NiCE Working Paper 06-105 December 2006

The inequality of the wage distribution in 15 European countries in the 1980s and 1990s

Joop Odink Jeroen Smits

Nijmegen Center for Economics (NiCE) Institute for Management Research Radboud University Nijmegen

P.O. Box 9108, 6500 HK Nijmegen, The Netherlands http://www.ru.nl/nice/workingpapers

Abstract Differences among countries and trends in inequality of hourly wages are studied for 15 European countries using additively decomposable inequality measures. We start with the decomposition of earnings inequality in the mid 1990s according to the key variables of the Mincerian model: education, age, and gender. Next, we test whether additional decomposition according to region, sector and occupation adds to the explanation of inequality on the basis of the Mincerian variables. Finally, we present trend information on earnings inequality and a decomposition of inequality according to education and age.

This paper is based on the work done by Joop Odink and Jeroen Smits as part of the EU funded TSER project Public Funding and Private Returns to Education (PURE). A reduced version of this paper appeared as Chapter 5 in the Finale Report of the PURE Project. Joop Odink is associate professor at the Faculty of Economics and Busines of the University of Amsterdam. Jeroen Smits is associate professor at the Nijmegen Center for Economics of the Institute for Management Research of Radboud University Nijmegen. Address of correspondence: Jeroen Smits, Nijmegen Center for Economics, PO.Box 9108, 6500HK Nijmegen, The Netherlands. E-mail: [email protected] Projectwebsite PURE-project: www.etla.fi/pure/

1

Introduction The distribution of income has always played an important role in economic theory and policy. In the 18th and 19th centuries the distribution of the national product over the different classes was a major issue. This distribution is known as the categorical distribution. Income shares are calculated as aggregates over all people belonging to each of the socio-economic classes. According to Ricardo, wages, profits and rents are attributed to labourers, entrepreneurs and landowners, respectively. According to Marx the struggle between the bourgeois (profits) and the proletariat (wages) determines the wage rate and the profit rate. Therefore, not only the income distribution matters, but also the remuneration or price ratio(s). In the 20th century, the distribution of incomes over persons – be it individuals, tax payers or households – became a major issue: the so-called personal income distribution. The income of individuals is the aggregate of all their income components. For many persons and households, labour income is by far the most important income component. Accordingly, it makes sense to analyse the personal distribution of wages separately. In this paper the distribution of the hourly wages is studied for 15 European countries: Austria, Denmark, France, Finland, Germany, Greece, Italy, Ireland, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom. One of the focuses of the analysis is the contribution of education and age (or work experience) to the total wage inequality. We also want to find out whether, besides these basic variables of the human capital model, other factors contribute substantially to wage inequality. Furthermore, we want to determine in what way wage inequality and the contribution of education and age to it have changed over the last decades. Hence the basic research questions of this paper are: - What is the degree of wage inequality in the 15 European countries in the mid 1990s and which part of this inequality is due to inequality between males and females and between educational and age groups within the countries. - To what extent contribute besides the standard human capital variables also differences between regions, sectors and occupational groups to the explanation of total wage inequality within the countries?

2

- To what extent has total wage inequality and the contribution of inequality between educational and age groups to it in the countries under study changed during the 1980s and early 1990s?

Wage inequality and inequality of the wage distribution Inequality has several aspects. It is necessary to make a clear distinction between wage inequality based on differences in hourly wages (or wage ratios) and inequality in the distribution of wages. Wage inequality refers to wage differences (only prices count). It can be measured using the wage equation. For inequality in the distribution of wages, both prices and quantities matter, resulting in income shares. For the measurement of inequality in wage distributions, inequality indices can be used. The following example further illustrates the difference between the two approaches. In a society with low-educated people earning 10 per hour, and high-educated people earning 20 per hour, the wage ratio is 2 irrespective of the population share of the highly educated. The inequality of the wage distribution, in contrast, will (in principle) be the higher the closer this population share is to 50%! When studying the rate of returns to education, the wage equation is the centre piece of the analysis. It is used to examine the rate of return on investments in education. An example of the wage equation for the Netherlands in 1996 is: lnWage = 1.583 + 0.063 Schooling + 0.326 lnExperience – 0.132DFemale (R2 = 0.531)

The wage equation also contains information about wage inequality: the differences in wages and wage ratios according to education, experience (or age), gender, and other factors.

Income inequality coefficients There has been a great variety in income inequality coefficients throughout the economic literature. The choice of measure has always been a tricky question. Because the income distribution is an emotional subject, many papers have been devoted to the merits and shortcomings of the different inequality indices. However, almost all authors agree about

3

three basic axioms (postulates, criteria) that a decent index or coefficient should fulfil. Specifically, the index (I) should fulfil the criteria of homogeneity, symmetry, and Pigou–Dalton. Homogeneity implies that if all incomes are multiplied by the same constant, inequality does not change. An important consequence is that inequality can be expressed as a function of income shares only. Symmetry means that a change of income between two persons does not effect inequality. According to Pigou–Dalton (see Kakwani 1980) a transfer from a high income to a low income has to reduce the value of the inequality index. Most of the existing indices do not satisfy these criteria. However, there are a few well-known which do: the Gini index, the Theil index and related indices, and the coefficient of variation. The Gini and Theil indices typically have been developed for income inequality measurement, while the coefficient of variation is a general statistical measure. If an additional criterion would be added, only a few or even none of the indices would remain. Kakwani (1980) adds measurement in a 0–1 scale as an additional criterion, which is met only by the Gini index. Foster (1983), in turn, proves that only Theil-related indices combine the three aforementioned criteria plus the additive decomposability criterion. Additively decomposable means that the index is equal to the inequality between different groups plus the sum of the weighted within-group inequalities. I = I between groups + wi* I within groups Because decomposability is an important aspect in our analysis, we choose to use the Theil indices and the Variance of log income. The last measure does not always satisfy the Pigou–Dalton criterion, but because this criterion is only violated in extreme cases, we adopt this measure as well. There are two Theil indices, Theil T using income shares as weights and Theil N using population shares as weights. Theil (1967) proves that, if the distribution of income is log-normal, the Theil T index is equal to half the variance of log incomes. The formulae of the three inequality indices are:

T=

1 n

n

yi

 yi 

  ln  

(1)

i 1

4

1 N= n

L=

1 n

n

 

 ln y 

(2)

i

i 1

n

 ln y  ln  * i

2

(3)

i 1

where yi is the income of worker i, n the size of the population,  and  * the arithmetic and geometric means of the distribution. To study the structure of earnings inequality in the countries, we conduct a decomposition analysis in which total inequality is expressed as a sum of within-group inequality and between-group inequality. The between-group component is most of interest because it shows which part of total inequality is due to the factor under study. Besides bivariate decompositions of inequality, in which inequality is decomposed according to one factor, also multivariate decompositions will be performed in which decomposition takes place according to several factors simultaneously.

Data and selections The data we use for our analyses were made available by the research groups which took part in the PURE project, an EU-sponsored TSER project (www.etla.fi/pure). Appendix 1 gives an overview of the data sets which were used. For each country at least one data set for the 1990s was available and for most countries also one or more data sets for the 1980s was available. For our analyses of country differences we wanted to use data for the mid 1990s. However, because only for eight of the 15 countries a dataset for 1995 was available, this was not completely possible. For three countries the data are for 1994, for two countries they are for 1993, for one country they are for 1996 and for one country they are for 1991. For one country (Spain) only one data set was available. For two countries, Switzerland and France, there were only data sets for the 1990s. The analyses for males are restricted to (almost) full-time working individuals (defined as persons working more than 34 hours in a week) the analyses for females include also part-timers working more than 8 hours in a week. The analyses are further restricted to persons aged 18 to 64. For most countries gross hourly earnings is used, but

5

some countries (Austria, Greece, Italy) use net hourly earnings. Further information on the data can be found in the country-specific chapters of Asplund & Pereira (1999) and Harmon, Walker and Westergard-Nielsen (2001).

Variables

For the decomposition analyses the following variables and categorisations are used: Education recoded into four categories: (1) primary, (2) lower secondary, (3) upper secondary, (4) tertiary. Given the large differences among the educational systems of the countries, the national research groups who did the recoding were allowed to use other categories, if these reflected the national educational structure better. Age, recoded into five groups: (1) 18-24, (2) 25-34, (3) 35-44, (4) 45-54, and (5) 55-64. Gender, (1) males and (2) females. Region: The national research groups were asked to make a division into four relevant areas. Private: (1) public sector, and (2) private sector. Manuf(acturing): (1) other sectors, and (2) manufacturing sector. Occup(ation): (1) white collar, and (2) blue collar. Parttime: (1) more than 34 hours a week, and (2) 34 hours or less in a week.

Figure 1. Theil T, Theil N, and Variance of log income on the basis of hourly wages for the 15 PURE countries around 1995

0,6

0,5

0,4

0,3

0,2

0,1

0 IRL94G

PT95G

UK95G

GER95G AUT93N

FIN93G

FR95G Theil T

GR94N Theil N

6

NL96G

DEN95G

Variance of log

CH95G SW E91G ITA95N

NOR95G SPA94G

Results We start with a comparison of hourly earnings inequality among the 15 countries in the mid 1990s. Figure 1 present the three inequality measures, for all respondents (1a) and separately for males (1b) and females (1c). The main conclusions that can be drawn from the figure are: - The differences between Theil T (using income shares as weights) and Theil N (using population shares as weights) are for all countries relatively small. - The variance of the log incomes measure is about twice as large as Theil T. - Spain, Greece, Portugal, Ireland and the UK are by far the most unequal countries with respect to hourly wages and Sweden is the most equal one. As a consequence of the first and the second conclusion, we shall concentrate on Theil T in the remainder of this paper.

Figure 2. Theil T on the basis of hourly wages for the 15 PURE countries around 1995

0,3

0,25

0,2

0,15

0,1

0,05

0 IRL94G

PT95G

UK95G

GER95G AUT93N

FIN93G

FR95G

GR94N

Total

Males

7

NL96G Fem ales

DEN95G

CH95G

SW E91G

ITA95N

NOR95G SPA94G

Figure 2 presents the values of Theil T separately for males and females. We see that in eight of 15 countries inequality is lower among females than among males and that the difference is relatively large in Norway and Sweden. In seven of the 15 countries inequality is lower among males, the difference being largest in Italy and Greece. In interpreting these and other gender differences in this paper, it should be kept in mind that the figures for males and females are not completely comparable because the male data are for full-time workers only whereas the female data are for full-time and part-time workers.

Decomposition analyses

When decomposing the Theil index according to gender, education, age, and the combination of these three variables, a total of four different between-group inequalities can be calculated. In Figure 3 those four between-group inequalities are expressed as a percentage of total inequality for each of the 15 PURE countries.

Figure 3. Decomposition of Theil T according to gender, education, age and a combination of the three variables for the 15 PURE countries, percentage of total inequality explained

50 45 40 35 30 25 20 15 10 5 0 IR L94G

PT 95G

U K95G

G ER 95G AU T 93N F IN 93G

FR 95G Sex

Edu

G R 94N Age

8

N L96G

D EN 95G

Sex+ edu+age

C H 95G SW E91G IT A95N N O R 95G SPA94G

The following conclusions can be drawn from Figure 3: - The combination of sex, education and age stands for about 30% to 50% of total inequality; Ireland with 48% being the highest and Denmark the lowest (28%). - The sex effect varies heavily; from almost nothing (France) to over 12% (Sweden). However, this outcome might be influenced by the composition of age and gender. - In Ireland and the Netherlands, age is more important than education; in almost all other countries the reverse is true. Figure 4a. Decomposition of Theil T of males according to education, age, and combinations of education and age with region, private sector, manufacturing, and manual/non-manual occupation, percentage of total inequality explained

55 50 45 40 35 30 25 20 15 10 5 0 IRL94G

PT95G

UK95G Edu

GER95G

AUT93N

Age

Edu+age

FIN93G

FR95G

Edu+age+region

GR94N

NL96G

Edu+age+private

DEN95G

CH95G

Edu+age+manuf

SWE91G

ITA95N

NOR95G

Edu+age+occup

In Figures 4a and 4b, the wage distributions of males and females have been further analysed by adding various variables to education and age in the decomposition analysis: region, sector (private versus public and manufacturing versus other) and occupations (manual versus non-manual). For females also fulltime versus parttime work is added. Figure 4a presents the results for males and Figure 4b for females.

9

SPA94G

Figure 4b. Decomposition of Theil T of females according to education, age, and combinations of education and age with region, private sector, manufacturing, and manual/non-manual occupation, percentage of total inequality explained

55 50 45 40 35 30 25 20 15 10 5 0 IRL94G

PT95G Edu

UK95G Age

GER95G Edu+age

AUT93N

FIN93G

Edu+age+parttime

FR95G

GR94N

Edu+age+region

NL96G

DEN95G

Edu+age+private

CH95G

SW E91G

Edu+age+manuf

ITA95N

NOR95G

Edu+age+occup

We can conclude that: - Adding more variables (region, sector, occupation) does not substantially increase the share of between-group inequality in total inequality. - This indicates that gender, education and age/experience are the top three components of income inequality between wage earners.

Trends

So far a static situation has been analysed. One of the main characteristics of the labour market in the second half of the 20th century is the increased schooling of the working population. According to the demand and supply models, one might think that wage inequality has therefore been reduced substantially. There are, however, forces that work in the opposite direction. Not only has there been an increased supply of higher educated people, but also an increased demand in the labour market for such skills. Wage differences decrease only if this “race between technological development and education”

10

SPA94G

(Tinbergen 1975, Ch. 6) is won by education. Since the eighties the rate of return on investment in education has been more or less constant in most PURE countries.

Figure 5a. Trend of Theil T for males in the 15 PURE countries

0,25

0,2

0,15

0,1

0,05

0 73

76

79

82

85

88

91

94

97

Ire land

Portugal

UK

Germ any

Austria

Finland

France

Greece

N etherlands

D enm ark

Switserland

Sweden

Italy

N orw ay

Spain

11

Figure 5b. Trend of Theil T for females in the 15 PURE countries

0,25

0,2

0,15

0,1

0,05

0 73

76

79

82

85

88

91

94

97

Ireland

Portugal

UK

Germany

Austria

Finland

France

Greece

Netherlands

Denmark

Switserland

Sweden

Italy

Norway

Spain

12

As long as wage ratios are constant, an increase in the population share of the highly educated from a low level to a substantial level will increase between-group inequality (see above). Furthermore, within-group inequality is generally highest in the groups with the highest wages. As the shares of these groups are increasing, the weighted sum of the within-group inequalities will rise as well. Figures 5a and 5b show the trends in wage inequality for 14 of the 15 PURE countries. These figure show that for males in most of the countries wage inequality has been more or less stable in the course of the 1980s and early 1990s, but has increased in Portugal, Ireland, Italy, UK and Greece and decreased in the last period in France. For females we observe an increase in Portugal, Ireland, UK and Italy and a decrease in Switzerland, Austria, and France (last period).

Some consequences for incomes policies

Three main groups of wage differences with respect to different political issues can be distinguished: - compensating differences - differences based on productivity differentials - differences based on imperfect market conditions. For socialists there is no problem if wage differences compensate for differences in effort or in the quality of the work (dirty, unpleasant). Wage differences based on productivity differences, in contrast, might be interpreted by them as being unfair. Liberals are in favour of good functioning markets. They will argue that productivity differences should be reflected in wages. If not, serious inefficiencies might be the result. Both socialists and liberals are in favour of elimination of differences based on imperfect market conditions. Therefore, in many countries the equity efficiency trade-off is a major political issue. What about wage differences related to the big three: education, experience, age and gender? If the rate of return on investment in education reflects the reference discount rate (e.g. a market interest rate corrected for (wage) inflation, uncertainty and the quality of the job), then we might argue that education-induced wage differences are compensating differences. Differences in experience reflect work done in the past,

13

implying that those differences might be interpreted as a compensation for this past effort. Lifetime wage incomes can be calculated by discounting wages over time. If the rate of return on investment in education happens to be the discount rate, then the differences arising from education and (calculated) experience will disappear out of the distribution of lifetime wages. The same is true for age-induced differences. So far we have been working with age differences and not with differences in (estimated) experience. However, in groups with equal age and education the calculated experience will also be the same. We can therefore state that the education-related wage differences found in the PURE datasets reflect one-third to one-half of wage inequality whether measured by indices or by wage equations. A substantial part of those differences might be interpreted as compensating wage differences. This result, however, does not mean that there is no task for the policy makers. The race between technology and education has not ended yet. The demand for higher educated workers continues to grow. The important task of the government is to stimulate education. If the supply side lags behind demand, this may lead to increased inequality, and also to problems between supply and demand that might generate substantial unemployment. The differences in inequality due to gender are substantially smaller than the differences in mean wages between genders. However, the resulting differences will probably still not be acceptable to politicians.

14

References

Asplund, R. & Pereira, P. T. (1999), Returns to human capital in Europe: A literature review. Helsinki: ETLA. Foster, J.E. (1983), ‘An Axiomatic Characterisation of Theil Measure of Income Inequality’, Journal of Economic Theory 31, 105–121. Harmon, Colm, Ian Walker and Niels Westergard-Nielsen (2001), Education and Earnings in Europe: A Cross Country Analysis of Returns to Education. Edward Elgar Publishing Ltd. Kakwani (1980), Income Inequality. Oxford. Theil, H. (1967), Economics and Information Theory. Amsterdam: North-Holland. Tinbergen, J. (1975), Income distribution. Amsterdam: North-Holland.

15

Appendix 1. Data information Country Austria

Denmark

Finland

France

Germany

Greece

Ireland Italy

Netherlands

Norway

Spain

Year 85 89 93 97 81 85 90 95 84 87 89 91 93 93 94 95 96 97 98 84 87 89 91 93 95 97 74 88 94 87 94 87 91 95 79 89 96 80 87 95 94

Wage Net Net Net Net Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Net Net Net Gross Gross Net Net Net Gross Gross Gross Gross Gross Gross Gross

N males 7645 7272 6620 5149 3898 4087 4114 4240 2223 1849 2064 1943 1140 13497 13781 13746 13992 13473 13290 1815 1500 1579 1354 1373 1356 1279 2164 1830 2083 1861 1553 3517 3790 3219 14293 12573 5730 1012 975 856 2370

16

N females 4584 4461 4373 3591 3357 3704 4043 3991 2212 1913 2052 2063 1287 11328 11572 12006 12122 11986 12101 1125 955 1057 992 981 976 947 917 1167 1489 1427 945 1862 2320 2302 5234 6728 4231 838 988 893 1247

N total 12229 11733 10993 8740 7255 7791 8157 8231 4435 3762 4116 4006 2427 24825 25353 25752 26114 25459 25391 2940 2455 2636 2346 2354 2332 2226 3081 2997 3572 3288 2498 5379 6110 5521 19527 19301 9961 1850 1963 1749 3617

Country Port

Switzerland

Sweden UK

Year 82 86 91 95 92 95 98 81 91 80 85 90 95

Wage Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross Gross

N males 7630 6337 6647 6793 3730 5043 3295 1637 1514 3583 3121 2912 2484

N females 3338 3233 4055 4660 3012 4337 2913 1570 1591 2706 2478 2670 2513

N total 10968 9570 10702 11453 6742 9380 6208 3207 3105 6289 5599 5582 4997

For further information on the data, see the country-specific chapters of Asplund & Pereira (1999) and Harmon et al (2001).

17