1
Development of the Dutch public-private wage differential between 1979-2009
Ernest Berkhout (SEO)
Wiemer Salverda (AIAS)
Compact version for NAD2013, further work is still in progress.
Please do not cite:for citations, please use the extended version which is available for download at or
Development of the Dutch public-private wage differential between 1979-2009
Ernest Berkhouta,* (SEO)
Wiemer Salverdab (AIAS)
a SEO Economic Research, University of Amsterdam.
bAIAS, Amsterdam.
* Corresponding author at: SEO Economic Research, University of Amsterdam. Roetersstraat 29, 1018 WB Amsterdam, the Netherlands. Tel.: +31-20-5251657. Email:
Abstract:
This paper investigates the long-run evolution of the pay gap between the public and private sector in the Netherlands overa three decade period. Using Oaxaca-Blinder decompositions to correct for compositional differences between the public and private workforces, we find that the ‘public premium’ has disappeared during the 1980s and became negative in the early years of this century. Although the balance was partly restored recently when wages in education improved, the public premium was still minus 2 per cent in 2009. The results seem largely driven by wage formation in collective labour agreements and the explicit link that was applied (in some periods) between the outcomes thereof and the growth of public wages.
Keywords: public sector labour markets, wage differentials, hourly wages.
JEL-code: J31, J45.
version 20130910 (please do not cite)
1
1. Introduction
Public wage regulationis currently an important elementin economic policy in the Netherlands, as public expenditure has to be brought down in order to comply with the EU budget regulations in short-term. But public wage moderation has become a more permanent phenomenon as wages have been put on hold since 2009. Therefore it’s role should be analysed in a more long term perspective. However, little empirical evidence exists to support such a debate. Most Dutch scientific contributions cover individual years within the period that stretches from the early 1980s to the mid-2000s, although Zaidi & Alessie (1998) and Berkhout et al (2010) cover a multi-year period. General findings are similar across these studies: women do better compared to men, young do better than old, and less educated do better than higher educated. But due to the variation in definitions and populations, the exact estimates of public-private wage gaps are difficult to compare. Even more so when taking part-time workers into account, which is a necessity in the Dutch situation. Alessie & Hoogendoorn (1999) for example exclude health care from the public-privatecomparison; together with Hoogendoorn (2001),Heyma et al (2004) and Berkhout et al (2006) they also exclude military and police while Salverda et al (2001) included them. They all find a raw differential that is mostly explained by compositional differences, since the 1990s the remaining pay gap is relatively small. Berkhout et al (2010) suggest thatthe increasing importance of the individual wage component on top of collectively agreed wage scales has increased private sector wages in prosperous periods. Hartog & Oosterbeek (1993) and van Ophem (1993) use less precise wage measures in smaller datasets, but add to the literature by indicating that self-selection into the public sector did play a role.A few other contributions that use different methods are worth pointing out. Van Schaaijk (1982) comparesaverage annual earnings for 1979 by gender-age-education cells for three public subsectors (government, education, health care and other services) to the private sector. After reweighting for composition differences he also finds an advantage for government employees and for those in education and health care. In a tentative extension (van Schaaijk, (1986)) he finds no longer a public premium in 1985.Theeuwes et al. (1985) cover 1965, 1972 and 1979and compare central-government civil servants to private sector employees. They find a clearly lower market value of university-level human capital at labour market entry among civil servants, and only slightly lower values for the lower and secondary educated.
Our paper contributes to the literature of public-private wage comparisons in adding a long term perspective without restricting the analysis to subgroups and including more recent high quality data.[1] Our main finding is that the ‘public premium’ has disappeared during the 1980s and became negative in the early years of this century. This findings is robust for various estimation methods.The remainder of this paper is organized as follows. Section 2 and 3 explain the data and empirical strategy. The main estimation results are presented in section 4. Section 5presents some sensitivity tests and section 6 concludes.
2. Data
2.1The main estimation sample
The key challenge is to find reliable data that are suitable for comparing wages over a longer period. Most research focuses on relatively short periods of time, and so do the underlying datasets. Longitudinal analysis is often constrained by the quality of the older datasets, because less detail is availableor explanatory variables are missing. For the Netherlands, however, it is possible to combine data from the late 1970s up to the 2000s, where nearly the same model specifications can be used for four different decades. Statistics Netherlands (CBS) has produced similar datasets for 1979, 1989, 1996 and 2002 to facilitate analysis of wage structures. From other wage data available at CBS we have constructed the appropriate datasets for 2004 and 2009 in the same vein.
All datasets are specifically designed for the purpose of calculating hourly wages for different groups of education, age gender etc. In 1979 and 1989 the information is gathered from a stratified subsample of employers, providing wage registers and additional information per employee. In later years the datasets are constructed by merging wage information from salary registers and information on education and job level from the Labour Force Survey. For the waves of 1996 and 2002 this data merge is carried out by Statistics Netherlands, using identifying characteristics like birth date, address, company name etc. This procedure is documented (in Dutch) in CBS (2007). For the year 2004 the merge is carried out by the authors, adopting a similar procedure but making use of anonymised social security ID-numbers. This approach has first been used by Heyma et al. (2004) and is now the standard procedure.[2]The merge for the year 2009 is also carried out by the authors, in a comparable but much more complex framework. Since 2006 wage information is collected directly from the Tax Registry and made available to external researchers through a collection of primary source datasets.[3] This paper is the one of the first to present a wage analysis based on the merging of these datasets for the year 2009, using both men and women from all ages between 15-64.
For reasons of comparability the agricultural sector has been excluded from the analysis, as it was not included in the 1979 data sample.[4] The research population is further restricted to employees in the age of 15-64; self-employed, firm directors/majority shareholders, and trainees are not included. Also temporary agency workers are left out as, unfortunately, it is unknown to which sector they are actually assigned.
2.2Populationweights
In 1979 and 1989 the sampling issue was solved by Statistics Netherlands by constructing ‘self-weighting’ subsamples of the original data collection. This means that in the dataset the distribution of important stratification variables (like sector, gender, firm size and age) is equivalent to the distributions in the target population. Of course, this procedure is quite inefficient as it disregards a lot of available observations. For the later years weighting variables were calculated and supplied with the original data, which allow the calculation of correct averages for the Dutch employee population. But, after merging the salary registers with the Labour Force Survey (LFS) the original weighting variables are not valid anymore so new weights had to be constructed. For 2004 and 2009 a procedure is applied similar to the procedure used by the Statistics Office for 1996 and 2002.[5] Weights are constructed in an iterative process using five ‘weighting tables’ that describe the population in several dimensions using the following variables:
a.hours worked (20 categories) * gender
b.sector * age class (5 categories) * gender
c.age class (5 categories) * wage class (50 categories)
d.age class (5 categories) * type of contract (full-time, part-time, flexible)
e.sector (18 categories) * firmsize (3 categories)
The resulting weight variable will be able to reproduce the entire target population of Dutch employees. In the analyses these weights are combined with the number of hours worked, in order to represent the supply of labour in hours instead of persons. It allows our conclusions to be formulated in terms of ‘the average hourly wage earned in the Netherlands’.
2.3Definition of the public and private sectors
A principal definition concerns the demarcation of the public sector from the private sector: which jobs and workers belong to the former, whichto the latter and which are left out from both?In some analyses the definition may be limited to the apparatus of the public administration; see for instance BeffyKamionka (2010) who are specifically interested in the position of civil servants. But the ‘civil service’ covers only part of public administration employment as it refers to a specific legal status,while employees may also be hired by the very same state apparatusin other types of contract. The German and French civil services are good illustrations of this (Bosch, 2012; Gautié, 2012).More often the research interest is in the services that are actually publicly provided to the population as one wants to know if these are efficiently and effectively produced. This regards in the first place (public) education. In virtually all countries this activity is largely in the hands of the state, be it directly – through direct ownership of the educational institutions and the concomitant direct employment relationship of the persons involved – or indirectly – via the virtually complete financing of legally independent institutions which formally are the employers. For the same reason health care may be included in the public sector, but here international differences are much larger than for education. Health care is sometimes a national provision organisationally, performed by institutions in public ownership such as the UK’s National Health Service NHS which is entirely part of the public sector. On other occasions, however, it may be provided by legally private institutions which are financed to a greater or lesser extent through the state. One example is the role of Medicare and Medicaid in the USA. Another example, of importance for this paper, is the ZVW compulsory health care insurance system that was introduced in the Netherlands in 2006 and whichis considered as a type of social security (at least in national accounts statistics)as it covers the entire population and most of health care spending.[6]The actual production of health care services though is entirely in the hands of private institutions and individuals.[7]
In this paper we define the public sector to include the three major sectors: public administration (including public security and defence), education, and health care (including social work and academic hospitals). The public part in the financing of health care and consequently the government’s treatment of wages of employees in the health-care sector provide important arguments for including health care in the definition of the public sector. The literature differs significantly in this respect and therefore we include subsectoral differences to help fathoming the effects of this inclusion. Further societal services such as of unions, associations and churches are included in the private sector.
2.4Measuring the wage variable
Wages are made up of various elements: monetary or non-monetary, collectively agreed or individually, monthly paid or yearly. Within the monetary elements one can distinguish employer contributions to pensions, holiday allowance and other collective yearly allowances, individual bonuses etc. next to the ‘normal’ monthly paid salary. Both its scope (gross or net, in-/excluding employer contributions, fringe benefits or non-monetary labour conditions) and its time basis (hourly, weekly, annually) vary significantly between studies – usually because of data constraints. In this paper the wage is defined as the gross hourly wage excluding all wages for overtime work, but including incidental wage componentslike bonuses and employee paid social security and pension contributions. The register data do not allow to take fringe benefits into account. Given the rapid growth of part-time employment in the Netherlands and the significant differences in the structure of working hours between the two sectors and the public subsectors, it is important to include persons working part-time and focus on the hourly wage.To calculate the hourly wage we divide the calculated yearly wage by the net hours worked. It means that we correct for hours not worked because of holiday or ‘collective working time reduction’ (adv).[8]Table 1 presents summary statistics of the dependent variable for the public and the private sector in our sample. Table 2 presents the average (geometric) mean for different subsectors. In both tables weights are used to control for sample selection and for part-time jobs. Over the years wages are clearly highest in the public sector, in particular in public administration and education. But since 1996 wages in the financial sector are even higher, while retail and hotels & restaurants on the contrary contain many low-paid jobs.
Table 1Summary statistics of dependent variable (gross hourly wage), per year
Private sector / 1979 / 1989 / 1996 / 2002 / 2004 / 2009arithmetic mean / 7.46 / 11.42 / 14.30 / 18.91 / 21.16 / 24.06
arithmetic s.d. / 3.32 / 5.20 / 10.10 / 11.48 / 14.07 / 21.17
geometric mean / 6.48 / 10.41 / 12.45 / 16.53 / 18.73 / 20.74
geometric s.d. / 0.38 / 0.43 / 0.53 / 0.53 / 0.53 / 0.52
Public sector / 1979 / 1989 / 1996 / 2002 / 2004 / 2009
arithmetic mean / 8.92 / 12.09 / 16.29 / 21.40 / 22.03 / 25.38
arithmetic s.d. / 4.00 / 4.85 / 11.83 / 9.50 / 10.02 / 10.80
geometric mean / 7.65 / 11.31 / 14.98 / 19.65 / 21.07 / 23.43
geometric s.d. / 0.38 / 0.36 / 0.41 / 0.43 / 0.38 / 0.42
* weighted by (population weights x number of hours)
Table 2Gross hourly wage, per subsector per year
1979 / 1989 / 1996 / 2002 / 2004 / 2009Manufacturing / 6.52 / 11.57 / 13.57 / 17.80 / 20.83 / 22.49
Public utilities / 8.42 / 12.65 / 18.30 / 24.70 / 26.57 / 30.31
Construction / 6.77 / 10.58 / 13.22 / 16.58 / 19.67 / 21.59
Wholesale trade / 6.38 / 10.41 / 12.41 / 16.87 / 18.51 / 20.24
Retail trade / 4.93 / 7.24 / 8.89 / 11.07 / 11.58 / 13.15
Hotels & restaurants / 5.10 / 7.72 / 8.87 / 11.49 / 12.23 / 12.42
Transport & communication / 7.49 / 10.73 / 13.57 / 17.13 / 19.07 / 21.35
Financial services / 6.58 / 12.11 / 16.17 / 23.59 / 26.46 / 30.42
Business services / 6.91 / 10.45 / 11.64 / 16.88 / 19.91 / 22.51
Other services / 6.26 / 10.04 / 13.28 / 16.74 / 20.04 / 20.67
Health / 6.27 / 9.84 / 13.04 / 17.70 / 18.83 / 21.02
Public administration / 7.99 / 11.72 / 16.09 / 21.25 / 23.46 / 25.51
Education / 9.55 / 13.48 / 17.66 / 21.57 / 22.96 / 26.46
* geometric mean, weighted by (population weights x number of hours)
2.5Control variables
Generally, an increase in the number of control variables will enhance the explained part and according to GregoryBorland often reduce the differential.[9]Alternatively, the exclusion of important explanatory variables(like educational level) will also affect the outcome.The striking differences in the gender composition of the sectors and their changes over time incite us to pay special attention to the gender dimension; the same holds for educational attainment. We also include occupational levels which help to correct for possible underutilisation of human capital which may distort educational differentials.We control for differences in the shares of part-timers, in tenure and in age (using dummies for ten-year brackets and for every age under 23). We also account for organisation size because of the role of wage hierarchy and labour-market similarities;our definition of the public sector ensures that not all employees work in big organisations. Interactions are modelled between age, education and gender, and between part-time and gender.Table 3 presents the weighted means of the explanatory variables in our main estimation sample. It shows that the public sector has become predominantly female and part-time when female participation has increased rapidly since the second half of the nineties. Higher educated and elderly are also overrepresented in the public sector.
Table 3Independent variable means in the main estimation sample, per year.
Private sector / 1979 / 1989 / 1996 / 2002 / 2004 / 2009female / 0,25 / 0,29 / 0,33 / 0,36 / 0,37 / 0,38
education = elementary / 0.19 / 0.09 / 0.10 / 0.12 / 0.06 / 0.06
education =primary / 0.44 / 0.32 / 0.27 / 0.25 / 0.23 / 0.24
education =secondary / 0.16 / 0.25 / 0.46 / 0.43 / 0.45 / 0.43
education =higher vocational / 0.04 / 0.07 / 0.12 / 0.14 / 0.16 / 0.16
education =academic / 0.01 / 0.02 / 0.05 / 0.06 / 0.09 / 0.09
education=unknown / 0.14 / 0.24 / 0.00 / 0.00 / 0.00 / 0.02
job level = secondary/tertiary / 0.52 / 0.57 / 0.55 / 0.53 / 0.58 / 0.56
job level = unknown / 0.00 / 0.00 / 0.00 / 0.09 / 0.05 / 0.05
fulltime / 0.85 / 0.78 / 0.66 / 0.62 / 0.58 / 0.57
tenure / 7.3 / 7.8 / 7.5 / 7.0 / 7.7 / 7.0
age / 34.6 / 34.3 / 35.2 / 36.5 / 37.2 / 38.4
firm size <=9 / 0.18 / 0.17 / 0.20 / 0.18 / 0.20 / 0.14
firm size <=99 / 0.33 / 0.31 / 0.30 / 0.31 / 0.33 / 0.35
firm size >=100 / 0.49 / 0.52 / 0.51 / 0.51 / 0.47 / 0.51
share of sector in total workforce / .71 / .71 / .72 / .71 / .69 / .67
number of observations / 20,719 / 19,893 / 97,856 / 41,513 / 50,741 / 95,100
Public sector / 1979 / 1989 / 1996 / 2002 / 2004 / 2009
female / 0,48 / 0,52 / 0,58 / 0,65 / 0,65 / 0,68
education = elementary / 0.08 / 0.03 / 0.04 / 0.04 / 0.03 / 0.02
education =primary / 0.27 / 0.13 / 0.12 / 0.12 / 0.11 / 0.10
education =secondary / 0.23 / 0.26 / 0.43 / 0.43 / 0.41 / 0.42
education =higher vocational / 0.11 / 0.18 / 0.28 / 0.28 / 0.30 / 0.30
education =academic / 0.05 / 0.07 / 0.13 / 0.13 / 0.16 / 0.15
education=unknown / 0.26 / 0.32 / 0.00 / 0.00 / 0.00 / 0.01
job level = secondary/tertiary / 0.73 / 0.94 / 0.80 / 0.76 / 0.81 / 0.80
job level = unknown / 0.00 / 0.01 / 0.00 / 0.06 / 0.02 / 0.03
fulltime / 0.75 / 0.59 / 0.48 / 0.41 / 0.37 / 0.42
tenure / 6.7 / 7.9 / 8.0 / 7.7 / 8.8 / 8.9
age / 36.0 / 36.8 / 39.5 / 40.8 / 41.2 / 42.3
firm size <=9 / 0.04 / 0.03 / 0.04 / 0.04 / 0.04 / 0.03
firm size <=99 / 0.12 / 0.10 / 0.16 / 0.11 / 0.09 / 0.07
firm size >=100 / 0.84 / 0.88 / 0.80 / 0.85 / 0.87 / 0.90
share of sector in total workforce / .29 / .29 / .28 / .29 / .31 / .33
number of observations / 8,551 / 8,169 / 45,746 / 41,473 / 56,850 / 53,786
* population weights are used to correct for sample stratification
2.6Intertemporal comparability issues
Although the dataset is constructed with special attention to comparability over the years, not every systematic difference could be eliminated especially in the earlier datasets. In 1979 we cannot correct nominal hourly wages for differences in holiday hours. Although the effect on adjusted wage differentials is minimal, possible differences in holidays should be kept in mind. In 1989 age is not a continuous variable, but limited to five-year classes (two-year classes for ages <25). In order to be able to run the same regressions, this categorical variable is replaced by a pseudo-continuous variable containing the average age within the age class. Also the dataset for 1989 did not contain information on the annual wage components (such as the holiday allowance). We have imputed this component from external information, for each subsector separately.[10] When interpreting the adjusted wage differentials for this year, possible intra-sectoral differences in annual wage components should be kept in mind.In the available data for 2004 it appears that apprentices are not included.For every year except 2009 the information is targeted at a reference week, somewhere at the end of the year. In 2009 every job that has existed at some point during that year is recorded. In order to prevent a domination of very small jobs in 2009, we select only those jobs that existed on December 31st.We believe these minor details are not affecting the main conclusions, but there is one major issue that should be mentioned here.