Was Thatcher Right?

Was Thatcher Right?

The impact of labor unions on unemployment

Author: Geert Nijenhuis S1632078

Supervisor: dr. T.M. Stelder July 2014

Abstract

This research tests the hypothesis whether union power has influence on the unemployment rate. Fifteen countries are analyzed between 1981 and 2010. Based on previous literature we conclude that centralization of bargaining is the best measure of union bargaining power. A negative relationship of centralization on unemployment is found. The hump-shaped relationship of Calmfors and Driffill (1988) cannot be confirmed by this paper. Furthermore, the impact of union density seems to be mitigated by the degree of centralization. However, this coefficient does not seem to be robust.

Table of Content

4. Data description and descriptive statistics ... 18

4.1 Centralization and Coordination ... 18

5.3 Fixed or random effects... 31

5.4 Expected signs ... 32

6. Results ... 33

6.1 Linear and non-linear base model ... 33

6.2 Alternative non-linear model ... 34

6.3 Interaction model... 36

6.4 Robustness checks ... 37

6.4.1 Time periods ... 38

6.4.2 Centralization/coordination alternatives ... 38

6.4.3 Sensitivity analysis ... 40

7. Summary and conclusion ... 41

1. Introduction

Reducing unemployment is a goal of almost every single government. It is one of the few issues in which labor unions and governments have common interests. Governments due to the simple fact that employed citizens bring in taxes and do not take the advantage of unemployment benefits. Unions, since it is normally in their members best interest to have a job with a decent salary. However the means of achieving this goal are heavily debated and often disagreed upon.

In order to change unemployment, we should know what causes unemployment. A great number of determinants of unemployment are identified by economic theory and previous empirical studies. Among others, the policy and institutional determinants include unemployment benefits, employment protection legislation, active labor market policies, minimum wages and housing policies, anti-competitive product market regulation, taxes, and labor union bargaining power (Bassanini & Duval, 2006). In this analysis we will focus our attention on labor union bargaining power.

Margaret Thatcher was not too fond of strong labor unions. “We had to fight the enemy without in the Falklands. We always have to be aware of the enemy within, which is much more difficult to fight and more dangerous to liberty” (Thatcher, 1984). This is what Margaret Thatcher said to the 1922 Committee1 when she was referring to coal miners during the 1984-1985 miners’ strike. The strike symbolized the power of labor unions and the dedication of Thatcher to reduce the power of labor unions in order to improve the labor market performance (Gamble, 1980). The rationale behind it was that wage increases would be restrained if market forces were allowed to play a larger role (Calmfors & Driffill, 1988). Union bargaining power is a rather difficult variable to capture quantitatively. A commonly used variable is the proportion of employees in unions, in other words union density. Another variable used in different studies is union coverage. This is the proportion of workers covered by collective bargaining agreement. However, much attention has been devoted to the degree of centralization of bargaining as a measure of union strength, which is later adjusted to

1 _{The 1922 Committee if formally known as the Conservative Private Members’ Committee, is a committee of}

Conservative Members of the British Parliament (Ball, 1990)

coordination by certain studies (Baker, et al., 2002). In this research the degree of centralization will be used as a measure of union strength.

The labor market performance can be measured in many ways. However, the main indicator of labor market performance is unemployment (Heijdra, 2009). Hence, in our analysis we use unemployment as a measure of labor market performance since a low unemployment rate would be a goal of labor unions, employers and government. By analyzing the influence of union bargaining structure on unemployment, we investigate whether or not labor unions indirectly improve or harm the unemployment rate.

Visser (2013b) developed a database containing more data than before. This database enables us to test whether the conclusions of the previous century are still valid in the twenty-first century. We will examine the following fifteen OECD countries. Austria, Belgium, Canada, Germany, Denmark, Finland, France, Republic of Ireland, Italy, Japan, the Netherlands, Norway, Sweden, United Kingdom, and the United States (OECD-15). We will use annual data from 1981 until 2010.

In the next section the theory on union power and unemployment will be presented. Section three will cover the available literature on this topic. In the following section we will discuss the data and the descriptive statistics. The fifth section will elaborate on the model that is used. Followed by a discussion on the results. And finally the summary and conclusion.

2. Theory

2.1 Unemployment

In order to understand unemployment, a few stylized facts should be considered (Layard, et al., 2005; Heijdra, 2009). Fact 1, unemployment fluctuates over time. Some short-term changes are reversed quickly, although some are large-circular changes. Figure 1 shows the unemployment rates of the Netherlands (NL), Germany (DE), and the average of the OECD-15 from 1960 until 2012. The fluctuations of the Netherlands are larger than the average fluctuations of all fifteen OECD countries together.

Figure 1: Rate of unemployment as a percentage of the civilian labor force 0 2 4 6 8 10 12 14 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 NL DE OECD-15 U n e m p lo y m e n t ra te i n % Source: OECD (2014)

Fact 2, fluctuations of unemployment vary more between business cycles than within business cycles. This is better explained with historical data. Unemployment in the USA rose much more between 1920’s and 1930’s (the Great Depression) and was extremely low during and right after the Second World War. Heijdra (2009) tests unemployment for persistence and concludes that unemployment persistence is high. This implies that it takes a long time for unemployment shocks to die out. This results in large fluctuations between business cycles. Fact 3, there is a strong correlation between the rise of unemployment in Europe and the increase of long-term unemployment. It illustrates that the rise of unemployment is not caused by people losing their job; it is caused by unemployed people not finding a job.

Fact 4, Unemployment shows no trend in the very long-run. In other words, in the very long run there is not a bigger or smaller chance of being unemployed.

Fact 5, there is a large variation in unemployment between different countries. As shown by figure 1, there is a large difference between unemployment of the different (group of) countries. Figure 1 illustrates the difference between the Netherlands, Germany, and the 15. We see that the Netherlands and Germany have some extremes, where the OECD-15 ‘floats’ around in the middle. Hence we can conclude that the other countries do not have the same unemployment rate. These deviations from the ‘group’ are to a large extent due to labor market institutions (Bassanini & Duval, 2006).

Fact 6, the vast majority of the unemployed want to be employed. Few of the unemployed quit their jobs voluntarily, to find another job for example. Most of the unemployed people are laid off by their employers (Layard, et al., 2005).

Fact 7, unemployment significantly differs between races, region, occupations, and age-groups. The unemployment rate of the age-group 15-24 is twice or three times as high as the age group 24-64 (OECD, 2014).

2.2 Unions

Labor Unions are organizations that represent the interests of their members which are the employees. These interests are for example wages, health care, holidays, safety standards, hours per week labor, and retirement conditions. They negotiate with representatives of a firm, an employers’ organization, or in so called tripartite negotiations with the government as a third party.

In the late seventeenth century, the first labor unions started in the United Kingdom. They were the successors of the guilds. Guilds started off as relatively closed organizations that organized the sales and production of a certain product or service. They evolved more and more towards an organization that took care of a member, in case of illness or other difficulties. During the industrial revolution, certain sections of guilds transformed to organizations that represented wage-earners (Reid, 2004). The first unions originate from the late seventeenth century and were formed during the industrial revolution. They became industrial organizations representing workers in semiskilled positions, in order to secure them from replacement by immigrants and unemployed. Later they became national organizations representing all workers from all work levels which wanted to be represented. However, the aim of the unions was egalitarian wages, which was not in the interest of skilled workers. As a result, unions were mainly the representatives of semiskilled and unskilled workers. Nowadays, unions are best represented in the public sector.

2.3 Bargaining power

the unions. These ‘free-riders’ are included in the union coverage, nevertheless not in the union density (Powdthavee, 2011).

Some relatively new and more complicated measurements are centralization and coordination. Centralization refers to the level of decision making, for example at company/plant level, industry level, central level or in between. Coordination refers the level at which contracts are negotiated. Coordination is normally executed at the same level or a higher level than centralization (Boeri & van Ours, 2013).

There are a number of models that elaborate on the way unions use their bargaining power. Three different models can be distinguished, the monopoly model of the labor union, the right to manage model, and efficient bargaining model (Heijdra, 2009).

The models have in common that they maximize the utility function of the union given by: 𝑽(𝒘, 𝑳) ≡_𝑵𝑳∙ 𝒖(𝒘) + �𝟏 −_𝑵𝑳� 𝒖(𝑩) (1)

𝑉(𝑤, 𝐿)= utility dependent on w(wage) and L(number of employed members). 𝐿

𝑁 = proportions of employed members, Where N = union members (L≤ N)

𝑢(𝑤) = indirect utility function of a union member receiving wage.

𝑢(𝐵) = indirect utility function of a union member receiving unemployment benefit

A representative firm is modeled in a similar way. First it is assumed that firms maximize their profits by maximizing the profit function given by:

𝝅(𝒘, 𝑳) = 𝑨𝑭(𝑳, 𝑲�) − 𝒘𝑳 (2)

𝜋(𝑤, 𝐿) = profit dependent on wage (w) and labor (L)

𝐴𝐹(𝐿, 𝐾�) = production function dependent on labor (L) and fixed capital (𝐾�) 𝑤𝐿 = cost function wage (w) multiplied by labor (L)

In order to find the right amount of labor to maximize profits, we have𝝅_𝑳 =𝝏𝝅

𝝏𝑳 = 𝟎, which

yields the labor demand function:

𝝅𝑳= 𝑨𝑭𝑳(𝑳, 𝑲�) − 𝒘 = 𝟎 ⟺ 𝑳𝑫 = 𝑳𝑫(𝒘, 𝑨, 𝑲�) (3)

𝐿𝐷 _{= labor demand of a representative firm, which is dependent on wage (w),}

productivity index (A), and fixed capital (𝐾�).

The first and the oldest model is the monopoly model (Dunlop, 1944). In this model it is believed that one monopoly union sells labor to firms. It faces the labor demand function (3), with this knowledge it sets its wage in order to maximize the utility function (1). Figure 2 illustrate graphically the monopoly union solution.

Figure 2: Monopoly Union Model

Source: Heijdra (2009)

𝐿𝐷 _{Is the labor demand function. 𝑉}𝑚 _{Is the indifference curve of the monopoly union. The}

combinations of labor and wage along the indifference curve create equal amounts of utility for the monopoly union. The wage is set by the monopoly union on 𝑤𝑚, given the demand function 𝐿𝐷, and the employer organization hires 𝐿𝑀 amount of labor. If labor was not unionized, the solution would be different. In that case, all assumed identical workers would work for their ‘reservation wage’ (B). Reservation wage is the wage for which workers are indifferent in being employed or receiving an unemployment benefit. As illustrated in figure 2, the monopoly union causes a higher unemployment and a higher real wage.

Figure 3 graphically shows the ‘Right to manage’ model (Leontief, 1946). In the model the amount of labor is still chosen by the employer. However, the employer and the union bargain about the wage. The value of the ‘markup’ depends on the bargaining power (λ,

where 0 ≤ λ ≤ 1) of the union, where the markup represents the difference between the wage and the reservation wage.

Figure 3: right-to-manage Model

Source: Heijdra (2009)

The solution still lies on the labor demand function. It depends on the bargaining power (λ) where the exact location of point R is. When λ=1 the solution is equal to the monopoly union solution (point M). When λ=0 the solution is equal to the competitive solution (point C). The major problem with this model is that the outcome of the wage and labor amount is not Pareto-efficient. Since, it is possible for one party to be better off without harming the other party. This can be demonstrated making use of figure 3. The indifference cure of the union is illustrated by VR and the firm is indifferent along the 𝜋𝑅-line, since all combination of labor and wage provide the same amount of profit. Hence, when we take the solution on point ER the firm is still equally satisfied with the same amount of profit. However, the utility of the union is increased, since the solution is on a higher indifference curve. In order to prevent this Pareto-inefficient outcome, McDonald and Solow(1981) developed the efficient bargaining model.

In the ‘Efficient bargaining model’ firms and union bargain about wages and the amount of labor. They maximize the utility of the union as well as the profit of the firm. This typically results in higher employment and lower wages.

Figure 4: Efficient Bargaining Model

Source: Heijdra (2009)

As shown by figure 4, the solutions do not lie on the demand function anymore. The so-called contract curve (CD-curve) is the Pareto-efficient solution line. Bargaining power does no longer result in higher unemployment. It is believed that higher bargaining power will move the wage-labor combination closer to point D. Point D is the full employment (FE) combination.

3. Literature review

3.1 Union power

In the previous section we talked about bargaining power and the influence on the different models and accompanying solutions. In order to create a valuable analysis we have to quantify the bargaining power variable. In different papers economists and sociologists used different quantitative measurements of bargaining power. These measurements are, strike activity, share of wages, union density, union coverage, centralization and coordination.

3.1.1 Strikes

was an appropriate measure, they test whether the rate of strike activity is positively related to the change of money wages. They find no relation between the rate of money wages and strike activity. However, what they do find is some association between strike activity and increasing union membership. They conclude that strikes generate (positive) publicity, which helps raising new members. The new members subsequently generate more bargaining power. Critics argue that strike activity should not be accepted as a form of union power. Since strikes should only be used as a deterrent.

3.1.2 Wage shares

In his paper Rowthorn (1999) uses wage shares (the proportion of the total wage bill of GDP) to measure bargaining power. Boeri et al., (2001) do find similar trends of union membership and share of wages. However, we did not find any strong evidence in the existing literature. Critics say that unions do not pursue an increase of the wage share, since they do not try to maximize the wage of the whole population but only of their members. Others state that the change of the wage share is just a result of a shift in technology (Blanchard & Wolfers, 2000).

3.1.3 Union density

A more frequently measure of bargaining power is union density. This is the proportion of union members of the total labor force. As figure 5 illustrates, in the last 53 years union density (UD) has steadily declined. In 1960, one of every three employees was a union member, followed by a decline during the late seventies. This resulted in 2012, in a density rate which was only fifty percent of its value of 1975. One of the reasons was the decline of employers in the manufacturing sector. The union density rate was traditionally high in the manufacturing as well as the public sector. However, the share of manufacturing workers declined rapidly over the last fifty year. For example, the share of manufacturing workers in the Netherlands declined from about 40 percent in 1960 to nearly 15 percent in 2010. Between 1995 and 2008, the manufacturing unions’ density was about 8 percent higher than the general union density (see appendix A for more details). Hence, the decline in union density was partly caused by a decline in manufacturing workers. As shown in figure A1 and A2 of appendix A, union density declined in the manufacturing sector as well. This is caused by other factors as members-age and the free-rider problem.

Figure 5 shows that while union density declined (UD), the unemployment rate (UR) first increased and later started ‘floating’ around 7 percent. This possible related phenomenon is one of the countless reasons for labor economists to analyze the relationship between unemployment and union density as a measure of union power.

Figure 5: OECD-15 countries (1960-2012)

Source: OECD (2014)

The IMF (2003) used union density in order to investigate the causes of the persistence of unemployment. They stated that the relatively high unemployment in Europe is caused by labor market institutions. Many different causes were reviewed; one of them was union density. They used data of twenty OECD-countries between 1960 and 1998. The dependent variable was aggregate unemployment versus union density as the independent variable. Highly significant evidence was found for positive relation between the unemployment rate and unionization (union density). This contradicts what figure 5 seems to project. Researchers believe that a higher density rate results in higher labor costs. They argue that the rise in unemployment was due to other external factors, such as the oil crisis. Consequently, without the decline in union density unemployment would even be higher. The IMF (2003) believes that greater unionization reflects less competition in the labor market, resulting in higher wages and lower unemployment. Various other studies found a highly positive relationship between unemployment and union density as well (Nickell, et al., 2001) (Boone & van Ours, 2004). Furthermore, Nickell, et al., (2001) found evidence that a higher unionization increases labor costs, resulting in a higher unemployment rate.

3.1.4 Union coverage

An even better measure of union bargaining power is provided by union coverage (Boeri & van Ours, 2013). This is the proportion of workers who are included in bargaining

agreements. Nowadays union coverage rates are often significantly higher than union density (see figure 6b). This is a relatively new phenomenon for two reasons. First of all, union density decreased over the last decades. Second, governments often practice laws that prohibit firms from excluding non-union members from the bargaining agreements, increasing union coverage relatively to union density. Although data for union coverage is hard to find, many economist see it as a worthy measure of union power.

Figure 6a: Coverage versus union density Figure 6b: Coverage, union density trend

Source: Visser (2013a) Source: Visser (2013a)

Figure 6a suggests a positive relationship between union coverage and union density. Figure 6b illustrates a more or less similar trend in union coverage and density. Furthermore, from the previous section we know that unemployment and union density are positively related. All this combined would imply a positive relationship between unemployment rate and union coverage. However, not all studies found this relationship. Belot & van Ours, (2001) find a significant negative relationship between union coverage and the aggregate unemployment rate. They argue that union coverage should be seen as a proxy of centralization and not as a variable related to union density. Centralization is a measure of union power which we will analyze below.

3.1.5 Union centralization/coordination

The most often used measurements of union power are union centralization and union coordination. The reason for this is the following. When there is complete centralization, there is only one union or one union federation active at a central bargaining level. As a result, this union or federation can exhibit its monopoly power as analyzed in the theoretical overview.

The definition of centralization is the actual level at which contracts are signed. Union coordination is the level at which negotiations takes place (Boeri & van Ours, 2013). In a way these variables are closely linked. However, in different studies they have different outcomes. The most influential and analyzed outcome is the hump-shaped relationship between unemployment and centralization developed by Calmfors and Driffill (1988). They find a significant parabolic relationship as illustrated in figure 7. This paper is still of great interest of many sociologists, political scientists and economists. Calmfors and Driffill (1988) believe that the high unemployment rate in the Netherlands and Belgium in the eighties might has something to do with corporatism. Although corporatism hard to define, they see the centralization of the wage-setting process as the key measure of corporatism (Heijdra, 2009). In order to analyze the effect, they create a centralization ranking of countries, using data between 1963 and 1985. With, among other countries, Japan, the United States, and Canada as decentralized countries Sweden and Austria as highly centralized countries and the Netherlands and Belgium as the intermediate countries.

Figure 7: Calmfors and Driffill model. Unemployment and centralization

Source: Mares (2006)

They find evidence that the intermediate case causes a higher unemployment rate. Centralized countries seem to have lower unemployment rates because they try to internalize the external effects. They do this mainly because high real wages result in a number of negative externalities (Booth, et al., 2000; Calmfors, 1993). The most obvious externality is a rising consumer price. A higher wage in a part of a country will put upward pressure on the price of the consumption basket. This will influence the purchasing power parity in the rest of the economy in a negative way. At the same time, this rise in prices will cause the aggregate demand to fall. This will result in less production accompanied by less employment. Another externality is the effect on input prices. A wage rise in a sector that produces input products will affect other sector because they are all interlinked. A different externality is the fiscal effect. A wage rise will cause higher unemployment (since higher labor costs means less labor), the total costs on unemployment benefits will go up and at the same time the tax base decreases (since less people are employed). In order to compensate for this loss, taxes have to go up, resulting in a lower disposable income. A psychological externality is the envy and jealousy effects. Employees always compare their wage to that of others. When wages increases at one firm, employees at the neighboring firm want the same wage. If the neighboring firm refuses to increase the wages, the productivity might be directly and negatively influenced.

Although the study of Calmfors and Driffill (1988) seems convincing and is supported by other studies, some find evidence for a negative relationship between unemployment and centralization/coordination. For example, Nickell (1997) finds a negative linear relationship between unemployment and union coordination. The OECD finds overall the same results (OECD, 1997). They find that high centralization or coordination is associated with low unemployment and high employment. The OECD (1997) further mentions that more analyses are needed in order to fully abandon the hump-shaped analysis of Calmfors and Driffill (1988), since not all of their findings are significant.

A more recent study found contrasting results (Di Tella & MacCulloch, 2005). They found no evidence for a hump-shaped relationship between unemployment and centralization/coordination. What they did find was a positive relationship between unemployment and centralization/coordination, contradicting existing theory and empirical evidence. Although this contradicts earlier results, it is in line with the trend toward greater bargaining decentralization in OECD-countries. An explanation for this trend finds its origin in the hump-shaped relationship found by Calmfors and Driffill. When countries are positioned in the middle of the curve and are bargaining is neither centralized nor decentralized, decentralization will result in lower unemployment. A more theoretical explanation is that centralized wage agreements are too general. In other words, the agreements are not optimal for a certain sector or area. By decentralizing wage bargaining, ‘custom made’ agreements for sectors and areas are easier to acquire (Boeri & van Ours, 2013).

3.2 Expectations

than just the wage setting. During crisis, attaining a high employment rate will be of more concern to unions than normally. This results in more wage flexibility and possible negotiations about hours worked and employment (OECD, 2012). Hence, we believe that highly centralized economies will attain lower unemployment compared to intermediate-centralized economies. Furthermore, for deintermediate-centralized economies we believe that market forces will be of great importance again. This will result in relatively low wages in exchange for relatively high employment. To conclude, for the new millennium we expect the hump-shaped relationship of Calmfors and Driffill (1988) to rise again.

4. Data description and descriptive statistics

In this section we will further elaborate on the variables that influence unemployment. First of all, the concept of centralization and coordination will be thoroughly discussed. Second, another frequently measure of union bargaining power will be discussed, union density. Thereafter, we will briefly discuss three control variables that influence unemployment, nonetheless not related to labor unions.

4.1 Centralization and Coordination

In order to capture the quantitative measurements of union power, the centralization and coordination scores are often used. These two measurements are not exactly the same, although they are highly correlated. Centralization is best explained as the level at which wages are bargained and agreed upon. To determine the quantitative measure of centralization, we should consider three important elements (Kenworthy, 2001a). The first one is the level at which the wage agreements are signed. This might be peak/central/intersectoral, sector/industry, or company/plant. The second one is the share of the workforce for which wages are negotiated at each level. If bargaining agreements only cover a small share of the workforce, but the agreements are signed high up in the economy, it is hard to call this a highly centralized bargaining structure. Third is the degree of horizontal centralization. In England it is common that wage agreements between a labor union and a firm only covers a share of the firm’s workforce. In Japan the majority of the wage agreements cover all employees of a certain sector or a firm. Hence, negotiations occur at the same bargaining level, however, the degree of centralization of Japan will be higher.

The first widely known measure of centralization was a measure originally created by Douglas Hibbs (Hibbs, 1976). David Cameron revised this measure in order to capture “the scope of collective bargaining, ranging from restrictions on collective bargaining on the one hand to economy-wide bargaining on the other” (Cameron, 1984). The measure is quantitated by an index, ranging for 0 to 1, started at totally decentralized until totally centralized, respectively.

Calmfors and Driffill (1988) wrote the well-known ‘hump-shaped analysis’ paper mentioned in the previous section. In their analysis they made use of a rank-ordered centralization measure. Ranking countries from the most centralized until the most decentralized country. They subdivided the countries in three different groups, centralized, intermediate, and decentralized. From that moment on, this measure was the most commonly used measure of centralization.

The Calmfors and Driffill (1988) ranking was the best measure until the OECD (1997) created a measure. The OECD (1997) assigned a score to every country, 1, 2 or, 3, representing company-level, sector-level, and economy-wide bargaining, respectively. In other words, 1 represented decentralized bargaining, 2 intermediate bargaining, and 3 centralized bargaining.

What these three measures all have in common, is that all the measurements are time-invariant. A country is assigned to an index, rank, or score, which does not change in time. Apart from that this is unlikely, it forced researchers to check whether the assigned index, rank, or score was still applicable. In order to tackle this inconvenience a number of time-variant indicators of centralization were created.

Golden, Lange, and Wallerstein (1998) developed three measures of centralization. Iversen (1999) developed one, Traxler, Blaschke, and Kittel (2001) another, and more recently Visser (2013a). As earlier mentioned these measurements improved the ones of the OECD, Calmfors and Driffill, and Cameron in the way that they are time-variant. This is not the only improvement, the new measurements of Traxler et al. (2001) and Iversen (1999) are tend to be grounded on more and better data (Kenworthy, 2001a).

to bargaining predominantly at central or cross-industry level, respectively. A country is assigned to a score (1, 3 or 5) when the agreement accounts for at least two-thirds of the total bargaining coverage rate. If it accounts for more than one-third, but for less than two-thirds of the coverage rate, an intermediate or mixed situation is assigned to that country, accompanied by a score of 2 or 4. Hence, this results in a division of the different bargaining structure quite similar to Calmfors and Driffill (1988). However, it is improved by making use of intermediate levels (the levels 2 and 4), like the OECD study (OECD, 2004).

In our view, the best indicator of centralization of bargaining power is the one of Iversen (1999) for the following reasons. First of all, it is time-variant, which enables us to determine whether a change in the degree of centralization results in a change of unemployment. This is helpful, since wage setting structure changed quite a bit during the last 30 years (see figure 8). Second, it is grounded on better and more objective data. Nowadays there is much more data available on the details of wage setting than twenty to thirty years ago. Hence, the objectivity of data can easily be improved. Furthermore, Iversen (1999) focusses on the on the degree of centralization itself, instead centralization-related activities by confederations like the Golden, Lange, and Wallerstein (1998) indicators (Kenworthy, 2001a). Finally, the data of Iversen (1999) is available for a large time-period as well as for a large set of countries. All this together brings us to the Iversen-index as the measure of centralization of wage bargaining.

Figure 8: Iversen Centralization index

A minor problem arises, since Iversen only made the data available for the period between 1973 and 1995. Fortunately, Visser (2013b) created data for a new period up to 2011 using the same method as Iversen. Hence in our analysis we will use the centralization measure developed by Iversen (1999), which is derived from the ICTWSS-database (Visser, 2013b). The centralization index mixes data that illustrate fragmentation or concentration of unions with data on the division of authority in the labor union between confederations, affiliated unions, and local or workplace branches. We use the following formula in order to calculate the index score:

𝐶𝑒𝑛𝑡 = �

+

𝑈𝑁𝑎𝑢𝑡ℎ𝑜𝑟𝑖𝑡𝑦 = Authority of union over their local workplace branches and

representatives. Consists of a score between 0 and 1, respectively low and high authority.

𝐻𝑎𝑓𝑓 = Membership concentration at industry level, within confederations.

An index between 0 and 1. This index is given by ∑ (𝑝𝑛_{𝑖 𝑖}2), where p is the share of total membership organized by the ith affiliate and n is the total number of affiliates.

𝐷𝐸𝑀𝐼 = Internal demarcations within union confederations. Score of 1, 1.5, or 2. 1 represents no cleavages (single jurisdiction, industry unions or enterprise unions). 1.5, moderate (skill-related or occupational) cleavages, associated with limited competition. 2 represents, sharp (occupational, skill-related, or organizational) cleavages, hence related to competition and conflict (within one company more than one union).

𝐶𝐹𝑎𝑢𝑡ℎ𝑜𝑟𝑖𝑡𝑦 = Authority of confederations over its affiliates. Consists of a score

between 0 and 1, respectively low and high authority.

𝐻𝑐𝑓 = Membership concentration at central or confederal level. An index

membership organized by the ith confederation and n is the total number of confederations.

𝐷𝐸𝑀𝐸 = External demarcations between union confederations. Score of 1, 1.5, or 2. 1 represents, no cleavages, hence unified confederation. 1.5, moderate (occupational, regional, linguistic, religious) cleavages, associated with limited competition. 2 represents, sharp (political, ideological, organizational) cleavages, hence related to competition and conflict.

Although, centralization and coordination are related, they must be distinguished from each other. Centralization is predominantly a matter of the level at which contracts are signed. Coordination is a matter of how wage negotiations, which are conducted by bargaining institutions, are coordinated across the country (Traxler, et al., 2001). More precisely, national coordination may take place regardless of the degree of centralization. For example, centralization may be extremely decentralized (firm-level negotiations), at the same time the negotiations might be coordinated at a national level. Although this is highly unlikely, since firm-level negotiations are hard to coordinate at a national level, this is theoretically a possible outcome.

In his influential paper, Soskice argued that the focus should be on bargaining coordination rather than centralization (Soskice, 1990). He states that centralization is only a part, although on important one, of the more valuable variable coordination. Apart from centralization, coordination consists of; state-imposed centralization, informal centralization, and pattern setting. State-imposed centralization means that governments set rules and laws in order to make negotiations binding for certain parties. This has happened for several countries (Denmark, Belgium, Canada, the Netherlands, and even United States) in the past. Informal centralization means the guidance of by employer organizations and high-level unions of lower level bargaining. For example, the guidance of national unions and employer organizations of industry level bargaining. We call it pattern setting, when smaller organizations or firms follow a bigger leader. According to Kenworthy (2001a) coordination should be seen as mainly a behavioral concept.

by peak confederations, respectively. His goal was to create a variable that focused on the degree of coordination rather than the type. The measure of Soskice seemed to be an appropriate one, however, according to Kenworthy (2001b) this variable may suffer from substantial errors in measurement. Hence he created a set of scores that represented wage setting that are “likely to generate more or less coordination” (Kenworthy, 2001b). Again, the data was only available for a small period of time. In order to make this data available for a larger time frame and a larger set of countries, Visser (2013b) developed a variable based on the Kenworthy-variable. “There are a few subtle differences in the wage coordination scale used here and the one used by Kenworthy and for some years and countries my scores and his differ” (Visser, 2013b).

The second variable of interest is the coordinating categorization of Traxler, Blaschke, and Kittel (2001). They did not appoint a score to certain bargaining structures, they only categorized them in six different categories. The focus was on the coordination activities of the larger employer organizations and/or unions, instead of the degree of coordination achieved. They coded the categories rather than scoring them, leaving aside whether the aim of coordination was actually achieved. However, Visser (2013b) did put them in a sequential order; appointing score 1 to uncoordinated bargaining, up to score 6 which represents state-imposed bargaining, making them more suitable for our empirical research. Furthermore, Visser (2013b) expanded the time frame up to 2011, and applied the codes to a wider range of countries.

Table 1 below shows the correlation between the two centralization variables (Iversen (1999) and Visser (2013a)) and the two coordination variables (Kenworthy (2001b) and Traxler, Blaschke, and Kittel (1998)). Although all correlations are positive, they are not as high as we might expect. The reason for this might be the more subjective nature of the coordination variable, or just the differences in the way of appointing scores, ranks, and indexes.

Table 1: Correlations centralization/coordination 1980-2010 1. 2. 3. 1. Iversen 2. Visser 0.51 3. Kenworthy 0.59 0.66 4. Traxler et al. 0.32 0.79 0.62 Source: Visser (2013b)

As stated before, in our opinion the Iversen-index, which is adjusted and made available for a larger time frame and set of countries by Visser (2013b), is best suitable as a measure of union power. First of all, since it is better grounded op objective data than the coordination variable. Second, it is an index. Hence, we do not have problem with countries ‘floating’ in between different scores.

4.2 Union density

As mentioned in earlier chapters, union density is the proportion of workers that are member at a labor union. Figure 9 below shows the density rates of Belgium (BE), the United Kingdom (UK) and fifteen OECD-countries (OECD-15) over the last 50 years. The figure shows a density rate which has an increasing trend up to the 1980’s. However, from that moment on the different density rate fluctuates increasingly. Belgian density rates are relatively stable, the union density of the UK experiences a sharp decline, and all fifteen OECD-countries combined show a steady decline. The fact that the Belgian density rate did not decline find its origin in the Ghent system. When the unemployment insurance systems were introduced, some people argued that the government should manage this insurance. Others argued that labor unions should take charge of managing and distributing of the unemployment insurance. This latter is called the Ghent system, since Ghent was the first city where such an agreement was signed. Today, Belgium holds a semi-Ghent system, where the unions play an active role (Dimick, 2002). Denmark, Finland and Sweden still holds the Ghent system, therefore their union membership rates did not decline over the last fifty year.

Figure 9: Union desnity rates of, OECD-15, UK and Belgium 24 28 32 36 40 44 48 52 56 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 OECD UK BE U n io n d e n s it y i n % Source: Visser (2013b)

After a period of increasing density rates, the rate in the UK sharply declined in the 1980’s. The strong decline is strongly related to the Thatcher-government that ruled during the 1980’s. As mentioned before, Thatcher was not fond of strong labor unions and aimed at breaking union power.

Many other density rates declined as well. This was moving simultaneously with the aging of the population. The average age of a union member was increasing, while at the same time the average age of a worker was declining. In other words, union were not able to attract new members (Boeri & van Ours, 2013). One of the reasons for this was the so called ‘free-rider’ problem. Many agreements did not hold only for union-members, but all employees of a certain firm or sector were included in the agreements. Hence, the additional value of becoming a union member was decreasing, unless for the countries that applied the Ghent system. On the other hand, membership dues were going down, which should encourage membership. However, simultaneously with the dues the total amount of services provided by the union decreased as well, which resulted in declining membership rates.

In the following sections we will discuss the control variables; change in inflation, trade, and government orientation.

4.3 Change in Inflation

approach of Belot and Van Ours (2001) and apply the change of inflation as a control variable. Theory suggests that an unexpected change in inflation is negatively related to unemployment. Since a sudden rise in prices combined with the rigidity of wages, leads to relatively cheap labor. Relatively cheap labor will result in a higher demand for labor followed by a decrease in unemployment (Belot & van Ours, 2001). For example, a rise of the inflation rate from two to three percent will have a declining effect on unemployment. If the following year the expected inflation will remain at three percent, the unemployment rate will not decline again, because this inflation rate is expected. For this reason we will control for the change in inflation, rather than the inflation itself.

Figure 10: Inflation rate in the USA and OECD-15

-2 0 2 4 6 8 10 12 14 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 USA OECD In fl a ti o n r a te i n % Source: OECD (2014)

Figure 10 show the inflation rate from 1960 until 2010. The lines represent the USA and the fifteen OECD countries, and follow more or less the same trend. This suggests that inflation shocks are relatively global. Furthermore, they follow the same trend as unemployment shown in figure 1. This contradicts negative relationship predicted by the economic theory. However, the peaks were the result of the Oil-crises of 1973 and 1979. Due to a sudden increase of the price of oil the all prices increased because production costs were rising.

4.4 Trade

unemployment and higher wages. This brings us to us to the second reason. Increasing trade means more competitions for domestic firms. Trade also enhances firms’ ability to move to foreign low-wage-countries. The abilities to move abroad motivated firms to demand lower labor costs in order to stay competitive. In other words, firms demand labor costs constraints. This again, leads to lower labor costs which result in higher demand for labor and lower unemployment. Furthermore, Feenstra (1997) found that the free trade agreement of Canada and the United States did not lead to an increase of unemployment in Canada, as expected by Canadian politicians. Hence, we strongly expect trade, as a percentage of GDP, to be negatively related to unemployment.

4.5 Government orientation

It is believed that a left-wing government has a higher priority to keep the unemployment low. A right-wing government has a higher priority towards economic growth in general and pays less attention to unemployment figures. Furthermore, there is support for that a leftish government is related to monetary growth (Chappell & Keech, 1988). Monetary growth will result in, ceteris paribus, an increase in inflation. The change in inflation will again result in a decrease in unemployment.

Figure 11: Average Partisan Score OECD-15

1.4 1.6 1.8 2.0 2.2 2.4 2.6 1975 1980 1985 1990 1995 2000 2005 2010 OECD p a rt is a n s c o re Source: Worldbank (2012)

Figure 11 shows the average partisan score of the fifteen OECD countries. Countries are appointed to a certain score, 1 for a left-wing government, 2 for a center government, and 3 for a right-wing government. The scores represent economic policy only.

4.6 Correlation

Table 2 below shows all the correlations of the fifteen OECD-countries between 1980 and 2010. Unemployment is the unemployment rate. Centralization represents the Iversen-index. Trade is the sum of imports and exports as a proportion of gross domestic product (GDP). Government represents a score of 1, 2, or 3 which is appointed to a government of a country.

Table 2: Correlation of the variables

Variable Source 1. 2. 3. 4. 5.

1. Unemployment OECD(2014)

2. Centralization Visser (2013b) _-0.26

3. Union Density OECD (2014) _{0.06 0.36}

4. ∆Inflation OECD (2014) _{-0.20 0.07 -0.05}

5. Trade OECD (2014) _{0.02 0.36 0.18 0.15}

6. Government Worldbank (2012) 0.16 -0.17 -0.13 -0.04 0.10

The table shows the correlations with their expected signs as described in previous literature, except for trade. Trade is positively correlated to unemployment, this contradicts the literature. However, the correlation is close to zero, therefore, we will not pay any attention to this. Furthermore, the unemployment rate is negatively correlated to centralization as well as change in inflation, and positively correlated to union density, and government orientation. Centralization is positively correlated to union density, which suggests that a higher union density results in higher centralization. This is in line with previous literature as well as with our expectations. Correlations between explanatory variables are not extremely high, this reduces changes in multi-collinearity. Furthermore, appendix B illustrates the ‘variation-inflation-factors’ (VIF) test. This test illustrates that none of the independent variables hold a higher VIF-value than five2.

2 _{The rule of thumb concerning the VIF-test is. If the VIF-value is higher than 5, the explanatory variable}

exhibits severe multi-correlations (O'Brien, 2007).

5. Model

5.1 Regression Model

In order to test whether centralization is linear, non-linear, or not related to unemployment we make use of multiple regression models. The regression models are based on the time-range between 1981 and 2010. This time range is chosen for two reasons. First of all, by starting at 1981 we do not have to include an explanatory variable to explain the rise between 1960 and 1980. Second, data is more widely available in the period after 1980 than before. Apart from the control variables mentioned in the previous section, we added the lagged unemployment as an explanatory variable. This is in line with research of Iversen (1999), Kenworthy (2002), and Nickell et al. (2005). The reason for this is that the lagged unemployment enables us to control for endogenous persistence of unemployment (Nickell, et al., 2005), because unemployment tends to be sticky over time in developed countries (Kenworthy, 2002).

5.1.1 Linear relationship

In order to detect a linear relationship, we have developed the following linear base model. 𝑈𝑖,𝑡 = 𝛽1+ 𝛽2𝐶𝑖,𝑡+ 𝛽3𝑈𝐷𝑖,𝑡+ 𝛽4𝑋𝑖,𝑡+ 𝛽5𝑈𝑖,𝑡−1+ 𝜀𝑖,𝑡 (5)

Where 𝑈_𝑖,𝑡 = Unemployment rate 𝐶𝑖,𝑡 = Centralization

𝑈𝐷𝑖,𝑡 = Union Density

𝑋𝑖,𝑡 = The control variables

𝑈𝑖,𝑡−1 = Lagged unemployment rate

𝜀𝑖,𝑡 = Error term

𝑋𝑖,𝑡 Represents the control variables and consists of change in inflation, trade as a share of

GDP, and economic government orientation (left, center, or right). The term i represent one of the fifteen OECD countries, and t is the annual term. Centralization will be our variable of interest, since in our analysis this will represent union power. We will also pay special attention to union density. Since some consider this to be an indicator of union power.

5.1.2 Nonlinear relationship

Besides the linear effect of centralization, we want to test for the hump-shaped analysis of Calmfors and Driffill (1988). In order to do so, we will modify the model in two different ways. First of all, we will test whether the slope depends on the centralization-variable itself. This means that the model should be extended with a squared centralization variable. The extended equation is shown below in equation (6):

𝑈𝑖,𝑡 = 𝛽1+ 𝛽2𝐶𝑖,𝑡+ 𝛽3𝐶𝑖,𝑡2 + 𝛽4𝑈𝐷𝑖,𝑡+ 𝛽5𝑋𝑖,𝑡+ 𝛽6𝑈𝑖,𝑡−1+ 𝜀𝑖,𝑡 (6)

A hump-shaped relationship between unemployment and centralization is only possible when the slope depends on centralization, since this means that the slope is not a constant. In other words, the first derivative with respect to centralization is an equation depending on centralization. This holds that both, 𝛽₂ and 𝛽₃ of equation (6), have to be significant. However, this does not conclude whether the relationship is hump-shaped or U-shaped. Hence 𝛽₂ has to be significant and negative, and 𝛽₃ should be significant and positive.

Second, assuming a hump-shaped relationship implies that the slope evolves from positive to negative along the line. Therefore, this means that from decentralized up to intermediate-centralized 𝛽₂ should be larger than zero, followed by a negative value of 𝛽₂ between intermediate-centralized and centralized. We can make use of dummy-variables to split the centralization into two parts. The model including the dummy-variables is shown below in equation (7):

𝑈𝑖,𝑡 = 𝛽1+ 𝛽2𝐶𝑖,𝑡+ 𝛽3𝐷𝐶𝑖,𝑡+ 𝛽4𝐶𝑖,𝑡∗ 𝐷𝐶𝑖,𝑡+ 𝛽5𝑈𝐷𝑖,𝑡 + 𝛽6𝑋𝑖,𝑡+ 𝛽7𝑈𝑖,𝑡−1+ 𝜀𝑖,𝑡 (7)

Where 𝐷𝐶_𝑖,𝑡 is a dummy variable with value, 1 for 𝐶_𝑖,𝑡 > ‘intermediate-centralization’, and, 0 for 𝐶_𝑖,𝑡 < ‘intermediate-centralization’. If the relationship of unemployment and centralization is explained by a hump-shaped form, 𝛽₂ should be significant and positive.

We should keep in mind that we do not know the possible location of ‘intermediate-centralization’ on the line (where the slope is equal to zero). To solve this problem, we will test the regression for three different index-scores of ‘intermediate-centralization’. The first value will be the average of all index-scores of centralization. Hence, for this single case we assume that on average, all countries together are intermediate-centralized. For the second case, we will test for the mean of the index-scores of intermediate-centralized countries,

according to Heijdra (2009) (the Netherlands and Belgium). For the third and last case we will test for the exact middle of the index, score 0.5.

5.1.3 Interaction model

Many economist and socialists argue that bargaining institutions mainly influence unemployment when they interact with each other (Nickell, et al., 2005; Kenworthy, 2002). Moreover, in the literature review we elaborate on the different views of union power. In the next model we test the relationship of unemployment and the interaction of centralization and union density (equation 8).

𝑈𝑖,𝑡 = 𝛽1+ 𝛽2𝐶𝑖,𝑡+ 𝛽3𝑈𝐷𝑖,𝑡+ 𝛽4𝐶𝑖,𝑡∗ 𝑈𝐷𝑖,𝑡 + 𝛽5𝑋𝑖,𝑡+ 𝛽6𝑈𝑖,𝑡−1+ 𝜀𝑖,𝑡 (8) 5.2 Stationarity

A possible problem is non-stationarity of the dependent or independent variables. Although we do not have any presumption, non-stationary variables might lead to biased results. In order to detect whether the variables are stationary we perform the ‘Augmented Dickey Fuller’ (ADF) test. In this test the null-hypothesis is; the variable has a unit root. The results are shown in table 3, in which 𝑈_𝑖,𝑡, 𝐶_𝑖,𝑡, and, 𝑈𝐷_𝑖,𝑡 represent the usual variables and ∆𝐼𝑁𝐹_𝑖,𝑡 represent change in inflation, 𝑇𝑅_𝑖,𝑡 trade as a proportion of GDP, and 𝐺𝑂𝑉_𝑖,𝑡 government orientation.

Table 3: Unit root test

Variable 𝑈_𝑖,𝑡 𝐶_𝑖,𝑡 𝑈𝐷_𝑖,𝑡 ∆𝐼𝑁𝐹_𝑖,𝑡 𝑇𝑅_𝑖,𝑡 𝐺𝑂𝑉_𝑖,𝑡

p-value 0.0001 0.0297 0 0 0 0.0025

Table 3 shows the p-values, which are all significant at a five percent significance basis. As a result, we reject the null hypothesis of a unit root (non-stationarity), and accept the alternative hypothesis of stationarity of all variables. Details of the test are available in appendix C.

5.3 Fixed or random effects

The next step of developing our regression model, is to determine whether we should include country-dummy variables, annual-dummy variables, or both. We expect the necessity to use cross-section, as well as period dummy variables. The reason for this is that country specific shocks in a certain period might not be explained by one of the explanatory variables.

Furthermore, the ‘steady-state’ unemployment rate might be different for every country and not be explained by one of the independent variables. To detect which one we have to include, we perform a Hausman test. A Hausman test tests the null-hypothesis of using random effects, against, the alternative hypothesis of using fixed effects. Table 4 shows the p-values of the Hausman tests.

Table 4: Hausman test.

P-value Cross section-random 0.0000

Period-random 0.0012

Table 4 shows us that we should reject the null-hypothesis of including random effects, and accept the alternative hypothesis of using fixed effect. This is in line with our expectations and with previous research (Bertola, et al., 2002; Belot & van Ours, 2001; Nickell, et al., 2005). In order to conclude whether the use of fixed effects as well as random effects is appropriate we have conducted a redundant fixed effects test (result in appendix D). This test enables us to conclude, at a 1% significance basis, that period fixed effects, as well as cross-section fixed effect are suitable.

5.4 Expected signs

To conclude, we expect the coefficients to be of the following form; from the standard base model (equation (5)), we expect centralization to be negatively related to unemployment, meaning that 𝛽₂ is expected to be smaller than 0. This is primarily based on previous literature. The majority of the existing literature found that high degree centralization is associated with a low unemployment level. Mainly because highly centralized organizations tend to internalize the externalities, which are thoroughly discussed in the literature review. Furthermore, we have found no recent papers that confirm the hump-shaped relationship as described by Calmfors and Driffill (1988). As a result, we expect insignificant coefficients, 𝛽3 and 𝛽2, of equations (6) and (7) respectively.

obligated to claim higher wages. As a result, labor costs go up followed by an increased unemployment rate. Although this effect might have been mitigated over the last decades, we expect to find a positive effect for our time frame 1981 until 2010. Therefore, we believe that 𝛽3 of equation (5) will be positive.

Furthermore, we have no reason to belief that the effect of the control variables on unemployment have changed in the last decade. Consequently, we expect the coefficients of change in Inflation and Trade as a share of GDP to be negative, and government orientation to be positive.

6. Results

6.1 Linear and non-linear base model

The results of equation (5) and (6) combined with the expected signs are shown below in table 5.

Table 5: Coefficient values of equation (5) and (6). Sample 1981-2010 – fifteen OECD-countries

(1)Expected sign (2)Linear model (3)Non-linear model

Centralization (-) -4.14*** -1.54 (-3.20) (-0.43) Centralization^2 … … -2.813 … (-0.79) Union Density (+) 0.04*** 0.04*** (3.11) (3.20) ∆Infation (-) -0.18*** -0.18*** (-5.30) (-5.31) Trade (-) -0.18*** -0.18*** (-4.95) (-4.80) Government (+) 0.02 0.01 (0.38) (0.22) Lagged unemployment (+) 0.88*** 0.88*** (44.42) (43.85)

Country fixed effects Annual fixed effects

    𝑅2 _{0.949 0.949} Total observations 440 440

Note: Numbers in parenthesis indicate the t-statistics. *, **, and *** represent significance from zero at a 10, 5, and 1 percent, respectively.

The regression results are in majority in line with the expectations. First, we analyze the linear model in the second column. Note that the signs of the coefficients are equal to the expected signs in column 1. More noticeable is that, apart from government orientation, all variables are highly significant. The coefficients are fairly straightforward. First of all, table 5 shows a highly significant negative relationship between unemployment and centralization. The centralization index takes all values between 0 and 1. Therefore, a shift from decentralized bargaining (e.g. index value 0.1) to intermediate-centralized bargaining (e.g. index value 0.5), results in an unemployment decrease of 1.6 percent. Please keep in mind that such a shift normally does not happen in one year. For example, the largest centralization change, of 15 countries in 30 years, took place in Japan in 1988. This shift was only 0.1 on the Iversen-index. Nonetheless, this should have led to an unemployment rate decrease of 0.4 percent. Furthermore, union density exhibits the opposite effect on unemployment. A decrease of 10 percent in union density would decrease unemployment by 0.4 percent.

The non-linear model in column 3 has different coefficients. The linear effect becomes even insignificant, followed by a highly insignificant quadratic effect of centralization. We highly suspect no hump-shaped form. However, another analysis will follow in which we try to find a positive effect of centralization on unemployment for relative decentralized countries. Moreover, we notice an extreme high value of 𝑅2. We should not pay too much attention to this figure, since this is mainly caused by the lagged dependent variable. We can see this strong relationship at the coefficient of lagged unemployment. If we drop lagged unemployment we find higher coefficients of nearly every explanatory variable (results in appendix E). Hence, centralized countries seem to be associated with low unemployment. However, the low unemployment is primarily the result of superior performance in earlier years. This is caused by unemployment persistence (Kenworthy, 2002), as discussed in stylized fact 2 of section 2.1.

6.2 Alternative non-linear model

test the model for three scores. The index contains a range from 0 to 1. The score that we test first is 0.5 (this is the middle of the index). Second, in previous literature the Netherlands and Belgium are used as an example for intermediate bargaining centralization. Therefore, we take the average of those two countries as a second intermediate value, which seems to be 0.5 as well. Third, we take the average of all countries, this is 0.4. As a result, we end up with two models, table 6 shows the outcome.

Table 6: Alternative non-linear model according to equation (7)

(1)needed sign

(2)dummy>0.5 (3)dummy>0.4 (4)dummy>0.4 (reduced) Centralization (+) -3.97** -2.60* -3.06** (-2.52) (-1.74) (-2.29) Centralization*dummy (-) -0.80 -1.33** -1.27** (-0.18) (-2.04) (-2.04) Union Density (+) 0.04*** 0.03*** 0.03*** (3.05) (2.74) (2.84) ∆Infation (-) -0.18*** -0.18*** -0.18*** (-5.29) (-5.28) (-5.57) Trade (-) -0.18*** -0.16*** -0.16*** (-4.81) (-4.49) (-4.50) Government (-) 0.02 0.02 … (0.37) (0.34) Lagged unemployment (+) 0.88*** 0.88*** 0.88*** (44.31) (44.65) (47.80)

Country fixed effects Annual fixed effects

      R2 _{0. 949 0. 949 0. 955} Total observations 440 440 440

Note: Numbers in parenthesis indicate the t-statistics. *, **, and *** represent significance from zero at a 10, 5, and 1 percent, respectively.

The regression results in table 6 are clear. The first column refers to the sign needed for a hump-shaped relationship. Looking at column 2 and 3, we can conclude that there is no evidence for this relationship. First of all, the centralization-coefficient has to positive to contain evidence for a hump-shaped relationship. As table 6 illustrates, this is not the case for neither of the two centralization-coefficients. Second, the centralization*dummy-coefficient has to be negative and significant. This is not the case for the model presented in column 2. However in the model of column 3, we attained a significant

coefficient. In other words, there is a significance difference in the slope for the first and second part of the line. As the centralization*dummy-coefficients in column 3 and 4 makes clear, the negative influence on unemployment is stronger for index values larger than 0.4. As a result, we can conclude that we have found no evidence that the analysis of Calmfors and Driffill (1988) still holds. However, we do find a non-linear relationship, where the slope becomes steeper after the index value of 0.4 (column 3 and 4).

6.3 Interaction model

As stated before, economists and sociologists often disagree on the best indicators of union power. In this section we use the interaction term which combines the characteristics of union density and centralization according to equation (8). Table 7 shows the results from the interaction model.

Table 7: Interaction of centralization and union density according to equation (8)

(1) Interaction model (full) (2) Interaction model (reduced)

Centralization -0.92 … (-0.45) Union Density 0.07*** 0.08*** (3.05) (5.41) Centralization* Union Density -0.06** -0.07*** (-2.03) (-4.06) ∆Infation -0.17*** -0.17*** (-5.04) (-5.33) Trade -0.20*** -0.20*** (-5.35) (-5.56) Government 0.00 … (0.05) Lagged unemployment 0.87*** 0.87*** (43.06) (46.35)

Country fixed effects Annual fixed effects

    𝑅2 _{0. 949 0. 955} Total observations 440 440

Note: Numbers in parenthesis indicate the t-statistics. *, **, and *** represent significance from zero at a 10, 5, and 1 percent, respectively.

of union density strongly depends on the degree of centralization and vice versa. For example, an increase of union density of 10 percent at a 0.2 centralization index value is associated with an increase of unemployment of 0.56 percent. However, the same rise of union density at an index value of 0.8 percent would increase unemployment by only 0.22 percent. To conclude, an economy can significantly mitigate the positive effect of union density on unemployment, by stimulating bargaining at a high and centralized level.

6.4 Robustness checks

In order to check the robustness of our regression, we perform a number of robustness checks. First of all, we split the sample in two time periods and see whether the conclusions still hold. Second, as mentioned before, the centralization index of Iversen (1999) is not the only measure of centralization. In order to check whether our conclusions are only applicable for this specific measure, we substitute the Iversen index for the measure of Visser (2013a). Moreover, since centralization and coordination are relatively correlated, we also test for two measures of coordination (Traxler, et al., 2001; Kenworthy, 2001b). Finally, we perform a sensitivity check, to test whether our regression are sensitive when we include another explanatory variable, tertiary school enrollment.

6.4.1 Time periods

Table 8: Linear model according to equation (5), different periods of time.

(1) 1981-2010 (2) l981-1995 (3) 1996-2010 Centralization -4.14*** -6.14** -6.23*** (-3.20) (-2.33) (-2.82) Union Density 0.04*** 0.13*** 0.04 (3.11) (5.40) (1.32) ∆Infation -0.18*** -0.11** -0.26*** (-5.3) (-2.45) (-5.02) Trade -0.18*** -0.48*** -0.13*** (-4.95) (-3.31) (-2.85) Government 0.02 0.04 -0.05 (0.38) (0.48) (-0.80) Lagged unemployment 0.88*** 0.82*** 0.88*** (44.42) (22.12) (23.79)

Country fixed effects Annual fixed effects

  

  

𝑅2 _0.949 0.956 0.949

Total observations 440 225 215

Note: Numbers in parenthesis indicate the t-statistics. *, **, and *** represent significance from zero at a 10, 5, and 1 percent, respectively.

Table 8 illustrates the regression outcome when we split the time frame. In general we can confirm that centralization as well as union density has a significant effect on unemployment for each period of time. Nonetheless, when we split up the sample, the model is slightly losing its strong significance. Centralization is still significant, however, not as strong as in the 30 years model. Furthermore, union density is not significantly in the time period of column 3. The interaction term only shows significance in the reduced form in the time period 1981-1996 (not illustrated in the table). It is difficult to draw any strong conclusion from this test. The slightly less significant effects could be the result of a smaller sample. However, it could also be a changing trend. It is possible that union-boards ‘internalized’ the finding of previous studies, and consequently mitigated the centralization effect as well as the union density effects.

6.4.2 Centralization/coordination alternatives

of centralization and coordination. This might give answer to the question whether the effect is due to the measure of Iversen, or due to the concept of centralization/coordination. In Table 9 we illustrate the results of four different regressions, respectively the Iversen Index (1999), Visser’s level score (2013a), coordination score of Kenworthy (2002), and the coordination measure of Traxler et al. (2001) (TBK).

Table 9: Models on different measure of centralization/coordination, according to equation (5)

(1) Iversen (2) Visser (3) Kenworthy (3) TBK Centralization/ Coordination -4.14*** -.24*** -0.16*** -0.11*** (-3.20) (-4.40) (-2.34) (3.01) Union Density 0.04*** 0.04*** 0.04*** 0.04*** (3.11) (3.15) (3.11) (3.01) ∆Infation -0.18*** -0.19*** -0.19*** -0.19*** (-5.30) (-5.78) (-5.55) (-5.66) Trade -0.18*** -0.14*** -0.15*** -0.15*** (-4.95) (-4.12) (-4.20) (-4.24) Government 0.02 0.04 0.04 0.04 (0.38) (0.92) (0.78) (0.81) Lagged unemployment 0.88*** 0.88*** 0.88*** 0.88*** (44.42) (45.69) (44.89) (45.28)

Country fixed effects Annual fixed effects

   

   

𝑅2 _0.949 0.950 0.948 0.949

Total observations 440 449 449 449

Note: Numbers in parenthesis indicate the t-statistics. *, **, and *** represent significance from zero at a 10, 5, and 1 percent, respectively.

Table 9 shows the same sign for all measures of centralization and coordination. For that reason, the concept centralization/coordination is negatively related to unemployment. Furthermore, we have not seen any significant changes in any of the other explanatory variables. In other words, the impact of one of the centralization/coordination variables is more or less the same.

The results on the first non-linear of equation (6) model are in line with the results for the Iversen-index (not illustrated in the table). Furthermore, the interaction term is insignificant for all three alternative measures. If we omit all insignificant coefficients, we do find a significant coefficient (not illustrated in the table). However, the effects are a lot smaller than in the case of the Iversen-index.

6.4.3 Sensitivity analysis

As an extra control variable we use school enrollment which is negatively related to unemployment (Mincer, 1991). We use tertiary school enrollment as a variable to test the sensitivity of our most significant models. The Worldbank (2014) defines tertiary education, as the share of people who finished secondary education and enroll for tertiary education. We have lagged the education variable for three years. Because it normally takes time to graduate from tertiary education. In order to determine how many years education should be lagged, we focused on the coefficient and the significance. As a result, we have chosen three years because it gives the highest coefficient as well as the highest significance.

Table 10 shows the sensitivity analysis for three models. The first is the base model, where we found a linear relationship between centralization as well as union density with the sign we expected. In column 2 we test the sensitivity for the alternative non-linear model. In this model, we did not find the hump-shape-relationship. However, we did find an increasing slope for value of the Iversen index larger than 0.4. In the last column we test whether the interaction coefficient still holds if we add a variable.