preschool is the basis level and all the other vaiables are dummies representing that level of education. *** significant at the 1% level.
The table shows that obtaining education above preschool significantly increases income. However the R2 is not that high, signalling that only a small fraction of the variation in income can be explained by education. This is in line with the low between educational group inequality of the Theil index in table 4.
T able 8: Av erage income ¯xj 1985 Sp ouses Ed ucation Education head Pre sc ho ol Lo w er high sc ho ol Higher high sc ho ol Higher education No partner Pre sc ho ol e 526 e 613 e 516 e 833 e 375 Lo w er high sc ho ol e 584 e 781 e 729 e 824 e 943 Higher high sc ho ol e 662 e 756 e 803 e 845 e 838 Higher education e 728 e 948 e 1175 e 1194 e 877 2010 Pre sc ho ol Lo w er high sc ho ol Higher hig h sc ho ol Higher education No partner Pre sc ho ol e 636 e 908 e 1192 e 1545 e 637 Lo w er high sc ho ol e 1112 e 1274 e 1612 e 1775 e 1003 Higher high sc ho ol e 1034 e 1502 e 1764 e 2078 e 1025 Higher education e 1025 e 2048 e 2167 e 2381 e 1791 The a v erage income of households b y their educational lev el.
T able 9: Group sp ecific The il indices Tj 1985 Sp ouses Ed ucation Education head Pre sc ho ol Lo w er high sc ho ol Higher high sc ho ol Higher education No partner Pre sc ho ol 0.1 48 0.135 0.16 3 0.058 0.177 Lo w er high sc ho ol -0.388 0.216 0 .141 0.452 Higher high sc ho ol -0.105 0 .125 0.241 Higher education -0.131 0.084 2010 Pre sc ho ol Lo w er high sc ho ol Higher hig h sc ho ol Higher education No partner Pre sc ho ol 0.0 26 0.193 0.08 3 0.069 0.298 Lo w er high sc ho ol -0.107 0.103 0 .067 0.166 Higher high sc ho ol -0.090 0 .100 0.168 Higher education -e 2381 0.112 The group sp ecific Theil indices o v er differen t households. The lo w er righ t columns are e mpt y b ecause for the calculation of the Theil index I did not sp ecify for groups whether the head or the sp ouse had the highest education. This means tha t for instance head:pre sc ho o l & sp ouse:lo w er high sc ho ol and head:lo w er high sc ho ol & sp ouse: pre sc ho ol for m one catergorie togeh ter. Sharing also the same gro up sp ecific Theil index.
