• Aucun résultat trouvé

209CHAPTER V WAGE COMPRESSION IN EUROPE: FIRST EVIDENCE FROM THE STRUCTURE OF EARNINGS SURVEY 2002 Gilles Mourre

N/A
N/A
Protected

Academic year: 2021

Partager "209CHAPTER V WAGE COMPRESSION IN EUROPE: FIRST EVIDENCE FROM THE STRUCTURE OF EARNINGS SURVEY 2002 Gilles Mourre"

Copied!
31
0
0

Texte intégral

(1)

CHAPTER V

WAGE COMPRESSION IN EUROPE: FIRST EVIDENCE FROM THE STRUCTURE OF EARNINGS SURVEY 2002

Gilles Mourre

(2)
(3)

1. I NTRODUCTION

Wage moderation in the euro area appears to have been a crucial element of macroeconomic stability in recent years and may partly explain the resilience of employment in the face of the economic slowdown. However, there is still a long way to go towards achieving Lisbon employment targets in Europe. Consideration should also then be given to the microeconomic structure of wages, which may also affect the macroeconomic performances. Indeed, there has been growing interest amongst researchers and policy makers in the labour market institutions, which are considered as a key factor influencing employment performance and wage determination in Europe (e.g. Nunziata, 2005). Wage distribution could be one of the channels through which some institutions, particularly those shaping the wage setting, impact labour market performances (Bertola and Rogerson, 1997).

Moreover, a problem often mentioned about the EU labour markets is the relatively low employment rate seen in specific groups, such as youth, women, older workers and the low skilled, while the prime-age employment rate is much higher and broadly similar to that of the US (Dolado et al., 2001). This is often attributed to the compressed wage structure in Europe.

Statistically, the dispersion of wage distribution is often claimed to be much lower in Europe than in the United States (Bertola, Blau and Kahn 2002a). In economic terms, “wage compression” means the difference in wages across workers or firms in Europe that does not reflect the (wider) difference in productivity. This mismatch can be understood in a static way comparing the level of relative wage and relative productivity, which is the focus of this chapter. It could also be defined in a dynamic way as the ability of relative wages to swiftly respond to shocks affecting relative productivity.

The release of the results of the Structure of Earnings Survey for 2002 (SES2002) by Eurostat in April 2005 sheds light on the wage structure in Europe. As well as yielding a snapshot of the overall wage structure with inter-percentile gaps, the survey provides useful information on average hourly and monthly wages across a number of relevant dimensions:

educational attainment, sector, type of occupation, firm size, gender or age group. Although

the data publically disseminated by Eurostat remain fairly aggregated and expressed as

average hourly wages, this information can be adjusted for the composition of the workforce

in terms of skills, occupations and sectors, which differs from country to country. Freeman

and Schettkat (2001a) emphasised the importance of such compositional effects, when

showing that the compression of wages in Germany compared with those in the US partly

(4)

comes from the compression of skill distribution in Germany. Moreover, the information on hourly wages is more relevant than that on monthly wages to study wage dispersion in Europe, since it is not distorted by the number of hours worked and in particular the impact of overtime and part-time, which vary a lot across European countries.

Based on the recent results of the SES 2002, the paper identifies the existence of wage compression in Europe using a labour demand setting. The methodology has two characteristics: the estimates of wage compression are model-based; the focus of the study is the EU27 (including Bulgaria and Romania).

The rest of the paper is structured as follows. Section 2 briefly surveys the literature on the causes and effects of wage compression. Section 3 presents the dataset used. Section 4 sets out the theoretical framework. Section 5 presents the econometric methodology used to compute the coefficent of wage compression. Section 6 presents and discusses the econometric results. This section looks at the existence of wage compression both across occupations and across educational attainments. Section 7 concludes.

2. O RIGINS AND EFFECTS OF WAGE COMPRESSION : A LITERATURE SURVEY

The economic literature has been very fertile in explaining the wage compression and in understanding the relationship between wage dispersion and employment. Economic theory offers two non-mutually-exclusive types of explanation for the existence of wage compression.

• The first one explains wage compression as being caused by exogenous labour market

institutions such as minimum wages (which affect the lower end of wage

distributions), trade-unions, central bargaining framework, governmental extension of

collective agreements or any institution which contributes to raise the reservation

wage, such as generous unemployment benefits. Koeniger et al. (2005) investigate the

importance of labour market institutions such as unemployment insurance, unions,

firing regulation and minimum wages for the evolution of wage inequality across

countries. Their estimates for 11 OECD countries suggest that labour market

institutions can account for a large part of the change in wage inequality across

countries after controlling for time and country effects.

(5)

• The second type of explanation identifies endogenous causes for wage compression.

For instance, Booth and Zoega (2002) provide micro-foundations for wage compression by modelling wage-setting in an imperfectly competitive labour market with heterogeneous workers and firms 1 . In their model, wage compression arises quite naturally in market economies and does not depend on the existence of ad-hoc institutional structures such as minimum wages and unions. Using a Nash Bargaining framework with on-the-job search, Shimer (2004) concludes that there can be wage dispersion in equilibrium even if all workers and firms are homogeneous. If firms are heterogeneous, more productive firms pay higher wages and workers switch employers whenever they encounter a more productive job. Ample empirical evidence gives support to the endogenous determination of wages, suggesting that employers pay different wages to similar workers, which potentially allows wages to deviate from productivity in some cases. Krueger and Summers (1988) show that, after controlling for personal characteristics, some US industries pay wages up to 20%

above and below the average wage. Similarly, taking a descriptive and macro approach at the euro area level, Genre, Momferatou and Mourre (2005) show that workers’ characteristics fail to fully explain wage differentials across industries, which are partly linked to firms’ characteristics, such as corporate size and capital intensity.

Two explanations are put forward in the literature: either employers pursue different wage policies or high-wage firms attract more able workers. Empirical studies by Abowd et al. (1999) and Abowd and Kramarz (2000a, 2000b), based on the analysis of matched employer-worker data for both the U.S. and France, conclude that the two are equally important as explanations of inter-industry differentials and that wage policy differences account for 70% of the firm-size differentials.

• Within the second type of explanation identifying endogenous causes for wage compression, a large vein of theoretical and empirical research considers and discusses the impact of the dispersion of wages within firms on firm productivity. Such relationship generates endogenous incentives for firms to modify - compress or stretch according to the sign of the relationship - their intra-firms wage distribution with a

1 Some firms hire higher ability workers who collectively perform better and this collective ability determines the

intra-firm level of task complexity. Because of the finite number of higher ability workers, firms able to do the

most complex tasks constitute a “narrow market” where they have a monopsony power. As a result, wages are

compressed within firms, so that low-ability workers are paid more, relative to their talent, than high-ability

workers.

(6)

view to optimising performance. Two types of models offer different interpretation and suggest the existence of opposite relationship: the ‘fairness, morale and cohesiveness’ models are opposed to the “tournament” models. On the one hand, Akerlof and Yellen (1990) argue that, in a firm where the workers’ characteristics cannot be fully observed and where the monitoring of their actions is not perfect, employers have to find well-designed incentives to maximise the workers’ effort.

According to the “fair wage-effort” hypothesis proposed by these authors, workers often compare their wages either with workers within the same firm or with workers in other firms. Workers decrease their effort if their actual wage is lower than the wage they regard as fair. A wage is generally considered as fair if the wage differential is lower than the performance differential. The ‘fair wage-effort’ hypothesis is based on the social exchange theory in sociology and on the equity theory in psychology.

Therefore, the effort function of a worker does not only depend positively on the wage

level but also on the degree of wage dispersion within the firm. Firms have a strong

incentive to compress their wage distribution, which improves labour relations and

stimulates the average workers’ productivity. In full contrast to the previous literature,

the ‘tournament’ model, developed by Lazear and Rosen (1981), points to the benefits

of a more dispersed wage structure, derived from a performance-based pay system. A

simple version of the tournament model suggests that managers should introduce a

large spread in the rewards of workers – in the form of a promotion or a bonus - in

order to stimulate their effort and should award the largest prize to the most productive

workers. Lazear (1989) recognises, however, that high within-firm wage dispersion

and the resulting rise in competition between workers may adversely impact the

performance of firms, when several workers are non-cooperative or adopt a sabotage

behaviour (“hawks”). The non-cooperative activities of some may then reduce the

total effort level of workers and counterbalance the virtue of an output-based pay

system on firm performance. This problem calls for some wage compression or,

alternatively, carefully screening workers before recruitment and adapting the

compensation scheme to the hierarchical level, where “hawks” are supposed to be

concentrated. Reflecting the ambiguity of the theoretical predictions, the numerous

empirical studies that investigate the relationship between wage spread and firm

productivity provide significantly different results across European countries. For

instance, for Belgium, Lallemand, Plasman and Rycx (2007) show the existence of a

positive but concave relationship and back the “tournament” theory while for Portugal,

(7)

Martins (2008) reports a negative relationship in support of the “fairness” theory.

Following Abowd et al. (1999), Martins (2008) argues that, owing to data limitations, most studies (e.g. Eriksson 1999 for Denmark, Winter-Ebmer and Zweimuller 1999 for Austria, Hibbs and Locking 2000 for Sweden) do not control for unobservable differences (e.g. ability, school quality), which can play a large role in explaining wage dispersion.

Besides the origin of wage compression, the second critical issue covered by the economic literature is the impact of wage compression on economic performance and, in particular, employment.

• A first vein of research looks at the impact of minimum wages on employment.

Although many economists traditionally argue that the effect of a binding minimum wage law is to reduce firms’ demand for low skill workers, Rebitzer and Taylor (1995) state that this prediction of worker displacement depends critically on the assumption that the productivity of employees is not dependent upon the wage. They find that in an efficiency wage model, a minimum wage may increase the level of employment in low wage jobs. More generally, such a result reflects either the case of an efficiency wage model with a large number of employers (Manning, 1995, Rebitzer and Taylor 1995) or that of labour demand under monopsony (Burdett and Mortensen 1998).

While recent empirical studies fail to reach a consensus on the issue of the employment effects of minimum wages (Neumark and Wascher 2000 and Card and Krueger 2000), Strobl and Walsh (2002) argue that, in a more general and realistic employment contract where not only a wage but also a set of working conditions are specified, an employment subsidy is a more effective way of improving welfare than minimum wages.

• A second strand of research focuses on the general impact of wage compression or

wage inequality on employment. For instance, Bertola, Blau and Kahn (2002a) show

that, controlling for country- and time-specific effects, high employment is associated

with high levels of wage inequality. They suggest that US relative unemployment of

most disadvantaged groups fell in recent years in part because the more flexible labour

market institutions prevailing in the US allow economic shocks to affect real and

relative wages to a greater degree than in other OECD countries. In another paper,

Bertola, Blau and Kahn (2002b) find that greater wage compression caused by a high

(8)

degree of involvement of unions in wage-setting would lead to relatively lower employment rates for young and older individuals than for the prime-aged (particularly for prime-age men), given the labour demand elasticities for these different groups 2 . They also find evidence that the wage compression induced by higher unionisation raises the unemployment rate of young men and prime-age women compared to prime-age men. Lindquist (2005) shows that, when labour markets are competitive, even low degrees of wage compression lead to large welfare losses, because wage compression brings about costly unemployment among low-skilled workers 3 . Overall, these analyses are mainly based on the theory that employment is mainly determined by firms’ labour demand.

However and as already mentioned in the case of minimum wages, the relevance of the usual labour-demand side story on the negative effect of wage compression should somewhat be qualified. In other models of employment determination at the micro level, such as efficient bargaining or employer monopsony models, wage increases may possibly induce employment growth (Card and Krueger 2000). For instance, compression in wage distribution may have a positive impact on firm-sponsored general training. Acemoglu and Pischke (1999) find that when labour market frictions and institutions compress the structure of wages, firms are encouraged to invest in the general skills of their employees, as the distortion in the wage structure turn “technologically general” skills into “specific” skills. On an empirical ground, Freeman and Schettkat (2001b) highlight that the differing dispersion of wages is not a major contributor to differences in overall employment rates between the US and Germany, although they acknowledge that changes (and not level) in relative employment are related to changes in relative wages, raising the possibility of some substitution behaviour. The job problem in Germany may be due to a general lack in overall demand for labour rather than an insufficient relative labour demand affecting the low skilled in particular. In a recent study carrying out a descriptive analysis comparing employment structure in Europe and the US, the European Commission (2004) finds no clear graphical evidence in support of the view that differences in employment structure are mainly due to a more compressed wage structure in the EU.

2 The wage compression, seen empirically, might result from the fact that, in maximising workers’ rent, trade

unions negotiate the largest wage premium for groups with very elastic labour supply such as women (reflecting

their high opportunity cost of employment, i.e. home production and child care).

(9)

3. D ATA

The Structure of Earnings Survey (SES) for 2002 is the first of a series of four-yearly surveys to be conducted under the Council Regulation 530/1999 and the Commission Regulation 1916/2000. The objective of this legislation is to provide accurate and harmonized data on earnings distribution in (current and future) EU member states for policy-making and research purposes. Therefore, the main advantages of the 2002 SES is to give detailed and comparable information for all European Union countries on earning distribution and the relationship between the level of remuneration, individual characteristics of employees and those of their employers 4 .

The survey reports gross wages only. It does not cover all labour costs and in particular excludes employers’ social security contributions and other non-wage labour costs 5 . Therefore, this study focuses on wage compression and not on labour cost compression, which can be somewhat different given the potentially strong effect of the tax system in Europe. The reference year is the calendar year 2002. The statistical units of the survey are both local units belonging to enterprises with 10 or more employees and employees having at least one working day paid by the employer at a full rate during October 2002. October has been chosen as the representative month, because it is least affected by absences owing to annual leave or public holidays. The limitation of the sample to private sector enterprises with 10 or more workers is partly due to the existence of large black economy in some countries, e.g. some Mediterranean countries. This limitation may somehow complicate the computation of an accurate and comparable wage dispersion indicator for all EU countries. Indeed, the low pay prevails in small firms and the proportion of employees working in firms with less than 10 employees varies much across countries. While it stands at around 20% in Belgium and the Netherlands, it is certainly much higher in some countries such as Italy. This would certainly affect the statistical dispersion of wages but not necessarily our indicator of wage compression which implicitly compares wages with productivity. Therefore, the wage

3 The effect of wage compression on the supply of skilled labour, however, is fairly small, since the disincentive effect of lower wages for the high skilled is largely offset by a lower opportunity cost of schooling owing to higher unemployment.

4 A first Structure of Earnings Survey was conducted in 1995. However, it has a lower geographical coverage and was not fully comparable with the subsequent surveys, which are to be carried out under new regulations aiming at providing more accurate and harmonized data on earnings in EU Member states.

5 Gross hourly wages (referred to as gross hourly earnings in the survey) are the remuneration in cash paid to the

employee directly and regularly by the employer at the time of each pay period, before deductions of any tax and

social security contributions payable by employee and withheld by the employer.

(10)

compression indicators are likely to be over-estimated in countries, where both the proportion of small firms and the firm-size wage premium are large and under-estimated in countries where the share of small firms is important and there is a high reservation wage (owing to the existence of a minimum wage or strong bargaining power). It is however difficult to a priori predict the impact of this sample selection on the bias in the estimate of wage compression.

For legal reasons, individual data are only accessible to researchers under restricted conditions and were not available at the inception of the project 6 . Therefore, we have used the semi-aggregated data released to the public and posted on the Eurostat website (in April 2005). They are aggregated in the form of group-average along different dimensions. In this study, we use two samples of the survey, both reporting information on average gross hourly wages (excluding overtime payments) and the number of employees. The first provides data broken down by occupation, firm size, gender and country, while the second consists of data disaggregated by level of education, industry, gender and country. In order to fully exploit the cross-sectional information of SES data, we use all the country data available: the paper not only covers the EU25 countries (excluding Portugal, Greece and Malta) but also includes the two recently acceded countries (AC2), i.e. Bulgaria and Romania 7 . The data are broken down according to the National Classification of Economic Activities (NACE Rev. 1, 2001), the International Standard Classification of Occupations ISCO-88 and the International Standard Classification of Education (ISCED-97).

The reference year is 2002. The statistical units of the survey are both local units with 1 or more employees belonging to enterprises with 10 or more employees and employees with earnings during October 2002, having at least 1 working day paid by the employer at a full

6 In fact, it is possible to have access to the individual data by going to the “safe-room” in Eurostat. For instance, some researchers involved in the European Central Bank Wage Dynamics Network have followed this procedure to work on the micro-data. Another possibility to have access to the individual data is to be a member of the LEED project coordinated by François Rycx and David Marsden amongst others. Within this research project, financed by the European Commission, a remote access system has been developed for the ESES 2002 and today approximately 20 researchers from various universities in Europe and the US have the possibility to do some analysis with SES 2002 at the individual level for 10 EU countries directly from their computer in their university. The idea is that researchers send Stata jobs by e-mail to Eurostat and receive the output a few minutes later. Within the LEED project (and its predecessor called PiEP) a lot of work has been done on wage compression/dispersion/inequality in European countries and in particular on the impact of wage dispersion on economic performance essentially assessed through productivity at the industry or firm level (cf. the LEED and PiEP websites: http://cep.lse.ac.uk/leed/ or http://cep.lse.ac.uk/piep/).

7 Data for Portugal, Greece and Malta were not available when the database was compiled and the bulk of the

current study was carried out. At the date of the publication, data were not available for Malta yet. Likewise, it

should be noted that the decomposition of earnings by education level and sector for the UK was not posted in

the Eurostat website when the current study was started.

(11)

rate. For more details on this information, see the Eurostat website. The rest of the subsection describes the data used in this analysis.

Employees are all persons who have a direct employment contract with their employer and receive remuneration in cash or in kind for certain quality and quantity of work done, irrespective of the type of work performed, the number of working hours (full or part-time) and the duration of the employment contract (fixed or indefinite). Other categories of workers include: apprentices and trainees with an employment contract with the reporting unit;

seasonal or occasional workers who are working pre-defined hours on a contractual basis with the local unit or the enterprise; outworkers, but only if there is an explicit agreement that they are remunerated on the basis of the amount of hours worked; employees on maternity leave as long as they receive remuneration from the employer. Conversely, the following categories of workers are excluded: employees, apprentices or trainees without an employment contract with the enterprise/local unit; seasonal or occasional workers who are employed without pre- defined working hours; persons (i.e. unpaid owners or directors or managers) remunerated by way of fees or commission; employees of the observation unit who have been working abroad for more than one year in an affiliated company; the self-employed; family workers and voluntary workers.

Hourly gross earnings are the average gross hourly earnings in October 2002 received for normal working hours (overtime excluded), i.e. for the number of hours which the employee is obliged to work in the reference month under the terms of the employment contract, regulation or rules in force in the local unit. Gross earnings are the remunerations in cash paid to the employee directly and regularly by the employer at the time of each pay period, before deductions of any tax and social security contributions payable by employee and withheld by the employer. The following are not included: payments paid in this period but relating to other periods, such as advances, or pay for holiday or sickness absence outside reference period; periodic bonuses and gratuities not paid regularly at each pay date;

payments for periods of absence paid by the employer at a reduced rate; statutory family allowances; the value of benefits in kind; reimbursements or payments (for travelling, subsistence etc.) or expenses incurred in carrying out the employer's business.

Figure 1 shows the variation of hourly wages across education attainments. The average

hourly wage of people with tertiary education is 84% higher in the EU25 than that of people

with lower secondary education or less. Figure 2 illustrates that wage dispersion is even more

(12)

acute across occupations. The average hourly wage of legislators, senior official and top managers is 2.3 times as high as that of elementary occupations. A natural question is what the “optimal” degree of wage dispersions could be.

Figure 1. Hourly wage by educational attainment

(Mean hourly earnings in euros, corrected for gender composition)

0 2 4 6 8 10 12 14 16 18 20

To tal P re-primary, primary and lo wer seco ndary

educatio n

Upper seco ndary educatio n

P o st-seco ndary no n-tertiary

educatio n

Tertiary educatio n

eu25 eu15 NMS10

Figure 2. Hourly wage by occupation

(Mean hourly earnings in euros, corrected for gender composition)

0 5 10 15 20 25

To ta l Le gi sl at or s, se ni or of fici al s a nd P rof es si on al s T ech ni ci an s and as so ci at e pr ofes si onal s Cl er ks Se rv ic e wo rk er s a nd sh op a nd C raft an d re la te d t rad es wo rk er s P lan t an d ma ch in e ope ra to rs a nd E lem entar y oc cu pat io ns A rme d f or ces

eu25 eu15 NMS10

4. T HEORETICAL FRAMEWORK : A SIMPLE LABOUR - DEMAND MODEL

In order to identify the degree of wage compression in Europe, we need to develop a simple model, which relates wages to the relative employment by using a standard labour demand framework. The idea is to examine whether the (absolute value of) wage coefficient in the relative employment equation is higher than that in the employment equation. If so, there is evidence of wage compression.

We first consider a CES production function with two production factors and constant

returns to scale, as proposed by Arrow et al. (1961). It provides a simple and standard

(13)

analytical framework to highlight the effect of the main determinants of labour demand, while allowing for a substitution between production factors, unlike the simpler Cobb-Douglas specification. Other more complex production functions, such as Translog, could have been chosen. However, assuming a specific production function could be problematic as the empirical results may be influenced by the assumed structure. We therefore prefer to stick to a fairly standard specification. The CES specification adopted is as follows:

1 1 1

) 1 ( )

(

⎥ ⎦

⎢ ⎤

⎡ + −

= σ

σ σ σ σ

σ α

α i i i

i a E K

Y

with Y standing for output, E for employment (i.e. labour), K for capital, a for employment- enhancing technical progress 8 , α for the employment-intensity of the method of production and σ for the elasticity of substitution between effective employment (aE) and capital. The subscript i represents a specific type of workers or firms (sector, age bracket, skill level, occupation, gender, contract type, etc). Assuming that firms do not necessarily operate in a perfectly competitive environment, the first order condition of firm’s profit maximisation leads to equate the marginal labour productivity to the mark-up adjusted real compensation per employee µ.(w/p) with µ being the mark-up over costs in the case of imperfect competition 9 . This gives the following expression:

σ σ σ

α σ

µ i 1 i 1 i 1

i i i i

i a E Y

E Y p

w − −

∂ =

= ∂ (1)

After rearranging and writing in logarithms, we end up with a standard employment equation:

log Ei= log Yi - σ log(w/p)i - (1- σ) log ai + σ log(α/µ) (2) The employment level in absence of wage compression depends on total output, real wages and employment-augmenting technical progress (i.e. Harrod-neutral technical progress). The elasticity of employment to real labour costs equals minus one times the

8 a represents the labour efficiency. It can also be seen as the degree of labour-augmenting technical progress (i.e. Harrod-neutral technical progress).

9 The SES 2002 data dealt with in this study cover gross wages, which exclude non-wage labour costs (mainly

employers’ social security contributions). However, we can arguably suppose that within a given country, the

share of non-wage costs in total labour costs is not substantially different along the wage distribution. In any

(14)

elasticity of substitution between employment and capital. The latter is conventionally assumed to lie between zero and unity, which reflects the imperfect substitution between production factors. This implies that the elasticity of employment to real labour costs is negative and lower than 1 in absolute value and that the coefficient of employment- augmenting technical progress is negative as well. This assumption is only called into question when the production factors are complements instead of substitutes, which is not plausible as shown by some studies (e.g. Bentolila and Saint-Paul, 2003).

If we consider two different types of workers or firms i and 0 with 0 serving as a benchmark, the relative mark-up adjusted wage can be written from (1) as a function of relative marginal return of employment:

⎟ ⎟

⎜⎜ ⎞

= ∂

⎟ ⎟

⎜⎜ ⎞

0 0 0

0 0

/

/ E

Y E Y p

w p w

i i i

i i

µ

µ (3)

Allowing for wage compression would mean that relative wage is lower than relative productivity. So, with wage compression coefficient c lower than 1, expression (3) can be rewritten in logarithm as:

⎟ ⎟

⎜ ⎜

⎟⎟ ⎠

⎜⎜ ⎞

⎟⎟ ⎛

⎜⎜ ⎞

⎟⎟ ⎛

⎜⎜ ⎞

− ⎛

⎟ =

⎜⎜ ⎞

− ∂

⎟ =

⎜⎜ ⎞

⎛ ⋅

− −

σ σ σ

σ

α α µ

µ 1

0 1

0 1

0 0 0

0 0

0 0

log ) 1 ( /

log ) 1 ( /

log Y

Y E

E a

c a E

Y E c Y

p w p

w i i i i

i i i

i

i (4)

We do not express the coefficient of wage compression directly but resort to an expression in log, so as to make the estimation possible. A linear expression will not be estimable, as the coefficient of compression c in the theoretical equation will simply disappear into the constant of the empirical equation.

If c equals 1, this expression comes down to (3) expressed in logarithm. If c ranges between 0 and 1, there is wage compression. Conversely, if c is negative, the elasticity of relative wages to relative marginal product of labour becomes higher than unity: there is a stretched wage distribution 10 .

case, this study focuses on wage compression and not on labour costs compression, which can be even higher given the effect of the tax system.

10 However, the compression coefficient could have been expressed directly as a linear coefficient and not in an

expression in log. This choice is mainly dictated by the feasibility of the estimation: there is unfortunately no

way to estimate a linear coefficient of compression with our type of approach.

(15)

After rearranging (4), we obtain the relative employment equation:

( 0 ) 0

0 log log

log 1

log i w i w i

E c

E σ γ

+

− −

=

− (5)

where:

( 0 ) ( 0 ) ( 0 ) ( 0 )

0

0 log log

log 1 log log

log )) 1 ( log 1 log

log

log σ σ σ α α σ µ µ

γ −

− −

− +

− +

− − +

= i i i i i

i p p a a c

Y c Y

or more simply: γ i 0 = log Y i − log Y 0 if we assume that p, a, α and µ are constant across the types of workers/firms considered 11 . In this setting, the higher the wage compression, the lower the relative employment rate. If there is no wage compression, the coefficient of relative wage equals the elasticity of substitution 12 .

We select the reference group 0 as the least influenced by the wage compression according to the literature. If we suppose that the reference group 0 is not affected by the wage compression, we could use the equation (2), which has been written in a simpler form:

' 0 0

0 log

log E = − σ w + γ (2’)

The coefficient of compression cannot be estimated directly as it cannot be disentangled from the elasticity of substitution in (5). It can, however, be calculated indirectly by comparing the results of the labour demand equation (2) and the relative employment equation (5). If the wage coefficient of the employment equation (2) and that of the relative employment equation (5) are denoted by W1 (i.e. -σ) and W2 (i.e. -σ/(1-c)) respectively, the coefficient of compression is equal to 1-W1/W2.

However, the assumption of no wage compression in the reference group might not be true in reality and the workers in this group might not be on the standard labour demand curve. In such a case, a coefficient of compression c 0 should technically be added to the first term of the right-hand side of the equation (2’). Therefore, the estimated coefficient of

11 If the elasticity of substitution is equal to unity, the production function becomes a Cobb-Douglas function and relative employment rate takes the following form: log Ei - log E0 = - 1/(1-c)[log(w/p)i - log(w/p)0]+ γi0

12 In this setting, real labour cost elasticity gives a measure of the elasticity of substitution σ. In economic terms,

this parameter means that a growth of 1% in the relative cost of labour to capital will lead to a growth of σ% in

the ratio of capital to labour.

(16)

compression must be interpreted cautiously as the “relative or additional wage compression”

vis-à-vis the degree of compression in the reference group.

Equation (5) also shows that data on relative productivity or relative output are not needed to compute c. It suffices to estimate the wage coefficients in (2’) and (5), as the output produced by each group i and more broadly γ i 0 are captured by group-specific fixed effects.

This is extremely important as either relative productivity or relative output is very difficult to measure. In particular, there is no available data on any of these in the SES 2002.

However, we have made the strong assumption in (2’) and (5) that the elasticity of substitutions between employment and capital is the same across the different groups of workers/firms considered. It is a convenient way to identify wage compression. We implicitly suppose that any difference in the elasticity of employment to wages across groups is explained by wage compression. While this could be reasonable across education levels, this might however be quite debatable across occupations or sectors, where the elasticity of substitution may vary owing to the use of different production methods. If we relax this hypothesis, expression (5) becomes (assuming for simplicity that p, a, α and µ are constant across the groups considered):

( 0 ) 0 0 0

0 log

1 ) log (

1 log log

log i i i i w i

w c c w

E

E σ σ σ γ

− +

− −

− −

=

− (6)

' 0 0 0

0 log

log E = − σ w + γ (2’’)

This means that the way to identify wage compression c remains broadly unchanged, as relative employment can still be written as a function of the coefficient of wage compression, the elasticity of substitution and relative wages. The condition to get unbiased estimate of c is that equations (2’’) and (6) should be estimated separately for each group i rather than over the full population.

There are a couple of caveats associated with this methodology. First, the specification

used may influence the empirical results which partly reflect the assumed structure. We

assume a specific CES production function, albeit very standard and allowing substitution

between factors. Second, we do not express the coefficient of wage compression directly but

resort to an expression in log, so as to make the estimation possible.

(17)

5. E CONOMETRIC STRATEGY

Two equations are estimated: an employment equation and a relative employment equation. They are run according to different techniques: simple OLS and OLS with dummies (LSDV) taking into account the heterogeneity in the sample.

The employment equation corresponds to the standard labour demand equation (2’), which is only estimated over the benchmark group, denoted oB. It uses the level of employment (in logarithm) as the dependent variable and also includes GDP level and labour costs. Thus, the following general specification is estimated, where E denotes employment, Y real GDP and w nominal hourly wage rate (weighted average of gross hourly earnings of all individuals included in the group considered, excluding overtime payment), α coefficients of dummies, the i, s and g are country, corporate-size and gender indices and Y an output variable (country GDP or sectoral value-added). ε is the residual and the equation is:

Then we estimate a relative employment equation, similar to equation (5) where oB denotes the occupation chosen as a benchmark for the computation of relative employment and relative nominal wages. For convenience, the notations presented above are again taken below but with an apostrophe to signal that this is a symmetric but different specification. In addition, o is the occupation index. The estimated equation is 13 :

From there and as derived in section 4, the coefficient of compression can easily be computed as:

' c

' with

c β β

β

β < < < <

= 1 0 1 0

However, if β has a positive sign (which does not comply with the theory) or if c is above 1 or below 0, the result for c becomes suspicious and would be considered as insignificant. As

13 The variable lnYi is not absolutely necessary as it could be included in the country dummies. However, it is directly derived from the equation (5). It is indeed interesting to specifically control for the “economic size” of countries or sectors (in the specification with education levels and sectors). Moreover, it is important to set up an identical form for both equations (relative employment equation and employment equation) in order to make the results as comparable as possible. In any case the results are not altered by the inclusion of this variable.

g s i oB i g

s i oB g

s i g s i

oB ln w ln Y

E

ln = α + α + α − β + γ + ε

oB o with '

Y ln w '

ln w ' ' ' ' E '

ln E i o i s g

g s i oB

g s i o g

s i o g s i oB

g s i

o ⎟ ⎟ + + ≠

⎜ ⎜

− ⎛ + + +

⎟ =

⎜ ⎜

⎛ α α α α β γ ε

(18)

shown earlier, positive c points to the existence of wage compression, while a negative c points to the existence of wage extension.

The two equations will also be estimated with a different version of the SES aggregated database, where occupation and corporate size are replaced by educational attainment and sector, while gender and country are still among the dimensions used in the sample.

As a robustness check, we also run the employment and relative employment equation on each occupation o (and alternatively on each level of education) in order to take due account of the fact that the coefficient of compression and the coefficient of substitution between employment and capital may significantly vary across occupations. This comes down to estimating equation (2'') and (6) described in section 4.

⎪ ⎪

⎪ ⎪

⎪⎪ ⎪

<

+ <

=

+ +

⎟ +

⎜ ⎜

− ⎛ + +

⎟ =

⎜ ⎜

+ +

− + +

=

1 c 0 ' with

1 '' c

' Y ln ' w

ln w ''

ln w ' ' ' E '

ln E

Y ln w

ln E

ln , oB o

0 o

o

g s i o i g

s i oB g

s i oB

g s i o o

g s i g s i oB

g s i o

g s i oB i g

s i oB g

s i g s i oB

β β

β

ε γ

β β

α α α

ε γ

β α α α

While possible pooling bias could be removed through group-specific estimations, the flip side of this is that the estimation would be much less efficient due to the much lower number of observations used. This is the reason for running both pooled estimation and group-specific estimation in this paper.

As a major robustness check, we also control for the endogeneity of wages by using an

instrumental variable estimation (IV). Indeed, wages are an endogenous factor in a standard

competitive market, and particularly at the more aggregate level (where firms cannot be

argued to be price takers). While wages could influence employment via the labour-demand

channel, employment could also affect wages via its macroeconomic impact on disposable

income, effective demand and eventually total output. Labour-supply determinants also

simultaneously influence equilibrium employment and wages, which are actually observed

empirically, as opposed to the theoretical employment in the labour-demand and labour-

supply curves. The choice of instrument is then crucial. In this field, there is no panacea. The

best instrument to wages that could be thought of is lagged wages, which are likely to be

tightly related with current wages without impacting current employment directly. Therefore,

we use the previous (and first) issue of Structure of Earnings Survey, which was conducted in

(19)

1995 as a pilot and presents an identical disaggregation of wages and employment (e.g. sector, educational attainment, firm size, etc). However, its shortcomings are that it does not cover the 12 new member states, it only contains information on monthly earnings of full time employees (not on hourly earnings) 14 , it has many missing values especially in the service sector 15 and is not fully comparable with the 2002 survey, which offers more accurate and harmonized data on earnings. Therefore, the IV technique will be run on a smaller sample, covering the sole EU15 and comprising around one third of the observations of the full sample of SES2002 only. The good news is that 1995 wages are highly correlated with 2002 wages with a correlation coefficient being higher than 75% 16

Lastly, it should be mentioned that the statistical significance of control variables X k i , which are generally available at a more aggregated level than the dependent variable, such as national GDP, national institutions or sectoral value-added, should be interpreted very carefully. Moulton (1990) finds that the calculated standard errors of the aggregate variables could be seriously biased downward under OLS, because they are likely to be affected by the correlation of disturbance terms across individual observations sharing the same value of the aggregate variable (common-group effects). As a result, the estimated t-statistics tend to be higher than their real value, which might lead to erroneously reject the null hypothesis (i.e. the coefficient considered is not significantly different from zero). Moulton (1990) shows that the covariance matrix should be estimated by the following formula: σ ˆ ( XX ) 1 [ 1 + ( m − 1 ) ρ ˆ ]

instead of σ ˆ ( XX ) 1 in OLS, where X denotes the matrix of explanatory variables, ρ ˆ the estimate of the correlation coefficient of within-group error term (intra-group correlation) and m the ratio of the sample size to the number of groups 17 . We have, however, decided not to adjust the standard errors because this inference issue only affects the statistical significance of the various control variables and not the wage elasticities of employment in both employment and relative employment equations, from which are derived the coefficient of compression. The estimation of the latter is indeed our primary focus. The addition of controls

14 For data broken down by occupation, we calculate hourly earnings using the available information on mean weekly hours.

15 Real estate, renting and business activities for some countries (Germany, Greece and Ireland). Data on non- market related services (Public administration and defence, compulsory social security; Education; Health and social work, Other community, social, personal service activities) are missing for all countries.

16 A Student’s t-test T(N-2) suggests that the correlation is significant at a level well below 1%.

17 This adjustment is valid as long as the number of observations is the same for each group. For a more

thorough discussion on the issue of within-group effects, see also Bertrand et al. (2004) and Wooldridge (2006).

(20)

mainly aims at checking the robustness of the estimated compression to the possible existence of omitted variables, which can render the estimates of the parameters of interest biased and inconsistent.

6. E CONOMETRIC RESULTS

This section will present the empirical estimates of wage compression, first across types of occupation and second across educational attainments, using the two available semi- aggregated samples.

6.1 Wage compression across occupations

The empirical findings on the wage structure by occupation confirm the existing findings of the economic literature that there is a compressed wage distribution in Europe. Although the evidence appears little robust to the inclusion of countries dummies, the results is confirmed by many of the econometric estimations carried out in this paper. There is more uncertainty surrounding the exact magnitude of this compression as the estimated coefficients vary significantly across the samples and specifications used. Overall, the compression coefficient is around 40% 18 , meaning that relative wages (in logarithm) are reduced by about 40%, compared with what the productivity level should allow.

The choice of the benchmark B 0 is a crucial assumption and should be done carefully. We assume that the occupational group “professionals” is the least affected by the wage compression, as the economic literature indicates that the compression of the wage distribution is stronger at its lower end. The idea is that this group corresponds to high skilled occupations, while being more representative than the small group “legislators, senior officials and managers” in terms of employment share. However and for sake of robustness, the coefficient of wage compression has been estimated when choosing other occupational benchmarks than the professionals, as shown in Table 1.

18 This is the simple average of the estimates shown in table 1 and 2 for the whole sample.

(21)

Table 1. Estimation with different occupational benchmarks for the computation for relative employment and relative wages

Relative employment equation in log

Benchmark occupation in computing relative wage

Legislators, senior officials and managers

Professionals Technicians and associate professionals

Clerks Service workers and shop and market sales workers

Craft and related trades workers

Plant and machine operators and assemblers

Elementary occupations

relative emp relative emp relative emp relative

emp relative emp relative emp relative emp relative emp Relative wage -0.921*** -1.474*** -0.691*** -0.448*** -0.562*** -1.250*** -0.209 -0.386***

(0.0965) (0.125) (0.130) (0.134) (0.127) (0.131) (0.139) (0.124) GDP 0.0709*** 0.0398** -0.0229 -0.200*** 0.0269* 0.0818*** 0.0265* -0.0331**

(0.0154) (0.0175) (0.0143) (0.0153) (0.0154) (0.0153) (0.0159) (0.0151)

Observations 1970 1970 1970 1970 1970 1924 1924 1930

R-squared 0.239 0.139 0.143 0.445 0.298 0.472 0.337 0.143

Employment equation in log

Benchmark occupation in computing relative wage

Legislators, senior officials and managers

Professionals Technicians and associate professionals

Clerks Service workers and shop and market sales workers

Craft and related trades workers

Plant and machine operators and assemblers

Elementary occupations

emp emp emp emp emp emp emp emp

Wage -0.903*** -0.891*** -0.550*** -0.210*** -0.339*** -0.832*** -0.688*** -0.386***

(0.0653) (0.0736) (0.0538) (0.0433) (0.0475) (0.0474) (0.0569) (0.0443) GDP 1.022*** 1.042*** 1.020*** 1.095*** 0.924*** 0.999*** 1.008*** 0.992***

(0.0408) (0.0488) (0.0353) (0.0287) (0.0329) (0.0320) (0.0376) (0.0310)

Observations 286 286 286 286 286 275 275 276

R-squared 0.748 0.666 0.776 0.875 0.775 0.862 0.806 0.823 Absolute value of t statistics in parentheses

* significant at 10%; ** significant at 5%; *** significant at 1%

Estimated with occupation, firms' size and gender dummies

Estimated coefficient of compression with different benchmarks for the computation for relative wages

Benchmark occupation in computing relative wage

Legislators, senior officials and managers

Professionals Technicians and associate professionals

Clerks Service workers and shop and market sales workers

Craft and related trades workers

Plant and machine operators and assemblers

Elementary occupations

2% 40% 20% 53% 40% 33% 0%

Although the estimated coefficient of compression appears somewhat sensitive to the choice of the occupational benchmark, the existence of wage compression remains broadly confirmed. When taking professionals, technicians, clerks and service workers as a benchmark, a degree of compression is 40%, 20%, 53% and 40% respectively. However, with very low-paid and low-skilled occupations as a benchmark, there is no evidence of wage compression. As confirmed later in Table 4, this suggests that the wage compression is not uniform across occupations and applies more to very low-skilled occupations 19 .

Table 2 shows the results with the professionals as the reference group using simple OLS and OLS with dummies capturing the various dimensions. From a technical point of view,

19 Indeed, if we suppose that wages are mainly compressed in the low skilled occupations, it is logical that wage

compression is more obvious when taking higher skilled occupations as a base, since the number of lower-skilled

occupations in the standard taxonomy (ISCO-88) is higher than the higher-skilled occupations (legislators-

managers, professionals and technicians).

(22)

OLS with dummies (LSDV) ‘a priori’ seems to be the soundest approach, as it permits to correct for the heterogeneity inherent in each dimension considered (i.e. firm size, occupation and gender). For this reason, it offers a far better goodness-of-fit, especially for the relative employment equation.

Table 2. Overall results for EU27 (benchmark: Professionals)

Relative employment equation in log (deviation from professionals' wage and employment) Total sample Men

OLS OLS with dummies OLS OLS with dummies

relative employment relative employment relative employment relative employment

Relative wage -0.915*** -1.474*** -0.679*** -1.541***

(0.0690) (0.125) (0.0790) (0.133) GDP 0.0486*** 0.0398** -0.00409 -0.0193 (0.0179) (0.0175) (0.0205) (0.0174)

Observations 1970 1970 986 986

R-squared 0.087 0.139 0.070 0.351

Standard employment equation in log

OLS OLS with dummies OLS OLS with dummies

employment employment employment employment

Wage -0.829*** -0.891*** -0.552*** -0.591***

(0.0825) (0.0736) (0.0976) (0.0856) GDP 1.041*** 1.042*** 1.064*** 1.058***

(0.0551) (0.0488) (0.0642) (0.0560)

Observations 286 286 143 143

R-squared 0.563 0.666 0.663 0.753

Absolute value of z statistics in parentheses

* significant at 10%; ** significant at 5%; *** significant at 1%

Estimated coefficient of compression

OLS OLS with dummies OLS OLS with dummies

Relative employment

equation 9% 40% 19% 62%

Technically, the existence of wage compression can be seen through the fact that the elasticity of relative employment to relative wages is higher in absolute value than the elasticity of employment to wages for the benchmark group, which is meant to capture the elasticity of substitution between employment and capital. OLS with dummies provides an estimated value of 40%, which is much higher than that derived from simple OLS. We also run the regressions for men only, as the rich literature on labour supply argues that they experience much less sample selection bias than women, especially married women. These regressions confirm the existence of wage compression and point to higher compression for men compared with the entire sample.

Table 3 shows that wage compression is higher in EU25 than in EU27, suggesting a lower degree of compression in the two acceding countries (Bulgaria and Romania).

Moreover, there is no evidence of compression in the new member states which acceded in

(23)

2004 (EU10). This confirms various indications regarding the decentralised wage setting in these countries, where wages are negotiated at the firm level (Pichelmann, 2001). The euro area experiences a higher degree of wage compression than in the EU15. These findings also hold true when considering the male sub-sample only.

Table 3. Wage compression by geographical areas (benchmark: Professionals)

Relative employment equation in log (deviation from professionals' wage and employment)

Total (eu27) eu25 eu15 euro area eu10

relative employment relative

employment relative

employment relative

employment relative

employment Relative wage -1.474*** -1.642*** -2.782*** -3.128*** -0.530***

(0.125) (0.137) (0.195) (0.218) (0.167)

Observations 1970 1690 949 725 741

R-squared 0.139 0.160 0.318 0.379 0.189

Employment equation in log

Total (eu27) eu25 eu15 euro area eu10

employment employment employment employment employment

Wage -0.891*** -0.890*** -2.552*** -2.486*** -1.137***

(0.0736) (0.101) (0.461) (0.560) (0.138)

Observations 286 246 140 108 106

R-squared 0.666 0.602 0.545 0.494 0.819

Absolute value of t statistics in parentheses

* significant at 10%; ** significant at 5%; *** significant at 1%

Estimated by OLS with GDP, firm size, gender and occupation dummies (but without countries dummies)

Estimated coefficient of compression

Total (eu27) eu25 eu15 euro area eu10

40% 46% 8% 21% °

Table 4 presents estimations of the coefficient of compression for each occupation separately (with the professionals as a benchmark). These allow the coefficient of substitution to vary across occupations, improving the identification of the coefficient of compression.

Indeed, in the pooled estimation, the value of wage compression might partly and mistakenly capture the differences in the employment-capital substitution coefficient across occupations.

Thus, Table 4 broadly confirms the widespread idea that wages are more compressed at

the lower end of the wage distribution. Indeed, legislators, senior officials and managers -

who correspond to the highest paid and highest skilled occupations - and technicians and

associate professionals record relatively low estimated coefficients of compression vis-à-vis

professionals. Conversely, wages for clerks show the highest coefficient of compression,

followed by service workers and shop and market sales workers, craft workers and elementary

occupations. Therefore, the three least paid and skilled occupational groups (elementary

(24)

occupations, craft workers and service and shop workers) turn out to have high relative wages vis-à-vis professionals.

Table 4. Wage compression by occupations for EU27 (benchmark: Professionals) (allowing for different coefficients of capital-labour substitution across occupations)

Legislators, senior officials and managers

Technicians and associate professionals

Clerks Service workers and shop and market sales workers

Craft and related trades workers

Plant and machine operators and assemblers

Elementary occupations

Relative employment equation in log (deviation from professionals' wage and employment)

Relative

employment Relative

employment Relative

employment Relative

employment Relative

employment Relative

employment Relative employment Relative wage -1.798*** -1.582*** -1.992*** -1.833*** -2.232*** -1.389*** -1.646***

(0.242) (0.233) (0.328) (0.271) (0.307) (0.304) (0.268) Benchmark -0.0307 0.291*** 0.656*** 0.590*** 0.0637 0.135* 0.545***

wage (0.0519) (0.0435) (0.0645) (0.0633) (0.0746) (0.0718) (0.0649)

Observations 286 286 286 286 275 275 276

R-squared 0.390 0.310 0.532 0.425 0.419 0.298 0.307

Employment equation in log (professionals)

Employment Employment Employment Employment Employment Employment Employment Wage -0.891*** -0.891*** -0.891*** -0.891*** -0.891*** -0.891*** -0.891***

(0.0736) (0.0736) (0.0736) (0.0736) (0.0736) (0.0736) (0.0736)

Observations 286 286 286 286 286 286 286

R-squared 0.666 0.666 0.666 0.666 0.666 0.666 0.666

Coefficient of

compression 50% 52% 66% 63% 60% 42% 59%

Absolute value of z statistics in parentheses: * significant at 10%; ** significant at 5%; *** significant at 1%

Estimated in log with country-specific GDP and firm size, gender and occupation dummies.

The results of the employment equation appear identical across all occupations as it only refers to the employment equation for the benchmark group (the professionals).

6.2 Wage compression across educational attainments

As far as the wage structure by level of education is concerned, the empirical findings broadly confirm that the distribution of wages is compressed in Europe. However, evidence appears very sensitive to the inclusion of various dummies. If education dummies and country dummies are added, the coefficient of compression becomes insignificant, in the sense that the underlying coefficient of either the relative employment equation or the employment equation turns insignificant or positive, which is difficult to reconcile with the economic theory. An explanation might be that wage differentials are very similar across educational groups and sectors within the same country, which would mean that using country dummies artificially removes part of the explanatory power contained in the wage variable. The most plausible explanation would, however, be that the variable “educational attainment” (referring to four levels of education only) is too coarse to capture the various levels of professional skills.

Moreover, the level of education attained refers to the potential skills of the worker rather

than the skills actually required for the job currently occupied, which is the most relevant

(25)

explanatory variable for the level of wage paid by the employer. In this respect, the phenomenon of over-qualification and of increasing enrolment in tertiary education in some European countries might blur the overall picture.

The choice of the educational benchmark B 0 is a crucial assumption and should be done carefully. Consistently with the assumption taken in section 6.1, we assume that the educational group “tertiary education” is the least affected by the wage compression. This assumption is made in accordance with the literature, which indicates that the compression of the wage distribution is stronger at its lower end. However, the coefficient of wage compression has been estimated when choosing other educational benchmarks than tertiary education, as shown in Table 5. The latter highlights the lack of robustness of the estimates:

there is no clear evidence of wage compression any longer when considering lower or medium education levels as a benchmark. This seems to confirm that wage compression is not uniform across the levels of education attained by employees and that wage compression applies more to the low-skilled. This may also suggest that the findings are fragile due to the coarse measure of skills provided by the educational level reached by the employee.

Table 5. Estimation with different educational benchmarks for the computation for relative wages

Relative employment equation Pre-primary, primary and lower secondary education - levels 0-2 (ISCED 1997)

Upper secondary education - level 3 (ISCED 1997)

Post-secondary non-tertiary education - level 4 (ISCED 1997)

Tertiary education - levels 5-6 (ISCED 1997) Relative employment Relative employment Relative employment Relative employment

Relative wage -0.0639 0.227* -0.127 -0.892***

(0.161) (0.126) (0.200) (0.161)

Value added -0.376*** -0.0395 0.415*** 0.150***

(0.0337) (0.0286) (0.0433) (0.0317)

Observations 1191 1193 897 1193

R-squared 0.221 0.046 0.139 0.179

Employment equation

Pre-primary, primary and lower secondary education - levels 0- 2 (ISCED 1997)

Upper secondary education - level 3 (ISCED 1997)

Post-secondary non-tertiary education - level 4 (ISCED 1997)

Tertiary education - levels 5-6 (ISCED 1997)

employment employment employment employment

Wage -0.0118 -0.375*** -0.951*** -0.488***

(0.0510) (0.0455) (0.111) (0.0480)

Value added 1.048*** 0.897*** 0.702*** 0.821***

(0.0358) (0.0297) (0.0760) (0.0272)

Observations 446 448 299 448

R-squared 0.813 0.813 0.424 0.852

Absolute value of t statistics in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%

Estimated by OLS with gender and sector dummies.

Estimated coefficient of compression

° ° ° 45%

°Not significant, as the wage coefficient of the relative employment equation or the employment equation is either not statistically

significant or of a positive sign (contrary to the theory).

Références

Documents relatifs

Les items qui contribuent le plus au facteur sont d’une part l’opposition à un droit absolu des femmes à s’habiller comme elles l’entendent, d’autre part le sentiment que

Work in [3] proposes a distributed approach that adapts to the size of the network, i.e., the number of coexisting WBANs, and to the density of sensors forming each individual WBAN

To meet this objective, I perform four main tasks. First, I obtain panel estimates of all the key components of the Mincerian wage regression function in a context where i)

First, as in Saunders, Strock, and Travlos (1990), Esty (1998), and Grossman and Imai (2010), among others, one can use the volatility of share prices (this corresponds to

wklv yduldeoh dhfwv wkh hyroxwlrq ri dfwxdo zdjhv/ surgxfwlylw|/ djjuhjdwh frqvxpswlrq/ rxwsxw/ dqg hpsor|phqw1 Lq Khoozlj dqg Luphq +5334, zh kdyh vkrzq wkdw xqghu odlvvh}0idluh/

The risk of dengue transmission is influenced by various factors, including trade of goods and human mobility, population density, urbanization, climate, presence of

biotechnologie végétale de l’université des sciences technologiques d’Oran, qui ma aidé surtouts pour la conservation des souches.  Je voudrais également

Celles-ci sont beaucoup plus présentes dans le cas d’un noyau pair-impair ou impair-impair que dans le cas d’un noyau pair-pair (essentiellement composé de transitions