Evolution of the Public-Sector Wage Premium in Poland

The aim of this paper is to estimate the adjusted sectoral wage gap in Poland from 1999 to 2012. We use a set of individual data from a survey on the structure of earnings by occupation (SEO) carried out by Poland’s Central Statistical Office (CSO) every two years for businesses with more than nine employees. We apply quantile regression and the Ñopo decomposition method to address several methodological problems of wage differential analysis. We show that, after controlling for structural differences in employment, there is no clear trend in the evolution of the adjusted public-sector wage premium in Poland in recent years. The parametric approach indicates a positive and growing premium, with significant variations across different parts of the wage distribution. The non-parametric approach yields different results, indicating a negative premium with no clear trend in the 1999–2012 period, with a declining public-sector wage penalty in recent years. This means that the Polish labour market is becoming increasingly similar to its counterparts in developed countries in this respect1.


Introduction
Despite the rapid growth of employment in the private sector during the last quarter-century, the public sector remains the biggest employer in Poland: in the first quarter of 2015 it employed 3.8 million individuals, which accounted for 24.2% of total employment.The role of the public sector in labour-market performance goes well beyond its borders, with the wage-setting mechanism being one of the main channels of its impact on other sectors of the economy.Consequently, it affects the overall competitiveness of the economy, the inflation rate and other macro-economic variables.The dynamics of public-sector wages is also crucial for the equilibrium of public finances, as wage bills remain one of the biggest expenditure items in both central and local government budgets.
Although, in the public discourse, public-sector employees often complain about being underpaid, it is an observed statistical regularity that the average wage in the public sector exceeds the average wage reported in the private sector.According to the CSO, the average gross monthly wage in Poland's private sector in the first to third quarters of 2015 was PLN 3,723.12,compared with PLN 4,415.01 for the public sector [CSO, 2016].However, such a simple comparison of raw average wages could lead to a misleading conclusion that the public sector offers higher wages than the private sector.Generally, individuals' wages exhibit considerable heterogeneity due to large differences in the personal characteristics of earners that are relevant for the labour market (e.g. level of education, experience, etc.).Due to the specific features of its services, the public sector attracts individuals with a higher level of education on average than the private sector.Also, the average age and firm-specific job experience are much higher for public-sector employees.Moreover, public entities are usually bigger and often operate in less competitive markets (i.e. they have higher monopoly power) than private firms.Nevertheless, the question remains: to what extent is the observed difference in wages between the private and public sectors a result of differences in the characteristics of employees?What is the role of other factors such as firm size or its area of activity?And finally, what is the direction and size of the adjusted public-sector wage premium that can be attributed to the unobservable characteristics of employees, firms or institutional differences between the two sectors?Any significant adjusted wage premium would be indirect proof of non-wage benefits offered to workers or significant obstacles to labour force mobility (job switching costs) that may adversely impact economic efficiency and growth.
There are a number of possible reasons why earnings differentials exist between the private and public sectors.While profit maximization is commonly regarded as the main goal of private companies, a large part of the public sector is focused on delivering public goods and services, the redistribution of wealth and the realization of social and political goals.Consequently, the wage formation mechanism in the public sector is to a large extent regulated by political considerations, while in the private sector it is strongly determined by profit constrains.In effect, wages in the public sector are poorly correlated to workers' marginal productivity [Fogel & Lewin, 1974].However, the co-existence of both sectors on the same labour market (competition for workers) makes the public sector subject to economic mechanisms as well [Lausev, 2014].The wage structure in the public sector is more compressed, which rewards less skilled workers while putting restrictions on high-skilled salaries, especially those of managers.Moreover, the principal-agent problem might be more relevant for the public sector.Better monitoring of efforts could reduce the need to pay more, but in the public sector it is not always clear who the owner is and who should monitor the effects [Bebchuk & Fried, 2004].
The public and private sectors differ in the structures of their economic activity.Industries where the public sector dominates are characterized by relatively stable demand for their services, which is often determined by political decisions rather than market forces 2 .Public-sector wages are less related to the business cycle, while private-sector wages seem to be strongly pro-cyclical.As a result, public-sector relative wages are counter-cyclical: they increase in times of economic slumps, which would justify the potential attractiveness of public-sector employment in times of economic crises.The literature has also pointed out that public-sector wages are related to electoral cycles [Borjas, 1984].Moreover, the public sector is usually more unionized than the private, which strongly affects the wage-setting mechanism, giving more power to the labour-supply side in wage negotiations3 .On the other hand, mostly in local labour markets, the public sector has monopsony power as it remains the only source of demand for workers with higher education [Mueller, 1998], which allows it to dictate wage levels.Lower wages can be somehow compensated by other employment benefits, such as greater job security or more flexible hours.
Lastly, there are significant differences in institutions that may explain the observed empirical regularities in the wage distributions.(For a review of different wage-setting arrangements in a cross-country approach, see Silvestre & Eyraud [1995]; Elliott et al. [1999]) Although many countries have implemented reforms to introduce more market-oriented mechanisms to the public sector, differences in the recruitment process, wage setting and collective bargaining coverage prevail.The public sector is characterized by a higher degree of job security, but also by strict rules of promotion and remuneration, 2 In Poland, almost two-thirds of public-sector employees belong to one of three segments: public administration and national defence, social security, or health and education.The transformation of the Polish economy resulted in partial privatization of the health and education segment; however, the public sector retains its dominant position in these areas of the economy.which are related mainly to job tenure, not efficiency [Makepeace & Marcenaro-Gutierrez, 2006; Burgess & Metcalfe, 1999] 4 .
The objective of this paper is to study the evolution of the public-private wage gap in Poland using a fairly comprehensive collection of data sets for the 1999-2012 period.This covers 13 years of the second phase of economic transition from a centrally planned to a market economy.
An adequate estimation of the adjusted wage premium is crucial for reliable answers to the above questions.There are many statistical procedures that are used for wage-gap analyses which allow us to isolate the effects of so-called observables on differences in earnings.However, there is probably no single method that would allow us to address all the methodological issues that arise while studying intersectoral wage gaps.Three problems are commonly discussed in the literature.First, there is non-random selection, in both overall employment and public-sector employment.Not taking this problem into account may result in biased estimates of the wage premium.Second, if public-sector employees are better educated on average (have longer job tenure, work in bigger entities, etc.), standard parametric approaches to estimating the adjusted sector wage gap may yield misleading results, due to the so-called common support problem.Namely, there may be no-or not enough-comparable private-sector employees to make the parametric estimates relevant.To address this problem, a non-parametric approach using the matching procedure may be used.Third, adjusted wage premiums may differ in different parts of the wage distribution.Conclusions drawn from simple ordinary least squares (OLS) estimates refer only to the average wage level and may therefore be highly misleading.Quantile regression is a methodological approach that may help to address this problem.
In this paper, we use the quantile regression method and the Ñopo [2008] decomposition method to estimate the public-sector wage premium.We show that, after controlling for structural differences in employment, there is no clear evidence concerning the sign and trend of the adjusted public-sector wage premium in Poland.The parametric approach indicates a positive and growing premium, with significant variation across different parts of the wage distribution.The non-parametric approach, however, yields different results, indicating a negative premium with no clear trend in the 1999-2012 period.To the best of our knowledge, this is the first study on the public-private wage gap over this period and also the first paper to apply the non-parametric decomposition methodology to the problem of public-sector wage premiums in Poland.
The paper is structured as follows.First, we discuss the relevant literature, focusing on previous research on the Polish labour market and the methodological aspects of estimating the adjusted sector wage premium.Second, we describe the properties of the dataset survey on the structure of earnings by occupation (SEO) and discuss differences in employment structures between the two sectors of the Polish economy.Finally, we report the estimates of an adjusted wage gap.At the preliminary stage, the simple Mincerian wage equation is used [Mincer, 1974], then quantile regression estimates are discussed.Since the results of the parametric approach are not fully reliable in the context of significant differences between the employment structures of the two sectors, we refer to the non-parametric analysis using the methodology first proposed by Ñopo [2008] for gender wage-gap analysis.The paper ends with conclusions and a discussion of ideas for further studies.
For the majority of developed countries, evidence for positive public-sector wage premiums has been documented, although its magnitude varies between the studies.Generally, low wage earners benefit mostly from employment in the public sector, while individuals with relatively high salaries earn more in the private sector.Most of the findings indicate that employment in the public sector is generally beneficial for women and people with lower levels of education.On the other hand, research for developing countries generally reveals a negative public wage premium, which seems to vanish over the years when countries reach economic maturity [Lausev, 2014].
It seems that the initial period of transformation adversely affected the public-sector wage premium (with the wage gap estimated, on average, to be about 20% in favour of the private sector).With progress in the transition process, the absolute value of the wage gap decreased over time to zero and became positive in some countries.
One of the most oft-cited papers in the literature is one by Adamchik and Bedi [2000].It uses individual data from the Polish Labour Force Survey of February 1996.The authors applied the endogenous switching regression model in order to control for a selection of public-sector employment.As an instrument, they used age and whether an individual entered the labour market post or prior to 1989.The results revealed a significant public pay gap (-7% for men and -10% for women).The gap was particularly large for tertiary education graduates (-22% for men and -21% for women for individuals with five years of work experience).The authors concluded that due to the low-wage attractiveness of public-sector jobs, the public sector would face serious difficulties in attracting and retaining qualified employees.In addition, low wages could encourage moonlighting, which might deteriorate the efficiency of the public sector.
Several studies of the sectoral wage gap in the Polish labour market were undertaken by Socha.In his study conducted with Newell [1998], based on LFS data from 1992 and the single equation estimation, a positive private-sector wage premium was revealed (5.1% for men and 8.6% for women).However, the data from the year 1996 did not confirm this result.In the study by Socha and Weisberg [2002], LFS data for November 1995 was used to show that, after controlling for employee characteristics, wages in the private sector are approximately 9.8% higher than in the public sector.Moreover, variables measuring human capital were found to be stronger determinants of wages in the private sector than in the public.A similar type of analysis with LFS data was conducted by Lehmann and Wadsworth [2000], Adamchik et al. [2003] and Newell and Socha [2007].The fact that the private sector in Poland offers higher rewards for higher qualifications than the public sector was also confirmed by Rutkowski [1996Rutkowski [ , 1997]], who showed that each additional year of schooling is associated with a rate of return 0.8-0.9percentage points higher in the private sector than in the public.
A different approach was implemented by Keane and Prasad [2006], who used a household budget survey (HBS) to estimate public-private wage premiums.Although an HBS includes information on wages (based on declarations) and a vast range of personal characteristics, it does not include data on working hours.It is then impossible to calculate an hourly wage rate.However, the results of this study were consistent with those based on LFS data.The public-sector wage premium was found to be negative at the early stage of transformation (at -13% in 1991 and -9% in 1992).
In recent years, research on public-private wage differentials in Poland has been scarce.Grotkowska and Wincenciak [2014] use LFS data for Poland for 2010 to show that, despite 15 years of economic transition, public-sector wage premiums in Poland are still negative.They use a methodology similar to that in most earlier studies (Mincerian wage regression); however, they address the problem of selection in employment (adding a Heckman correction) and the potential variation of the premium across different parts of the wage distribution (by using quantile regression).The public-sector wage penalty was found to be particularly strong for women, young people, and those with higher levels of education.The size of the penalty is clearly differentiated along the wage distribution: for a large part of the distribution it is not significantly different from zero.
A critical analysis of the literature on the public-sector wage premium in Poland thus leads us to the following conclusions: • There is relatively strong evidence that in the first few years of economic transition the public-sector wage premium was negative, particularly for persons with a tertiary education and women.• There is not much evidence for the development of the public-sector wage premium in the second and third decades of economic transition; some preliminary results indicate that the premium is still negative, although it varies in different parts of the wage distribution.• The majority of the research in this area used simple parametric techniques (based on the Mincerian wage equation), estimated with OLS, which approximates the conditional mean of the response variable, given certain values of the predictor variables.In only a few studies was the selection process (in employment and the public sector) taken into account and alternative measures of central tendency and statistical dispersion were used to obtain a more comprehensive analysis of the wage differences between the two sectors (quantile regression).• The studies for Poland used LFS or HBS data that-apart from wage information-provide comprehensive data on an employee's characteristics; however, wage data from both sources is based on respondents' own declarations.Furthermore, the scale of missing information on wages is significant (e.g.27.6% in 2013 for LFS) and possibly a non-random distribution of missing data can be expected; SEO seems to be an interesting alternative to supply-side labour market surveys.• With severe data limitations, modelling the selection to the public sector is quite challenging (if possible at all), and therefore alternative non-parametric methods that would ensure the comparability of populations in both sectors seem an interesting methodological option.
In this paper we want to address some of the above issues; first, by using wage data reported by firms (SEO); and second, by confronting a parametric approach with non-parametric methods that allow us to deal with the common support problem, which seems to be a severe methodological obstacle in the analysis of public-private sector wage differentials.

Data and empirical analysis Data and empirical strategy
Firm survey data is an interesting alternative, although it has been used incidentally to asses sector wage differentials.As for transition economies, several papers used the firm-level harmonized Hungarian Wage Survey [e.g.Telegdy, 2006;Lausev, 2012].In Poland-to the best of our knowledge-this paper is the first to use firm survey data to investigate the nature of the public-private wage gap.
The SEO data set is based on a large-scale firm survey carried out by the CSO every two years.It is a sample survey and it covers businesses with more than nine employees (full-and part-time).Details of the survey methodology can be found in CSO [2015].
The SEO data provides information about gross wages (basic remuneration, payment for overtime work, duty or work seniority duty allowances, earnings for periods longer than one month recalculated per month, bonuses).One of its advantages is that it is a large data set comprising several hundred thousand observations.In 2012 (the last year for which the individual data is available) the sample included data for 725,200 employees.The biggest advantage of this data set is that it provides accurate information on gross salaries and its ingredients.It is more accurate than declarative information available in either the LFS or HBS, where the problem of approximation often occurs.However, the information on the personal characteristics of employees is less comprehensive than in the LFS (e.g.no information on family background or labour market history).Information on a worker's human capital is also limited; it includes the level of education, job tenure and occupation.There is also some basic information on firm characteristics (size by the number of employees, type of industry).One should also bear in mind that wage data in SEO, since it is reported by the employer, may be somehow biased and not fully reliable.Since the shadow economy is limited only to the private sector, we could suppose that data on wages (and on working time) might be underestimated.
Since SEO includes information only on employees, it is not possible to control for selection in employment.There is also a considerable problem in finding an appropriate instrument to control for selection in public-sector employment.A selection equation in the switching equation model requires at least one variable that affects the probability of employment in the public (or private) sector but has no direct impact on the wage level.Moreover, this needs to be a variable not related to a worker's current employment status5 but to the moment of undertaking their current job.Since we failed to find an appropriate instrument, we decided not to control for selection in the parametric approach.An attempt to model selection in employment in the public sector can be found in the paper by Grotkowska et al. [2016], where LFS data was used and information on the learned profession was used as an instrument.
Our empirical strategy included two stages.In the first we applied a parametric approach, focusing on the problem of the variation of the adjusted wage premium across the wage distribution.In the preliminary stage, we used a simple OLS single equation model to choose explanatory variables.Then, a quantile regression model was used in order to estimate the public-sector premium parameters across the wage distribution.A comparison of the results of estimations on the SEO datasets for the years 1999-2012 allowed us to analyse the evolution of the public-sector wage premium in Poland, not only for the average level of wages, but also for different moments.
Due to significant differences in employment structures and a lack of an adequate instrument to control for selection, in the second stage we decided to use a non-parametric method of wage distribution analysis proposed by Ñopo [2008].

Comparing the incomparable? Employment structures in the public and private sectors
The average wage in the public sector is heavily affected by the specific structure of employment in this part of the economy.Therefore, before estimating the size of the public-sector wage premium in Poland, we wanted to determine the extent to which both populations of workers are similar and comparable.
Table 1 reports key structural characteristics of workers in the public and private sectors in Poland.For the sake of brevity, we report data for 2012 in the table.One of the most characteristic features of the public sector is its feminization.According to the data in our sample, almost 65% of employees in the public sector are women.In the private sector this share is slightly above 40%.Due to changes in the structure of economic activity in the public sector (the decline of industrial production and the rise of non-market services, which are highly feminized), the structure of employment by sex has changed significantly over the last several decades with a significant increase in the share of women.On average, the public sector employs older workers than the private sector: the much higher share of employment in the public sector (compared to the private sector) is attributed to people of "immobile age" and the age of formal retirement, although the share of the latter group in employment is negligible.The average age of a worker in the public sector is 43.8 years, versus 38.9 years in the private sector.
Perhaps the most important aspect of the difference in the structures of employment in the public and private sectors is related to qualifications, and in particular to the level of education and occupations.In our sample (SEO 2012 data), more than half of the public-sector workers have a higher education and almost 30% have a vocational secondary education or post-secondary education.In the private sector, the largest group are workers with a vocational education (or lower).Simultaneously, the public sector remains the main source of demand for higher qualifications (although its position is less monopsonic than at the beginning of the transition period).
Structural change in the public sector and an increased average level of education of its employees led to a significant change in the occupational structure of the worker population.In recent years, the share of craft and related trade workers and technicians and associate professionals has fallen, while the share of professionals has expanded.Professionals are now the largest group of public-sector workers (44%).On the other hand, the private sector has a more dispersed structure in terms of occupations, with craft and related trade workers, professionals, machine operators and salesmen being the most numerous.
Currently (SEO 2012 data), more than 70% of employees in the public sector work in one of three segments (public administration and national defence, social security, or health and education).Employment in the private sector is concentrated almost exclusively in the manufacturing industry, construction and market services (66%).Surprisingly, there is not much difference between workers in the public and private sectors in terms of the size of their employer.However, our sample is restricted to entities with at least 10 employees.

Public-sector wage premium: unadjusted wage gap and preliminary OLS estimates
Table 2 provides statistics on the distribution of hourly wages in the population of workers, in both the public and private sectors in our sample (SEO 2012 data).The average hourly wage is generally higher in the public sector, with the average wage ratio equal to 1.21.The difference is even more substantial if we compare median wages (1.34).The difference is higher for women and for older workers.It is also more substantial for individuals with a vocational education, for craft and related trade workers, and plant or machine operators and assemblers.The gross public-sector wage premium is also relatively high for individuals employed in the smallest entities in the sample (10-50 workers).There are only a few labour force groups with a negative gross public-sector wage premium.These are individuals with a tertiary education, managers, technicians and persons performing elementary occupations.The distribution of wages in the public sector is less dispersed than in the private sector, with a considerably smaller relative standard deviation for women and individuals performing occupations typical of lower levels of education (see Figure 1).
Moving to the econometric analysis, we started with a simple Mincerian wage equation estimated with the OLS method.Although serious methodological reservations can be raised with regard to simple OLS estimation, we find it desirable to start with this method as it will make it possible to select relevant explanatory variables for two other methods of our study.The results of the OLS regression can also be compared to the results of other studies in the literature.The basic equation that we use is stated in the following form: ln w j = X j β + u j , u ∼ N(0, σ ).
Our explained variable ln w j is a log of an hourly wage rate, while explanatory variables (X j ) include the basic demographic variables of sex and age (specification 1), supplemented by the human capital variables of the level of education and tenure with the current employer (specification 2), dummies for occupational groups (2-digit ISCO classification, specification 3), employer's characteristics (variable for the size of a firm measured with the number of employees and dummy variables for industries -specification 4), and contract characteristics (dummy variable for the type of contract: permanent versus temporary and dummy variable for non-standard working time organisation -specification 5).Specification 5 is the most comprehensive (with the highest level of explanatory power); however, we will treat specification 4 as the benchmark, since the conditioning variables used in this specification were also included in earlier editions of the SEO databases.It will be possible then to compare the results for 2012 with earlier periods when using other econometric techniques.
Table 3 shows the estimates for all five specifications.For the sake of brevity, we omitted the parameters for occupational and industry dummies.The results are consistent with the expectations.As for control variables, all of them were found to be significant and sensible predictors of wages (negative sign for women, positive returns to education and firm size, concave relationship to age and tenure).Each specification revealed the positive public-sector wage premium.It is worth noting that adding simple demographic variables makes an adjusted premium even higher than observed in the raw data, while controlling for human capital variables makes it significantly lower.It suggests that the public sector attracts relatively more workers with characteristics of human capital being better rewarded in the labour market.Including the employer and contract characteristics resulted in an adjusted wage gap of 5-7%.These results suggest that public-sector employment is concentrated in firms of the size and industry offering, ceteris paribus, lower wages.
A simple linear regression model is far too simple to fully capture the nature of the public sector wage premium.Controlling for all relevant worker and job characteristics may not be enough to address the issue of the common support problem, therefore we propose the Ñopo decomposition technique in the last step of our empirical analysis.Additionally, the linear regression model is based on averages, which masks the different effects of the premium in different parts of the wage distribution.To address this issue we move to the quantile regression approach.

Public-sector wage premium: results of quantile regression
The econometric approach reported in the previous subsection relied on averages.The quantile regression approach allows us to extend the analysis for the whole distribution of wages.It is a well-documented fact for various countries and periods that the public-sector wage premium differs for different parts of the wage distribution, usually being positive for individuals with low wages and negative for the right tail of the wage distribution.
Again, let ln w j be the log of hourly wage, while X j the vector of regressors capturing individual characteristics of employees as well as the job attributes.The quantile regression model estimates the θ -th quantile of the distribution of ln w j conditional on X j as a linear function of the regressors: The set of regressors in X j include the variables used in specification 4 for the OLS model.A full set of estimates for the model are available from the authors upon request, and for the sake of brevity are not included in this paper as in reporting the results we focus on the ownership dummy only (Figure 2).Starting with analysis for the year 2012, the last for which data is available, there is a clear relationship between the wage decile and an adjusted public-sector wage premium.It is the highest for individuals with the lowest wages (first decile), where it is equal to 13.3% and it monotonically falls with growing income.For the fifth decile (median) it is positive, but it is equal only to 5.4%.For wages in the upper half of the distribution, it falls even further to become negative for top earners (ninth decile).However, the penalty for working in the public sector does not seem to be severe, since it is equal to only 1.3%.
When we compare the adjusted wage premium for different deciles in 2012 with previous years, we notice a significant upward shift of its curve (Figure 2).Top earners working in the public sector are definitely the group that benefited the most from the wage structure changes in recent years (as compared to their counterparts from the private sector).Between 1999 and 2012 the share of earners that exhibited a negative public-sector wage premium diminished gradually; yet in 1999 the public-sector wage premium was negative for about 40% of workers, while in 2012 the respective share was only 17%.
The quantile regression method is still plagued by the problem of common support, which cannot be totally eliminated by including explanatory variables as controls.Another issue is the fact that all untreated (private-sector employees in our study) individuals receive the same weight in parametric methods when estimating the value of the public-sector wage premium.The Ñopo [2008] decomposition comes as a solution to both of these issues.

Non-parametric approach: Ñopo decomposition
All parametric estimations of wage differentials rely heavily on the overlap assumption, which is rarely verified empirically.There may be combinations of characteristics for which it is possible to find workers in the public sector but not in the private sector, and vice versa 6 .In such cases, parametric estimations may become useless.Moreover, the decision as to whether the overlap is sufficient in the multivariable context is discretionary.A second issue is the selection problem resulting in the endogeneity of how individual and employer characteristics affect each other and how they affect wages.In such a case the parametric estimates may be biased.
Ñopo [2008] proposed a solution to avoid the problem of common support by constructing an implicit decomposition based on matching.It is based on a technique allowing the researcher to find matched samples with "similar" observable features, except for one particular characteristic, the "treatment", which is used to divide observations into two sets, the treated and the control group.After controlling for differences in the observed characteristics, it is possible to measure the impact of treatment alone.Ñopo [2008] considered 6 An extreme example of this situation is the lack of public administration workers employed in the private sector.
the gender variable as a treatment and used matching to select sub-samples of males and females in such a way that there were no differences in observable characteristics between "matched" males and "matched" females.Afterwards, the wages of each woman were compared with the average wages of men with exactly the same characteristics.The same procedure may be applied to sectoral wage differentials analysis.
The advantage of the decomposition approach is to split the difference in the earnings between public and private sector workers into a part attributable to differences in endowments and a part attributable only to different pricing of these characteristics across the two sectors.In gender wage-gap studies, the differences in pricing are usually considered to be "unjustified", or interpreted as discrimination (or its upper boundary).In the case of public-sector wage premiums, the interpretation could be different.In fact, it is either indirect proof of significant obstacles to intersectoral labour force mobility or proof of non-wage benefits associated with employment in the sector characterized by a negative wage premium.
In the case of parametric methods, it is assumed that estimates of the wage equations are also valid from the support of the observable characteristics for which they are estimated.It may result in an overstatement of that part of the wage differential unexplained by differences in endowments.Ñopo in his decomposition proposes to break down the wage differential into four terms, two of which are analogous to the elements of the traditional decomposition technique proposed by Oaxaca-Blinder [Oaxaca, 1973], while the other two account for differences in the supports.
Adopting the Ñopo methodology in the case of public-sector wage premiums, the public-private sector wage differential can be expressed in terms of four additive elements: where ΔEXP is the part of the average public-private sector wage gap (Δ) that can be explained by differences in the distributions of the characteristics of workers in the public and private sectors over the common support; ΔPUB is the part of the wage gap that is explained by the existence of public-sector employees that cannot be found in the common support of the distributions of characteristics; and ΔPRIV is the part of the wage gap that can be explained by the presence of private-sector workers out of the common support.Finally, ΔUNEXP is the unexplained part of the wage gap (i.e. that part of the difference that is attributed to unobservable characteristics or to different returns between the two groups in the observable characteristics).The sum of the first three elements may be understood as the component "explained" by differences in the characteristics, while the fourth term corresponds to the "unexplained" component.
The Ñopo [2008] procedure involves applying matching methods to simultaneously identifying the common support (i.e.groups of workers in the public and private sectors with similar observable characteristics) and decompose the wage differential into four components without imposing any restrictions on the way in which the explanatory variables affect the dependent variable.The main drawback of the method is the problem of high dimensionality.The inclusion of a large number of explanatory variables for matching may substantially reduce the number of observations found in the common support.
The most important advantage of Ñopo's methodology of decomposition is that it accounts for differences in the supports of the distribution.If the unmatched public-sector workers have average wages above the average wages of their matched peers, then estimating earnings equations for public-sector workers without accounting for this regularity tends to overestimate the unexplained component (ΔUNEXP).On the other hand, if there are private-sector workers who to a greater extent than public-sector workers exhibit characteristics that the labour market rewards (as was probably the case at the beginning of the transition period), the unexplained component could actually be underestimated.
The raw observed wage gaps for the years 1999-2012 are in the range of 18.2% to 28.8%.Controlling for matching variables significantly reduces the value of the gap as reported in Table 4. Inclusion of sex and age is generally not enough to explain the difference as the unexplained gap remains close to the raw value.Extending the specification for human capital variables (level of education and job experience) allows us to explain much of the raw gap as the unexplained part generally turns even negative.It allows us to conclude that public-sector employees are endowed with human-capital characteristics which would justify even higher wage differentials than observed in the raw data.The inclusion of occupation dummies (at the 2-digit level) generally does not produce a significant impact on the unexplained wage differential.However, adding variables related to job characteristics (NACE section and firm size) brings about a large decrease in the value of the unexplained wage gap.The common support in this case, however, is severely limited, which suggests that public and private employment is far more differentiated over these characteristics than over any other.
Contrary to the results of the parametric approach (in which the problem of common support is evidently present but neglected), the average adjusted public-sector wage premium has remained negative in recent years.Moreover, there is no clear trend when we look at the results for the 1999-2012 period.

Conclusions and open issues
The objective of this paper was to investigate the evolution of the public-private wage gap in Poland using a collection of data sets covering the span of 1999-2012.We applied the quantile regression method and the Ñopo [2008] decomposition method to show that, after controlling for structural differences in employment, there is no clear evidence concerning the signs and trends of the adjusted public-sector wage premium in Poland.We used samples of individual wages reported by firms as a representative data set on wages in Poland, although only companies employing more than nine persons are surveyed.Small companies are concentrated in the private sector and the positive wage returns to firm size are well documented in the literature.This probably leads to the underestimation of the raw wage gap in our study.
The parametric approach indicates a positive and increasing premium, with significant variation across different parts of the wage distribution.Quantile regression results indicate that the public-sector wage premium for the median earner ranged from 0.6% in 2001 to 5.0% in 2012.The non-parametric approach yields different results, indicating a negative wage premium with no clear trend in the 1999-2012 period.However, we might see a declining public wage penalty in the years 2006-2012.The apparent inconsistency of the results may be explained by the fact that in the case of the non-parametric approach non-random selectivity to the ownership sector is not an issue, while in the regression analysis it could not be properly addressed due to the lack of a valid instrument.If workers select themselves to the ownership sectors in a non-random way, part of the effect attributed to the public-sector wage premium is in fact a result of specific characteristics of employees.To the best of our knowledge, this is the first study on the public-private wage gap over this period, and also the first paper to apply the non-parametric decomposition methodology to the problem of the public-sector wage premium in Poland.
The results obtained here are clearly different from the findings of studies for developed economies, although quite typical of countries in transition.The public-sector wage penalty is particularly strong for top earners.Further analysis should, however, include quantile selection models to check whether the non-random selection process affects quantile effects for wage premiums.This would require a different data set, covering more detailed labour force characteristics, and enabling the identification of reliable instruments to control for selectivity.
Our results raise several new questions.One of them is related to the nonwage benefits of employment, particularly in the public sector.One might expect some compensating benefits of employment in the public sector for those particularly affected by strong wage penalties.These can come in the form of greater job security, employment stability, prestige, career prospects and personal development, in addition to flexible working hours and many other benefits.Another open issue is the question of barriers to intersectoral labour mobility and its limitations.

Figure 1. Wage distributions in public and private sectors
Explanatory note: Young = individuals aged 30 years old or less; High skilled = individuals with tertiary education; Low skilled = individuals with education levels lower than secondary Source: Author's computation based on SEO data, 2012.