THEORETICAL APPROACHES TO WAGE DISPERSION AND THE ROLE OF INSTITUTIONS

18.4.1 The Wage Inequality Debate 1980-2000 and the Role of LMIs

We start this paragraph with a brief introduction of the evolution of the literature on wage dispersion up to 2000.

The literature in which the contemporary discussion on the dispersion of wages is rooted took off seriously in the course of the 1980s, kicked off with a detailed picture of changing male and female inequalities in the United States by Peter Henle and Ryscavage (198 0).^[168] This literature focused initially on the factual question about whether inequality had increased or not, and it took some time before the factual doubts about that growth dissipated,^[169] though factual questions have remained on the agenda throughout. It did not take long before the “why?” question started being asked and answers were sought in many directions—more often than not in different directions at the same time incurring a risk of ad hocery. From the start, some of these routes led to what in due course have become known as LMIs. For example, Plotnick (1982) attrib­uted the (slow) increase in the variance oflog (male) annual earnings—“earned income” as it was usually called—between 1958 and 1977 entirely to the differential effects ofthe level of unionization, the dispersion in weeks worked, the age distribution of workers, and the inequality of education. Dooley and Gottschalk (1984) looked for a demographic explanation (viz. cohort size and the baby boom), and Dooley and Gottschalk (1985) added—on the factual side—the insight that real earnings of men’s earnings below a low-pay threshold were lagging behind.^[170] Bluestone and Harrison (1982) launched the thesis of deindustrialization and expansion of low-wage employment, which in a later book (Bluestone and Harrison, 1988) turned into the famous “Great U-turn” of growing inequality.

Thus, education, institutions, demographics, and the composition of the economy made an appearance in the literature almost from the start. From the early 1990s, inter­national trade and especially the competition with low-wage countries were added as other explanations (e.g., Wood, 1995). However, this is outside the focus on LMIs of the current chapter.^[172] Though the interest in demographics may have waned (apart from the fact that gender has become a staple ingredient), the attentiveness to educational dif­ferentials, institutions, and the composition ofthe economy—industries and sectors at the time, occupations and tasks nowadays—has grown into a vast literature over the 1990s. The interest in education focused on the demand for skills in the economy on the one hand and their supply by the labor force on the other hand (e.g., Juhn et al., 1993). This has ushered in the thesis of “skill-biased technological change” (SBTC for short) (Bound and Johnson, 1989, 1992; Levy and Murnane, 1992) as the driving force behind the demand for skills. For some time this became the canonical model for explaining growing wage inequality. Evidently, the composition of the economy is not unrelated to this, if only because the skill structure differs between industries.

Technological change and economic composition together have become the sup­porting vector for the attention paid in the literature to supply and demand—or “market forces”—as explanatory factors for differences and changes in wage inequality. In their first overview of the literature, Levy and Murnane (1992) recommend future research to “get inside the black box ofthe firm” (p. 1374) and pursue the hedonic theory of labor demand that views a worker as possessing a fixed package of separate productive abilities and the wage as the sum of payments to them.

Parallel to the market view, the institution-focused approach thrives. Fortin and Lemieux (1997) convincingly demonstrate its relevance basing themselves on both cur­rent American experience and effects of the Great Depression. There are some empir­ically obvious candidates for explaining wage inequality—the extent of union membership and the minimum wage given the clear declines in both the United States and the significant international differences, including the (de)centralized nature of wage bargaining. Freeman (1991) finds an important but not overwhelming role for unioni­zation in relation to increasing inequality; by contrast, DiNardo and Lemieux (1997) explain two-thirds of the American-Canadian difference in male wage inequality growth from the faster decline of American unions; Blackburn et al. (1990) attribute only a small role to the declining minimum wage. DiNardo et al. (1996) find a substantial contribu­tion to increasing wage inequality for the declining level of the minimum wage between 1879 and 1988. Though Levy and Murnane (1992) mention these institutions in their overview, they barely touch upon institutions—and then for the United States only— and apparently see no clear research agenda there. During the rest of the 1990s, however, the interest in institutions widens far beyond the above-mentioned. This receives great stimulus from international comparisons where differences in institutions can more easily get into the limelight. Working under Different Rules (Freeman, 1994), Differences and Changes in Wage Structures (Freeman and Katz, 1995), and Blau and Kahn (1996) bring together important empirical studies of a variety of countries. On a somewhat different tack from wage inequality, Freeman (1988) and Richard Layard and Stephen Nickell (1991) contribute on LMIs and macroeconomic performance in OECD countries. At the same time, Card and Krueger (1995a) dispel the consensus regarding the negative employment effects of the American minimum wage.

18.4.2 Defining and Analyzing LMIs

Given the importance of the subject for this chapter we will pay attention to the issue raised by the treatment of institutions before turning to the discussion of recent devel­opments in the inequality literature in the next two sections. We distinguish between conceptualizing (method) and analyzing institutions and consider this at a general level in the present subsection, which boils down to the question of how economic analysis accounts for the existence of institutions. More specifically, their role in wage-inequality analysis is discussed in Section 18.4.5.

18.4.2.1 Method

First, the definition of institutions, and of LMIs in particular, seems in need of more pre­cision, generically as well as in specific cases.^[173] Too often in the literature they seem to be considered too obvious—and therewith perhaps too difficult—to warrant explicit defi­nition. As Nickell and Layard (1999) say, “It is difficult to define precisely what we mean by labour market institutions, so we simply provide a list of those features of the labour market which we shall consider.” In the words of Freeman (2007), “While economists do not have a single tight definition of an institution, per Justice Potter’s famous statement about pornography, they know institutions when they see them, and they see them everywhere.” Those who actually venture a definition come up with very broad ones, such as “A labour market institution is a system of laws, norms or conventions resulting from a collective choice and providing constraints or incentives that alter individual choices over labour and pay” (Boeri and Van Ours, 2008, 3), that may be begging impor­tant questions—in this case: Is the alteration of choice a practical matter or a theoretical issue? If the latter, this poses the risk that by definition any institution will imply a devi­ation from the theoretical ideal and also that institutions will be conceptually difficult to endogenize, meaning that an institution might endorse changes in behavior that have already taken place instead of causing such changes.

Institutions started their career in the inequality discussion as union density and the (de)centralized nature of wage bargaining. These are factors that can obviously influence individual wages, but they are neither law, nor regulation or rule, and seem closer to physical organization. The minimum wage, another institution also present from the start, is (often) in the law though. Not only in the inequality debate but more broadly, institutions have come to encompass a wide array of factors, and there seems to be no clear defining limit as to what can qualify as one or not. This makes institutions not only difficult to delineate but also tends to lend them a fundamentally ad hoc character. The selection of LMIs does not necessarily result from a systematic scrutiny but seems to reflect trial and error based on constrained and sometimes even biased knowledge. How otherwise to explain the immensely strong focus on unions while employer asso­ciations hardly figure, in spite of the fact that they are equally involved in collective bar­gaining and that “employer density”—the percentage of workers in a sector who are employed by the negotiating firms—and not union density may actually provide the basis for declaring a collective agreement generally binding.^{[174] [175] The attention paid first to union density and only much later, when this appears to fail as an explanation, to the coverage of collective bargaining, is another case in point.[176] Last but not least, it is difficult to under­stand, given the highly central role of educational differentials in the wage-inequality lit­erature, that the educational system figures so little as an institution in this literature.[177]}

In addition, in the literature, LMIs can be viewed as specific factors with an actual origin and presence in the labor market and only there (e.g., wage bargaining), but equally they may be factors that affect the labor market from outside that market (e.g., income taxation or the tax wedge). The latter definition, as a factor affecting the labor market irrespective of its origin, opens the door to a myriad of institutions and, unsurprisingly, many other factors have been added in the 30-year course of the debate. For example, Nickell and Layard (1999, 3047) include home ownership in their list of labor market “institutions” (their quotation marks), inspired by the finding of Oswald (1996) that it is one of the most important barriers to the geographical mobility of labor.^[178] Freeman (2007) lists, without presumption of being exhaustive: mandated works coun­cils, employment protection laws, minimum wages, extension of collective-bargaining coverage, lifetime employment, peak-level collective bargaining, wage flexibility, teams, job rotation, temporary employment contracts, social dialogue, apprenticeship programs, occupational health and safety rules, defined benefit, and defined contribution pension plans. To name a few other examples: Oliver (2008) draws attention to industrywide wage scales, Boeri (2011, 1183) adds regulations on working hours, Blau and Kahn (2002, 4) include the public-sector share of employment. Antidiscrimination measures easily come to mind as a further example,^[179] and we will meet more below when we discuss recent contributions to the literature. Obviously, the discussion of earnings in relation to household income in the preceding section suggests new candidate institutions such as parental leave, child care provisions, or individual entitlements to the choice of working hours.

Second, the analysis of the effects of institutions still demand attention even if they were clearly defined, for several reasons: the actual significance of individual institutions and the type of effects they may have, their embedding in a larger set of institutions and also in the wider economy, and the potential pitfalls of international comparisons which have taken center stage in the literature.

In many studies institutions seem to be taken at face value and equated with what they look like dejur, and, naturally, it is their de facto implementary force or “bite” that counts (Eichhorst et al., 2008, 18). That bite may depend on its particular enforcement—laws and rules can be strictly enforced or they may be a dead letter to which no one pays atten­tion. Enforcement may be automatic when it is the responsibility of a supervising inspec­torate, or it may be costly and cumbersome if it is the responsibility of the individual who feels duped—minimum wages and other provisions diverge importantly in this respect (Benassi, 2011). Institutions may also be general provisions whose precise nature is filled in by actual policy. The minimum wage is again a case in point when the law establishes its mere existence while its actual (uprated) level is determined by policy making—the United States being the leading example.^[180] For policies the bite is the heart of the matter.^[181] Note that this further blurs the distinction between institution and policy. Institutions may also differ in the nature of their implementation: legally prohibiting or prescribing certain behavior, or economically encouraging or discouraging it, and consequently in the type of effects they may have. An example may be a prescriptive rule of employment protection versus a hiring subsidy for disadvantaged groups of active labor market policies. Note that the implementation of an institution may not be either black or white but can have different shades, a cross section within as well as between countries, and can also differ over time.^[182] Evidently, the de facto significance of an institution may come close to actually measuring its effects, and this can pose a methodological problem.

Often the effects are scrutinized for individual institutions as there is no clear theory on their coherence or interactions as a (national) set (Eichhorst et al., 2008, 17, 24, 29). However, institutions may partly balance or reinforce each other, say a country with strong employment protection could be mitigating the possible upward pressure on wages by collective bargaining, or employment “at will” may be neutralized by individ­ual contracts. It is easier to expand union membership if workers are protected from the threat of being fired. Diffusion of part-timers may reduce the quest of work leave permits. Generosity of unemployment benefit schemes may increase voluntary mobility and raise the demand for publicly provided training. Similar functions may be provided by differ­ent institutions. For example, Garnero et al. (2013) conclude to the functional equivalence with regard to earnings inequality of statutory minimum wages in some European countries and minimum-pay provisions of sector collective labor agreements combinedwith high bargaining coverage in other countries.^[183] Gottschalk and Smeeding (1997, 647) warn for the risk of double counting the effects of institutions when consid­ered in isolation (union density and the minimum wage in their example).

There may be deeper dangers in international comparisons. Institutions may catch the eye more readily than other international differences and be considered in relative isola­tion, enhancing the risk of their effects being overestimated. Blau and Kahn (1999) focus their overview on some 20 OECD countries stating that this selection permits utilizing “the similarity in educational levels, technology, living standards and cultures among these countries as de facto controls in examining the effects of institutions.” Freeman (2007,18) cautions against the methodological implications of the fact that the number of countries is small compared to that of institutions. Conversely, appearances may be deceptive and the potential dissimilarities of institutions that may look the same at first sight, need to be taken into account. We have already seen the possible divergence between educational attain­ment and skill levels in spite of the extensive efforts spent on a standardized measurement of educational systems (ISCED). Freeman’s (2000) plea for a metric may be more demanding than thought but above all it may be necessary but not sufficient. The above observations about the enforcement and bite of institutions apply particularly in a com­parative context, to prevent comparing apples with pears. More importantly, there may be deep-seated differences in the general economy as have been illustrated forcefully in recent years by, for example, the havoc wreaked by the larger propensity to consume by private households in the United States as compared to many other countries.^[184]

In sum, the study ofthe role of institutions needs to account for the force of those insti­tutions, their mutual interactions at the national level, supply and demand in the labor market, and also the broad structure of the economy. The latter potentially puts on the research agenda institutions that affect the economy more broadly such as those governing the flexibility of exchange rates^[185] or international capital movements, which have under­gone important liberalization in many European countries since the end ofthe 1970s and may have weakened employees and unions vis-a-vis employers. An important lesson of the minimum wage debate and the contribution made by Card and Krueger (1995a), who no longer started from the a priori of a negative effect on employment, is to prevent a stack­ing ofthe cards against institutions. As Freeman (2007,2) observes, “many adherents to the claim (that labour institutions impair aggregate performance, authors) hold strong priors that labour markets operate nearly perfectly in the absence of institutions and let their priors dictate their modelling choices and interpretation of empirical results.”

18.4.3 Why Do LMIs Exist?

The economic rationale and potentially beneficial effects of creating and/or preserving LMIs need to be accounted for from the start.^[186] According to Freeman (2007) there are three ways in which institutions affect economic performance: by altering incentives, by facilitating efficient bargaining, and by increasing information, communication, and pos­sibly trust. In his cursory review, the evidence shows that labor institutions reduce the dis­persion of earnings and income inequality, which alters incentives, but finds controversial effects on other aggregate outcomes, such as employment and unemployment.^{[187] [188] In his opinion, the modest effect would be attributable to the fact that “the political economy of institutional interventions rules out collective bargaining settlements and regulations that are truly expensive to an economy. No country would impose a minimum wage that dis­employed a large fraction of the work force; and no union or employer would sign a col­lective bargaining agreement that forced the firm to close.” In this perspective, a positive contribution of institutions would be observed whenever and wherever they solve trans­action cost of individual bargaining, according to the prediction of the Coase theorem.}

Regulations in the labor market as defined by Botero and coauthors (2004) emerge by government desires to protect the weaker side in a labor relationship. ⁵ They show that the orientation of governments to the political left is often associated to more stringent labor market regulations (politicalpower theory), but find the legal origin to be even more relevant to accounting for cross-country variation (especially when considering the trans­plantation of legal systems in the colonial era, much in line with sociological theories of path dependence—legal origin theory). According to the latter, common-law countries tend to rely more on markets and contracts, whereas civil-law (and socialist) countries on regulation (and state ownership): as a consequence, civil-law countries do regulate labor market more extensively than common law ones. The legal origin, possibly adopted for efficiency reasons in mother countries, becomes exogenous for former colonies, thus allowing a study of its causal impact on the origin of institutions.^{[189] [190] Following this line of argument, several papers account for the endogenous emergence of some LMI as an (optimal) solution for at least of a subset of agents. A controversial contribution to this approach is given by Saint-Paul (2000), who aims to identify gainers and losers of a given institution. In his view, each institution creates a rent (i.e., a difference between the paid wage and the outside option), which is unevenly distributed in the workforce. Because the employed workers enjoy most of the benefits of these rents, they obviously represent the largest constituency advocating the preservation of institutions (political insider mech­anism). This has to be traded-off against the rise of unemployment, which is associated to higher wages, and this represents the most serious threat to the continuation over time of an institutional setup. Ifwe accept Saint-Paul’s view that the most relevant conflict within the workforce is between the skilled and the unskilled, then “labour market rigidities mostly redistribute between skilled and unskilled labour” (Saint-Paul, 2000, p. 6). Ignor­ing within-group inequality, this means that institutions affect earnings inequality by affecting the skill premium and the (unskilled) unemployment rate. In this perspective, institutions emerge when the constituency represented by the employed unskilled dom­inates those of the skilled and of the unemployed (which is a different coalition than the one supporting fiscal redistribution, for example). The relationship between inequality and institutions becomes ambiguous: institutions create or enhance wage differences, but wage inequality may support the introduction of LMI as an alternative device for redistribution.8 Similarly, different institutions may reinforce each other, revealing the potential existence of a politicoeconomic complementarity, which contributes to explain­ing why empirically we observe clusters of institutions, often indicated as social models (Amable, 2003; Hall and Soskice, 2001). For reasons of viability, labor market reforms are more likely to emerge after a period of crisis, when the bias toward the status quo is weakened and the rise of unemployment allows the formation of alternative constit­uencies. A rather different view on rents in the labor contract, however, is offered by Manning (2011). According to him, rents are pervasive in the labor market, because of frictions in hiring and recruiting, separation costs due to investment in specific human capital, and collusive behavior on both sides (employers and employees). If imperfect competition is therefore taken as the relevant paradigm,[191] [192] the regulation of this market (via wage bargaining or wage setting by public authorities, as in the case of minimum wage) acts as a second best device, which may achieve Pareto improvements (as in the case of the minimum wage under monopsony).}

In a recent contribution, Aghion et al. (2011) frame the existence of labor market reg­ulations as an incomplete and less-efficient substitute for the quality oflabor relations. They rationalize their argument with a model of learning of the quality oflabor relations: the unionization decision is seen as a costly experimentation device aimed at finding out more about cooperation at the workplace. Thus, the existence oflegal provisions (such as a min­imum wage) reduces the learning incentive. Because beliefs are gradually updated based on past experiences, the authors obtain the prediction of a coevolution ofbeliefs (as measured by the quality oflabor relations perceived by top executives) and institutions (as measured by the stringency of minimum wage). ⁹ As a consequence, distrustful labor relations lead to low unionization and high demand for a direct state regulation of wages. In turn, state reg­ulation crowds out the possibility for workers to experiment with negotiating and to learn about the potential cooperative nature oflabor relations. This crowding-out effect can give rise to multiple equilibria: a “good” equilibrium characterized by cooperative labor rela­tions and high-union density, leading to low-state regulation (the Nordic countries), and a “bad” equilibrium, characterized by distrustful labor relations, low-union density, and strong state regulation of the minimum wage (some of the Mediterranean countries, and especially France). Their empirical application covers 23 countries over the period 1980-2003, and shows that the quality oflabor relations is negative correlated with either union density or state regulation of the minimum wage (while controlling for other insti­tutional measures such as unemployment benefits and the tax wedge).

Also recently, Alesina et al. (2010) have proposed a model where the emergence of employment protection and minimum wage provision is accounted for by cultural traits, namely the strength of family ties. In their theoretical model, individuals are born with different preferences with respect to family ties: those characterized by weak ties are geo­graphically or sectorally mobile and achieve an efficient allocation by being matched to jobs providing the highest productivity; however, those characterized by strong ties rationally select labor market rules (such as firing restrictions and minimum wages) that restrain the monopsonistic power of local employers while accepting a less-productive allocation. Another rationale of living close to family members is that it provides addi­tional insurance against unforeseeable shocks (including unemployment). The authors prove the existence of two stable Nash equilibria: one where everybody chooses weak family ties, votes for labor market flexibility, and changes her or his initial location (high mobility); another where everyone chooses strong family ties, votes for stringent labor market regulation, and stays in the original (birth) location. In the latter case the labor market is monopsonistic because workers are immobile, and workers limit employers’ power by means of labor regulations. Empirically, they show the existence of positive cross-country correlations between the strength of family ties and labor market rigidities. More convincingly, they also find that individuals who inherit stronger family ties (i.e., second-generation immigrants from countries that record high preferences for family values) are less mobile, have lower wages, are less often employed, and support more stringent labor market regulations.

In their historical review of the introduction of severance payment schemes in 183 countries, Holzmann et al. (2011) suggest three rationales for the introduction of such schemes: (1) as a primitive form of social benefits (anticipating the introduction of ben­efits for unemployment and retirement), thus providing an answer to a demand for insurance; (2) as an efficiency-enhancing human resource instrument (a sort of bonding between workers and firms, to minimize the loss of firm-specific knowledge) solving the hold­up problem; and (3) as a proper job-protection instrument, intended to enhance perma­nence in employment of main earners in the household.

If we restrict ourselves to the minimum wage, the historical account provided by Neumark and Wascher (2008) suggests that this institution has emerged as a counterbal­ance of power in the labor contract, preventing the exploitation of child labor (minimum-wage settlement power assigned to law courts in New Zealand in 1894 and in Australia 2 years later) or women (Fair Labor Standards Act, introduced at the fed­eral level in the United States in 1938). Viewed in this perspective, the minimum wage would represent a device aimed at preventing a “race to the bottom” competition among firms, more than a measure aimed at sustaining the incomes of poor families.⁹⁰ Seen from

In a similar vein, Agell and Lommerud (1993) proposed a model where setting higher wages promoted higher growth by eliminating low productivity enterprises.

the side of union leaders, minimum-wage legislation represents an improvement in the outside options of workers, inducing an increase in their bargaining power. The sum of these two effects may create an unusual coalition of large companies and worker unions supporting the introduction and/or the periodical updating of wage minima.⁹

18.4.4 Do LMIs Matter for the Economy?

In their overview of LMIs, Blau and Kahn (1999), with a careful discussion of the ratio­nale, draw some implications for studying the causalities. They look back at “an explosion of research” on the economic impact of institutions and conclude that institutions do appear to matter. In their view, the evidence across the literature that institutions affect the distribution of wages is more robust than for employment levels. Freeman (2001) supports this, saying that institutions identifiably affect the distribution, but that other effects on the macroeconomy and on efficiency are hard to discover and modest at best. Later he states even more forcefully that “institutions have a major impact on one impor­tant outcome: the distribution of income... By contrast, despite considerable effort, researchers have not pinned down the effects, if any, of institutions on other aggregate economic outcomes, such as unemployment and employment” (Freeman, 2009, pp. 19-20; see also Freeman, 2005). Nickell and Layard (1999, p. 3078) seem more ret­icent about the role of institutions when they conclude that “[m]ost of the gross features of unemployment and wage distributions across the OECD in recent years seem expli­cable by supply and demand shifts and the role required of special institutional features such as unions and minimum wages is correspondingly minimal.” These are not the last words about the role of institutions with regard to the dispersion of wages—let alone that of earnings incorporating the hours dimension which we deem of special interest here— as we will see when we turn to more recent contributions to the literature in the next two sections and to our empirical approach in Section 18.5.

So over the 1980s and 1990s a vast literature has grown, which seems to tend into two main directions: supply and demand on the one hand, institutions on the other. Each side acknowledges the relevance of the other, there is talk even of an SDI (supply-demand- institutions) model (Freeman and Katz, 1995; Katz and Autor, 1999; see also Lemieux, 2010) but little has grown out of that since, and in reality—understandably given the above-mentioned concerns—the prime focus of the market view and the institutional view seem to have grown more independent of each other. The flurry of institutions make them look overdetermined, and, by comparison, technological change—the driver of supply and demand—underdetermined. Over the 2000s many new arguments have been developed: polarization of the distribution, offshoring of productive activities, sharp growth in the upper tail of the distribution, top taxation, focus on tasks and skills, two-tier

See their review of empirical evidence based on minimum wage voting across US states (Neumark and Wascher, 2008, chap. 8).

nature of reforms of institutions, growing importance of performance pay, rise of “new institutions,” and, last but not least, new contributions have been made with regard to the minimum wage. These contributions seem firmly placed in either one or the other of the two main directions. In this respect the recent Volume 4 of the Handbook of Labor Economics repeats the preceding Volume 3. Acemoglu and Autor (2011) hardly even touch upon institutions in their conclusions, whereas Boeri (2011) focuses exclusively on aspects of institutions.

We think that Manning’s (2011) approach of imperfect competition in the labor mar­ket, which aims to leave behind the thinking in terms of canonical models and departures from these, may indicate a third route that can provide a different and ultimately more unified perspective. From the starting point that rents are inevitable and pervasive— though it is unclear how large they are and who gets them—Manning (p. 996) suggests that their very existence creates a “breathing space” in the determination of wages and allows the observed multiplicity of institutions on efficiency grounds. He concludes (p. 1031) that “[o]ne’s views of the likely effects of labour market regulation should be substantially altered once one recognizes the existence of imperfect competition.”⁹² An important corollary seems that institutions do not “cause the labour market to func­tion differently from a spot market” (Blau and Kahn, 1999, p. 1400) but that this market should not be considered a spot market but instead needs institutions for its proper func­tioning from the very start. Thus, a better principle for analyzing supply and demand as well as institutions may be that institutions are equally pervasive: every act of supply and demand goes together with an institution of some kind, and their existence and effects shall be accounted for from the start.

18.4.5 Recent Theories Based on Demand and Supply of Labor Inputs The review of theories of earnings inequality provided by Neal and Rosen (2000) a decade ago focused on the allocation of workers to jobs (the Roy model), on individual human capital accumulation (the Ben Porath model), on the search models (yielding var­iations in tenure—for a recent review see Rogerson et al., 2005 or Rogerson and Shimer, 2011), and on imperfect observability of either ability or effort (efficiency wage and con­tract theories). They adopted an individual perspective of wage determination, which did not allow great scope for the institutional framework to affect the resulting earnings dis­tribution. In such a perspective, wage inequality can be considered as the outcome of changes in the relative demand and supply of labor inputs. Starting from the original paper by Katz and Murphy (1992) and the literature originated since then (reviewed in Katz and Autor, 1999), the so-called canonical model predicts that the wage differential between skilled and unskilled workers accommodates an expanding demand for skilled

Note, however, his observation that the actual effects of (or, for that matter, the limits to) institutions are an empirical matter.

labor (SBTC, induced by introduction of computers in production) and a contraction of the demand for unskilled labor (due to increasing competition by developing countries). Demographic changes (variations in cohort size, immigration) and/or educational choices may partly attenuate (or even offset) these changes. The resulting dynamics of inequality can be predicted by tracking down these movements (Acemoglu, 2003).

In this framework, wage-setting institutions affect the flexibility of relative wages, creating a trade-off between wage differential and relative unemployment; when consid­ering inter-industry wage differentials, it translates into lower employee quit rates and longer queues ofjob applicants.⁹ Consider, for example, an increase in the relative demand for skilled labor (upskilling), at given supply of labor inputs. If the wage differ­ential cannot adjust the relative excess demand for skilled labor (because minimum-wage legislation prevents a downfall of the unskilled wage and/or union bargaining prevents an excessive rise of the skill premium), then the unskilled workers will experience an increase in their relative unemployment rate. This effect will be more pronounced the higher the substitutability between labor types.

It did not take long into the new century before Card and DiNardo (2002) mounted a fierce critique ofthe thesis of skill-biased technological change. Their arguments are both theoretical and empirical. From a theoretical point of view, a constant SBTC rate does not yield a permanent skilled/unskilled wage differential, as long as the relative supply is sufficiently elastic (see Atkinson, 2007b). On the empirical side, they revisit the evolution of American wage inequality since 1967, almost back to the starting point of the literature but now extending to include more recent occurrences over the 1990s. This refers to the problem already mentioned that technological change lacks a positive identification in economic models but is commonly subsumed in the unexplained leftovers. To avoid the tautology that this implies, they look for independent empirical measures of techno­logical change that can be incorporated in the model: the introduction of PCs and the Internet, the size of the IT sector in the economy, and the use of computers by individuals at work—particularly disaggregated by personal characteristics.^{[193] [194] From this material, the general trend in technological change seems unabated over the 1990s, if not increasing because ofthe Internet. The disaggregated use of computers points, among other things, to a larger role among women than men, particularly among the less-educated women whereas the best-educated men have closed the gap to their female counterparts. From this, Card and DiNardo conclude that computer technology should have widened gen­der differentials for the most highly educated and narrowed them for the least educated.}

On the inequality side, they argue from a fresh inspection of the data (using different sam­ples, sources, and inequality measures), “viewed from 2002” as they say, that there has been a pattern of a strong episodic rise in inequality in the 1980s, preceded by near stability before and after, during the 1970s and 1990s respectively. ⁵ From a comparison of the two, demand and supply, they conclude to “a fundamental problem... that rises in overall wage inequality have not persisted in the 1990s” and also to various puzzles, including the fact that the gender differential has diminished irrespective of education. In summary, they find the evidence for SBTC to be surprisingly weak. They do think there has been substantial technological change but deplore that this has diverted attention away from inequality trends that cannot be easily explained by this. The critique of SBTC is the main point of their contribution, not the design of an alternative explanation of inequality. However, Atkinson (2007b, 2008) points out that their critique of SBTC ignores the dynamics of the process and implicitly assumes a curve of skilled labor supply whose speed of adjustment is inversely related to the distance from an infinitely elastic one. International differences in the wage differential may reflect differences in the speed of that adjustment. Card and DiNardo end their contribution by teasing the reader with a quick exercise about the min­imum wage that shows a strong correlation between the evolution of its real level and aggregate hourly wage inequality (P90:P10) over the entire period 1970-1999.

Autor et al. (2006, 2008) have shown that the period of rising earnings inequality in the US labor market during the 1970s and the 1980s has been replaced by job polarization (simultaneous growth of the share of employment in high-skill/high-wage occupations and low-skill/low-wage occupations) in the following two decades. Despite the fact that the emergence of polarization crucially hinges on the procedure according to which occupations are ranked (educational attainment, wage rank, task content), also many European countries feature similar patterns: the decline in blue-collar jobs (mostly held by uneducated men) and the expansion of service jobs (mostly held by women and youngsters). One suggested interpretation (Autor et al., 2003) points to the increase in productivity of information and communications technology (ICT), which would have replaced middle-skilled administrative, clerical, and productive tasks with computer­operated machines.

Autor et al. (2008) have taken up the challenge of what they call a “revisionist” lit­erature of both the description and the explanation of US wage inequality since the 1970s. They object to the episodic interpretation of the rise in wage inequality; that is, they contrast this with ongoing inequality growth in the top half of the distribution combined with initially (1980s) increasing and subsequently (1990s) declining inequality

⁹⁵ Lemieux (2006a,b) finds a concentration of the increase in the 1980s together with a concentration of within-group inequality change among male and female college graduates and females with some college, implying an increasing concentration of wage inequality at the very top of the wage distribution. In addi­tion, Lemieux (2006c) finds a role for changes in the composition of the labor force after the 1980s. in the bottom half.^{[195] [196]} They view that initial lower-half increase as episodic indeed and incorporate the minimum wage as a potential explanatory factor in their approach; how­ever, they find only a modest role when modeled together with relative supply and demand. For the 1990s they agree that the slowing down ofinequality growth poses a prob­lem for the SBTC thesis, but only for the “naive” SBTC story as they call it, which is based on a dichotomy of high skills and low skills. They aim to improve on this by arguing a more detailed approach, based on the dispersion of occupations by their skill levels, measured as the mean years of schooling of an occupation’s occupants (weighted by their hours worked), and distinguishing between different types of tasks that can be performed in an occupation, showing that this works out differently between the 1980s and the 1990s.

The occupations and tasks approach can be viewed as a step along the route for further research pointed out by Levy and Murnane (1992), opening up an important black box albeit at the level of industry and not of the firm. In principle, though not always in practice, it also advances on the traditional SBTC approach by distinguishing between properties of the occupation and of the worker. Routine tasks were first stressed by Autor et al. (2003), polarization by Goos and Manning (2003, 2007). The approach aims to provide an answer to the problem posed to the SBTC thesis by the strong slowdown in wage inequality growth after the 1980s. It implies a significant shift in the SBTC thesis and the underlying empirics. Modern technology is complementary no longer to higher levels of skills and education but to nonroutine types of work. Although before work­place computerization was indiscriminately interpreted as skill biased and furthering the demand for higher skills, it is now taken to substitute for routine tasks that are defined as cognitive and manual activities that can be accomplished by following explicit rules. Therewith it reduces the demand for workers predominantly performing such activities, implying a more polarized effect on educational levels. Autor et al. (2003) focus on American employees, and the period 1960—1998 and combine CPS data with Dictionary of Occupational Titles (DOT) classifications. They analyze the shift in tasks that has resulted from both compositional changes across occupations and changes in task com­position within occupations, and find strongly diverging trends: negative for routine cog­nitive tasks from the 1970s and routine manual tasks from the 1980s and strongly positive for nonroutine cognitive tasks whereas nonroutine manual tasks decline steadily and strongly over the entire period.^[197] Note that they focus on employment effects and do not link the results to wage inequality,^[198] though the implication is clear and to some extent spelled out in Autor et al. (2006): low-wage and high-wage employment both

expand while jobs with intermediate pay contract. Notably, employment trends do not seem to differ between the 1980s and 1990s, though, naturally, the gaps between increas­ing and decreasing types of tasks become much wider.

Goos and Manning scrutinize the UK data for similar developments between the mid-1970s and the late 1990s, using a variety of data sets, samples, and methodological approaches. They find a clear polarization across the distribution of occupations and, linking to wages, also across the wage distribution. They check various tenets of the SBTC thesis. They discuss first whether labor supply may have contributed to the polar­ization, because of the rapid growth in female workers, and better-educated workers, but they find these changes unable to explain the polarization pattern. As to educational attainment, they find an increase in almost all occupations. This may be due to either rising requirements of the jobs or overeducation of the occupants. The data are insuffi­cient to decide between the two hypotheses though the authors seem inclined to opt for the second one. On the demand side they touch upon other factors than technology that may have contributed: trade and especially the structure of product demand—though these are not necessarily fully independent from technology—but find no explanation for polarization either. From a counterfactual exercise of the wage distribution over the 1975-1999 period restricted to changes in the occupational distribution only they conclude that polarization can explain large fractions of the rise in wage inequality (51% lower half, 79% upper half). They underline the important implication that the contribution of within-job inequality is minor. This contrasts sharply with established explanations in terms of education and age where most of the action is within groups, and they point out that the between/within conclusion is sensitive to the choice of con­trols included in the earnings function. They leave open the explanation of inequality change in the lower half of the distribution which may be due to imperfect competition, including institutional changes such as declines in unionization or the minimum wage. Goos et al. (2009) show a polarization of employment by occupations for 16 European countries between 1993 and 2006; Goos et al. (2010, 2011, 2014) extend the analysis to include relative wages and also capture effects of product demand, induced by a lowering of relative prices in industries with routine tasks, and institutions. They find that relative occupational wage movements in Europe are not strongly correlated with technology and offshoring, which may be due to wage-setting institutions, and therefore consider relative wages as being exogenous. They conclude that the thesis of routine jobs is the most important explanatory factor for increasing polarization, and product demand shifts across industries mitigate it.

Dustmann et al. (2009) find increasing wage inequality for Germany in the upper half of the distribution over the 1980s and 1990s.¹⁰⁰ This is attributed partly to composition

Spitz-Oener (2006) looks at the employment side of occupational polarization in Germany over the 1980s and 1990s.

changes and largely to technological change, as occupations at the top grow faster. For the lower half they find increasing inequality only in the 1990s, not before. For this they suggest possible episodic explanations such as a decline in unionization and an inflow into the country of low-skilled labor after the demise of the communist regime; the latter lends a role to the relative supply of skills. Following Autor et al. (2008) they conclude that the naive or canonical SBTC hypothesis cannot explain these trends, but they find support for the “nuanced” tasks-focused hypothesis as they note that occupations in the middle of the distribution decline compared to those at the bottom. In summary, they believe that the German results add unifying evidence to the pattern of polarizing effects of technological change already found for the United States and the United Kingdom.

We conclude our discussion of this stream of the literature with its current culminat­ing point, the overview and further development of the task-based approach to SBTC by Acemoglu and Autor (2011) for the latest Handbook of Labor Economics (Ashenfelter and Card, 2011).^[199] Note, however, that Mishel et al. (2013) provide various arguments why the evidence for the job polarization of these is weak. Although the canonical model builds on the unity of skills, tasks, and job (better-educated/talented workers obtain skilled jobs where they perform more complex tasks), the task-based approach considers a job as a collection of tasks, which can be executed by workers of different abilities, though at different level of productivity, and even by machinery. The empirical classi­fication of tasks is still in its infancy; they are classified according to three attributes: rou­tine, abstract, manual. “Offshorability,” meaning that the performance of certain tasks is internationally footloose, is added as another important job dimension, which can over­lap with each of the three types of tasks.^[200] This theoretical approach improves upon the canonical model by accounting for job polarization, real wage decline for some groups of workers (but not in a monotonic relationship with skill ranks), and offshoring as an alternative explanation of reductions in jobs to technical change.

Acemoglu and Autor’s model considers a continuum of tasks (unit of work activity that produces output, similarly to occupations) and different levels of skills (capability to perform various tasks); given existing supply of skills in the labor market, profit maximiz­ing firms allocate skills to tasks, given existing prices. Capital and/or offshoring may replace workers in performing tasks. The key assumption is the existence of comparative advantage of skills in executing tasks: more skilled workers are more productive in exe­cuting more complex tasks when compared to less-skilled workers. This structure creates a sort of hierarchical sorting associated to comparative advantage.

Wage flexibility ensures full employment of all workers. Given perfect substitutability among workers in task assignment, wages dynamics depend on the relative supply of skills (as in the canonical model) and on task assignment rules, which then allow for a potential competition in task execution posed by technological progress and/or offshorability. With their model they make a sharp prediction: “[I]f the relative market price of the tasks in which a skill group holds comparative advantage declines (holding the schedule of comparative advantage constant), the relative wage of that skill group should also decline—even if the group reallocates its labour to a different set of tasks (i.e., due to the change in its comparative advantage)” (p. 1152). The impact on the overall wage inequality is hard to predict, because the relative wages (high to medium skill and medium to low skill, when only three skill levels are considered) can move in opposite directions.

Acemoglu and Autor do not incorporate LMIs in their framework, which as they observe “depends crucially on competitive labour markets” (p. 1159) and can be thwarted by labor market imperfections of search and information and institutions such as collective bargaining by unions. The impact of certain LMIs may be enhanced by the way these affect the assignment of tasks to labor or capital as, for example, they may restrict the substitution of machines for labor for certain tasks, or conversely they may change the return to unionization, thus feeding back onto union density. The authors see this as an area for further research.

18.4.6 RecentTheories Based on LMIs

The other main current in the literature does take the existence and effects of LMIs into account. Also here, interesting contributions have been made throughout the 2000s. At the start of the new century, Blanchard and Wolfers (2000) launched their hypothesis that the internationally differential effects of institutions can be found particularly in countries’ responses to shocks. This view can offer a solution to the problem that, on the one hand, shocks alone cannot explain country differences, and, on the other hand, institutions on their own cannot explain long-run country performances. Their focus is the macroeconomy and unemployment, not wage inequality. Blau and Kahn (2002) connect to the latter in much of their book, and later extend this further by accounting for demographic shocks (with Bertola, 2007). However, their strong focus on the inter­national comparison of institutions may be the reason that they seem to overlook the shift­ing trend in the evolution of American wage inequality after the 1980s. The issue of this shift has been taken up by Lemieux (2008), for the United States. He objects to the con­sensus view on inequality growth that had taken root in the early 1990s which views this growth as secular and all-pervading. As we have seen his contribution (2010) extensively revisits the American data on the evolution of wage inequality, and pays particular atten­tion to the very top of the wage distribution, improving on the traditional adjustment for top-coding. From this, he concludes that in the 1970s inequality change was not all­pervading, whereas it was in the 1980s (though it also already showed more convexity at the top than at the bottom), and that since the 1990s inequality growth has been con­centrated at the top of the distribution. Growth in residual (“within”) wage inequality is general in the 1980s, although later it is largely confined to the college-educated category. In particular, relative wages continue to grow for postgraduates and their annual returns to education compared to high-school returns double between the mid-1970s and the mid- 2000s.

Lemieux (2008) also questions the consensus explanation of SBTC on the basis of this, but also because it leaves no room for a role of institutions in spite of the research that has shown the effects of unionization and wage-setting. He advocates an explanation that can account for both the above findings and the international differences and explores the possible contributions of institutions as well as of supply and demand.^{[201] [202] He finds that deunionization can explain one-third of the expanding inequality in each of the two halves of the distribution, and is also consistent with the divergence of English-speaking countries, where top incomes grew much more, from other countries. In addition to this the decline in the minimum wage has augmented lower-half inequality in the 1980s.104 On the side of supply and demand he thinks that more empirical research is needed before the tasks-based development of the SBTC thesis can be accepted as an explanation. That research should account for the fact that, contrary to what one would expect, the relative wages of occupations at the core of the IT revolution are suffering, and it should also answer the question why the process should not have occurred already during the 1980s. In addition, it should account for the growth of within-inequality at the top. For the latter he suggests modeling heterogeneous returns to education, which have as a key implication that both the level and the within-dispersion of pay of the better educated can rise relative to the less educated at the same time.[203] [204] [205]}

18.4.6.1 TopIncomes

Interestingly, Lemieux’s conclusion about upper-tail growth is consistent with the find­ings in the top-incomes literature (Alvaredo et al., 2013; Atkinson and Piketty, 2007, 2010; Atkinson et al., 2011, especially the summary Chapter 12; Piketty and Saez, 2003, 2006). Often a strong rise in labor incomes at the top is found, particularly in the United States, but not only there.^{1 6} This literature is suggestive of the role of yet another institution: income taxation, not as the traditional tax wedge but as marginal tax­ation at the top. “Higher top marginal tax rates can reduce top reported earnings through three main channels. First, top earners may work less and hence earn less— the classical supply side channel. Second, top earners may substitute taxable cash com­pensation with other forms of compensation such as non-taxable fringe benefits, deferred stock-option or pension compensation—the tax-shifting channel. Third, because the marginal productivity of top earners, such as top executives, is not perfectly observed, top earners might be able to increase their pay by exerting effort to influence corporate boards. High top tax rates might discourage such efforts aimed at extracting higher compensation” (Atkinson et al., 2011). Thus, the rise in top incomes and pay may have been encouraged by the lowering of top marginal tax rates. However causation may also run in the opposite way, because the rise of capital incomes in recent decades may have produced pressure for tax reductions. In a recent series of papers, (e.g. Piketty and Saez, 2013) have proposed formal models where the relationship between taxation and earn­ings has been carefully scrutinized. Most of the argument is a supply-side story, in the presence of imperfections: a reduction in the degree of progressivity would stimulate more effort and bargaining of CEOs and high-rank cadres with stakeholders, thus raising earnings inequality. Piketty et al. (2011) show a strong negative correlation between the Top 1% share and the top tax rate for a set of 18 OECD countries since 1960; the cor­relation also holds for CEO pay after controlling for firm characteristics and performance. The element of luck in CEO pay seems to be more important when tax rates are lower. It may point to more aggressive pay bargaining in a situation of lower tax rates. The high top tax rates of the 1960s were then part of the institutional setup putting a brake on top compensation through bargaining or rent extraction effects. In their view, the SBTC explanation seems to be at odds with international differences in top pay shares as well as their correlation to tax rates.

18.4.6.2 Minimum Wage^[206]

New contributions to the literature ofinequality and institutions are also found for var­ious other individual LMIs. First and foremost, we consider the literature on the effects of the minimum wage—an old debate by now (as old as the Department of Labor (viz. 1913) according to some)^[207] that nevertheless continues to attract passionate contribu­tions. The combination of wage and employment effects taken together determines the effects on annual earnings and, ultimately, incomes. Especially the impacts of a min­imum wage on employment remain a bone of contention—“the canonical issue in wider debates about the pros and cons of regulating labour markets” in the words of Manning (2011, p. 1026). A complication is that the employment effects likely relate to the level of the minimum wage and also differ between worker categories (e.g., Abowd et al., 1999; and also Philippon, 2001).

Neumark and Wascher (2008) hold a very critical attitude with respect to minimum wages. Exploiting cross-state and temporal variations in the United States, they conclude that minimum wages are ineffective in raising low wages and reduce employment opportunities for their earners.^[208] However, Dolton and Bondibene (2011) analyze employment effects for 33 OECD countries over 1976—2008 and find that existing evi­dence of negative effects is not robust. Dube et al. (2010) generalize Card and Krueger’s comparison of minimum-wage policy differences across US state borders and find no employment effects over 1990—2006, whereas Neumark et al. (2013) dispute their method and results. Allegretto et al. (2011) find no employment effects (including the hours dimension) distinguishable from zero over 1990—2009. Slonimczyk and Skott (2012) use US state variation to confirm their model predictions of a negative effect of minimum wages on the skill premium, owing to increasing overeducation of college- educated workers following an increase in mismatch. The overall effect is that a mini­mum wage would lead to a rise in both total and low-skill employment, accompanied by a fall in earnings inequality. Giuliano (2013) studies personnel data of a large US retail firm and finds no aggregate employment effect but composition effects that run contrary to standard theory. Interestingly, Dube et al. (2012) focus attention on effects on employ­ment flows. In the view of Richard Sutch (2010), the disemployment effects of pricing low-skill jobs out of the market may create incentives to invest more in human capital.

Most of the recent discussion revolves around whether there are spillover effects on wages higher than the minimum. A higher statutory minimum wage in itself compresses the wage distribution as it prohibits paying lower wages. However, the higher minimum rise may send ripples up the wage distribution—in the most extreme case all wages could be increased to the same extent and the dispersion of wages would remain unchanged. The minimum wage debate of the 2000s has generated new contributions particularly on this spillover or knock-on issue. Wages higher up may be raised for several reasons (Stewart, 2012, 618): the higher price for low-skilled labor incites substitution demand for higher-skilled workers, realignment of the marginal product of minimum wage workers affects the marginal product of other workers, firms maintain within-firm pay dif­ferentials for motivation, and reservation wages increase more broadly in certain sectors.

During the 1990s, spillover effects were detected in various contributions. Card and Krueger (1995a, 295) conclude to no effect at or above the 25th percentile of the wage distribution, which is well above the relative position of the minimum wage. Lee (1999) endorses an approach that compares to an estimated “latent” wage distribution (in the absence of the minimum wage). He finds effects beyond the P50:P10 ratio on other per­centile differentials across the entire distribution. At the end of the 1990s, the consensus view agreed to spillover effects though not extending high up the distribution (Brown, 1999, p. 2149).^[209] Over the 2000s, views on this have changed. Neumark and Wascher (2008, Section 4.3.2) discuss the previous literature and observe that the percentile approach as used by Lee may conflate spillover effects with disemployment effects of the minimum wage: as some of the least-paid lose their jobs, wage levels may increase at all percentiles of the distribution. Neumark et al. (2004) do not link to the wage dis­tribution but look instead at actual impacts on workers with wages up to eight times above the minimum wage, using US states with no rise in their minimum wages as con­trols. They find a wage elasticity with respect to the minimum wage of 0.25 at 1.5 times the minimum wage and much smaller effects above that level. Autor et al. (2010) are puzzled by Lee’s effects on the upper half of the distribution and attribute these to an omission of variables and the insertion of the median wage on both sides of the equation (division bias). They stick to Lee’s basic approach but propose econometric corrections and demonstrate the effects using a longer panel of US states with more variation in state minimum wages. They find substantial widening effects on the lower tail (P50:P10) of the decline in the real minimum wage over 1979—1988, but these effects remain well below those found earlier in the literature; they find only small effects for 1988—2009. Then they are puzzled by the large and increasing effects even at the 10th percentile in spite of the fact that currently the minimum wage is received by less than 10% of workers. They confront those effects with the possibility of mismeasurement and misre­porting of lower wages in the data and conclude from a detailed analysis that it cannot be ruled out that all of the spillover found is actually the result of such data problems.

111

Stewart (2012) adopts the direct estimations of Neumark et al. (2004) over a range of fractions of the minimum wage extending up to six times the minimum wage using differences-in-differences for comparisons between these factions. In addition, he exploits comparisons between minimum wage upratings that have differed in size (including no change period before the introduction of the minimum wage in 1999) while accounting for differences in general wage growth. Using British data, he concludes to no spillover effects. As the level of the minimum wage is steadily below the 10th percentile, he draws the logical inference that the changes in the minimum wage have not affected lower-half wage inequality as measured by P50:P10. That seems fair enough, but it also puts on the table the strength of this inequality measure as evidently the minimum wage may signif­icantly affect the within-distribution of the bottom decile. The top-to-bottom ration S10: S1 may be better suited to capture such effects. Butcher et al. (2012) revisit the effects on wage inequality and spillovers for the United Kingdom and do find spillover effects up to the first quartile of the distribution. In their view, decades of discussing the employment effects of the minimum wage—with very little to none as the consensus outcome—have been focusing on second-order effect, and instead they advocate developing a theoretical framework for thinking about its first-order effects on wage inequality, which, naturally, should be able to allow the possible absence of employment effects. They develop a non­competitive model with wage-posting instead of bargaining^[210] with imperfectly elastic labor supply to the individual firm. The authors elaborate on their model to consider the spillover effects to wage levels above the minimum wage. They derive those from a comparison between the actual wage distribution at and above the minimum wage and a counterfactual latent wage distribution derived with the help of the distribution preced­ing the introduction of the minimum wage in 1999. They find higher levels for the former compared to the latter up to 40% above the minimum wage, which corresponds with the 25th percentile of the aggregate wage distribution.

Finally, Garnero et al. (2013) show that statutory minimum wages (or equivalent sys­tems) represented by sectoral minimum rates combined with high coverage of collective bargaining—see also Boeri (2012)—are very effective in reducing earnings inequality. They combine harmonized microdata from household surveys (EUSILC), data on national statutory minimum wages and coverage rates, and hand-collected information on minimum rates from more than 1100 sectoral-level agreements across 18 European countries over several years (2007—2009—see also Kampelmann et al., 2013). Alternative specifications confirm that institutional variants of setting a wage floor reduce both between and within-sectors wage inequalities.

18.4.6.3 Union Presence

Card et al. (2004) study the relationship between wage inequality and unionization in the United States, Canada, and the United Kingdom over the period 1980-2005, showing that within narrowly defined skill groups, wage inequality is always lower for union workers than for nonunion workers. For male workers, union coverage tends to be concentrated at the middle of the skill distribution, and union wages tend to be “flattened” relative to nonunion wages. As a result, unions have an equalizing effect on the dispersion of male wages across skill groups. For female workers, union coverage is concentrated near the top of the skill distribution, and there is no tendency for unions to flatten skill differentials across groups. The effect of deunionization on US wage inequality is stronger at the top end of the distribution than at the bottom, as shown by Lemieux (2008) when updating the DiNardo et al. (1996) decomposition. In addition, the increase ofperformance pay schemes may have enhanced the within-group wage inequality at the top end of US distribution.^[211]

The decline in workers’ bargaining power in the Anglo-Saxon world is recorded by several authors (see, for example, Levy and Temin, 2007), but we have not found any convincing decomposition of the relative contribution of each specific institutions. However, when taking the dynamics of the wage share in the domestic product as an overall indicator of workers’ bargaining power, one would recognize a clear declining trend in most countries over the past decade, though some reversal can be recognized during the crisis period (ILO, 2008, 2010).^[212]

A parallel decline in workers’ bargaining power can underlie the decentralization of wage bargaining. Following recent changes in industrial relations in Denmark, Dahl et al. (2011) show the existence of a wage premium associated with firm-level bargaining rel­ative to sector-level bargaining, and a higher return to skills under more decentralized 115

wage-setting systems.

18.4.6.4 Unemployment Benefit

Even if unemployment benefits and employment protection are negatively correlated in the data (Bertola and Boeri, 2003), in principle they do respond to the same problem of reducing the intertemporal variability of workers’ earnings (Blanchard and Tirole, 2008).^{[213] [214] This may explain why research has paid less attention to the contribution of unemployment schemes to inequality reduction. Corsini (2008) studies the dynamics ofthe college premium in 10 European countries over the last decade of previous cen­tury. He finds a positive impact of the generosity of unemployment benefit (but a neg­ative correlation with duration), which is interpreted as the outcome of wage bargaining that takes into account the outside option.[215] Ifwe shift to individual data analysis, the results of Paul Bingley et al. (2013) on Danish data show that access to unemployment insurance is associated with lower wage-growth heterogeneity over the life cycle and greater wage instability, changing the nature of wage inequality from permanent to tran­sitory. Given data limitations, the authors are unable to control for moral hazard behavior of unemployed, who may be induced to lengthening their permanence in unemploy­ment, thus increasing cross-sectional inequality.[216]}

18.4.6.5 Employment Protection Legislation

Recent cross-country evidence has been summarized in the following way by World Bank (2012, 262): “Based on this wave of new research, the overall impact of EPL and minimum wages is smaller than the intensity ofthe debate would suggest.” However, Martin and Scarpetta (2011) express a different view, arguing that EPL reduces workers’ reallocation and prevents efficiency gains for highly productive workers, while avoiding job losses and/or real wage reductions for unskilled workers.^[217] In their review they list a series of papers based on changes in dismissal regulation, which find mixed evidence of EPL impact on labor productivity (see, among others, Bassanini et al., 2009; Boeri and Jimeno, 2005; Kugler and Pica, 2008; Schivardi and Torrini, 2008). Productivity dynam­ics may translate one to one into wage dynamics in a competitive environment; in non­competitive models, firing restrictions raise the bargaining power, creating artificial divisions among workers when groups of firms are exempted (see Leonardi and Pica, 2013). Similarly, EPL exemptions for firms may create artificial wage differences among workers, due to their differential cost, thus enhancing wage inequalities; for example, Karin Van der Wiel (2010) provides evidence referring to a policy reform of terms of notice in the Netherlands. A further connection between EPL and wage inequality can be found in comparative analysis: Bryson et al. (2012) show that higher labor (and product) market regulation is associated with lower use of incentive pay (ranging from 10% of covered workers in Portugal to 50% of the workforce in the United States). Inas­much as incentive-pay schemes increase within-group earnings inequality (Lemieux et al., 2009), this induces a negative correlation at the aggregate level between earnings inequality and EPL indexes.

18.4.6.6 Labor Market Policies

Kluve (2010) provides an extensive meta-analysis based on a data set that comprises 137 active labor market program evaluations from 19 countries. Four main categories of ALMP are considered across European countries: (i) training programs, (ii) private-sector incentive schemes (such as wage subsidies to private firms and start-up grants), (iii) direct public employment programs, and (iv) “services and sanctions,” a category comprising all measures aimed at increasing job search efficiency, such as counselling and monitoring, job-search assistance, and corresponding sanctions in case of noncompliance. His main finding is that traditional training programs have a modest significant positive impact on postprogram employment rates, but both private-sector incentive programs and ser­vices and sanctions show a significantly better performance. Evaluations of direct employment programs, on the other hand, are around 25% points less likely to estimate a significant positive impact on postprogram employment outcomes. Although effective­ness is here defined in terms of employment impact, they can be easily mapped one-to- one to wage inequality whenever the unemployed are taken into the picture.

18.4.6.7 Stepwise Institutional Change

A new and different line of argument regarding institutions is nicely summarized by Boeri (2011). After reviewing existing institutional differences among European countries and stressing their persistence over time, he proposes a taxonomy of institutional changes (reforms), in terms of orientation and phasing-in. The orientation concerns the question whether they reduce (e.g., by making employment protection less strict and/or unem­ployment benefits less generous or by expanding the scope of activation programs) or increase the wedge (e.g., by increasing labor-supply-reducing taxes on relatively low- paid jobs) introduced by LMIs between supply and demand. Boeri accordingly classifies a reform as either decreasing or increasing the (institutional) wedge. The second charac­teristic relates to the phasing-in of reforms: this can be either complete or partial. In the former case, the change in the regulation eventually involves everybody. In the latter case, even at the steady state, the reform is confined to a subset of the population. The timing is also important. Even a complete phasing-in may involve a very long tran­sitional period, so that the steady-state institutional configuration is attained beyond the planning horizon of management’s potential involvement by the reform (Boeri, 2011, 1184). A two-tier reform is then defined as the case involving either a partial phasing­in or when its complete phasing-in requires more than 30 years, the average length of the working life in many countries. According to data collected over the period 1980—2007 for the European Union, the two-tier pattern is prevailing in most of the institutional dimensions. This has obvious implications in terms of earnings inequality, especially between insiders and new entrants (typically women and youngsters). With the help of a search model a la Pissarides-Mortensen, Boeri shows that institutions affect the threshold below which it is no longer convenient for either the employer or the employee to continue the work relationship. Even if the underlying inequality pattern depends on idiosyncratic shocks hitting individual productivity, the boundaries of the distribution of realized wages are institutionally determined, owing to variation in the equilibrium unemployment. According to the model an increase in unemployment benefits raises the reservation productivity at which matches are dissolved as the outside option of workers has improved: in equilibrium there is a higher probability ofjob loss, a lower job finding rate, higher unemployment and average wage.^[218] Conversely an increase in firing taxes has the opposite effect of maintaining alive jobs with a lower match productivity. This reduces the gross job destruction rate and positively affects wages. An increase in employment conditional incentives (modeled as an employment subsidy) makes the labor market tighter, and increases the duration of jobs at the expenses of a decline in entry wages. Finally, an increase in the activation scheme reducing recruitment costs features higher job finding and job-loss rates, whereas the effects on unemployment and the average wage are ambiguous. When liberalizing (wedge-reducing) reforms are applied to only a fraction of workers (temporarily creating a dual labor market), then earnings inequality expands: insiders enjoy a surplus over outsiders at the same productivity levels, which is increasing in the difference in replacement rate offered to the unemployed (coming from long-tenured jobs with respect to those coming from short-tenured jobs), in the employ­ment conditional incentive and in firing taxes, which matter more when workers have more bargaining power.

Returning to the more general, internationally comparative literature, developed by Blau and Kahn (2002) and others, we find the contribution of Koeniger et al. (2007) who look beyond cross-sectional differences at the comparative evolution of wage inequality over time, and extend to more OECD countries over a longer period, focused on overall wage inequality of males taken from the OECD database. They treat the various insti­tutions (union density, union coordination/centralization, the minimum wage, employ­ment protection, unemployment benefit generosity and duration, and the tax wedge) simultaneously and also model some interactions. On the demand side they control for the aggregate economy (unemployment rate), the relative supply of skills, interna­tional trade (import intensity), and technology (R&D intensity). They add some coun- terfactual simulations, including one that attributes US institutions to the other countries. They find compressing effects on the wage distribution of most institutions which explain at least as much as trade and technology do on the demand side. Applying American reg­ulations would increase wage inequality in Continental Europe by 50-80%. The authors observe, however, that endogenizing the institutions, that means accounting for their dependence on supply and demand, will likely reduce the effects somewhat.

Finally, as we have observed above, the context of household (joint) labor supply potentially augments the number of institutions that need to be addressed, adding paren­tal leave, maternity leave, part-time work regulations, and any other institution affecting the flexible use of working hours. Analyses of this (e.g., Dupuy and Fernandez-Kranz, 2011; Thevenon and Solaz, 2013) are few, and they are focused on employment chances and/or pay penalties of gender/motherhood/family, not on the wage dispersion.

18.4.7 Summing Up

Over time the literature seems to have gone in two different directions that tend to grow further apart—not in the sense of interactions (one retorting to the other) but in the sense ofintegrating the approaches into one framework. Freeman (2007, p. 24) signaled the risk of creating the social science equivalence of “epicycles”—aimed at preserving Ptolemaic views on the earth as the center of the universe—for the institutional approach. How­ever, the same danger may be looming for the supply-and-demand approach, which has been adding tasks, offshoring, and consumer preferences, in an attempt to dispel doubts about the relative demand of skills as a tautology. The institutional approach faces an abundance of institutions for which it lacks a clear criterion of choice; the supply- and-demand approach by contrast is challenged by the need for finding better empirical measures of technological change. However, a fortunate effect of the interactions just mentioned has been the great interest that is now taken in the very data on wage inequal­ity. The take on the data’s properties, advantages, and disadvantages has greatly improved over time. Consideration of the data at later points in time alter the stylized facts and also show that consensus explanations may be temporary and can break down when data for later periods become available and shine a different light on preceding periods. In spite of this, the prime aim of future work on both sides should be to integrate the other side into the framework. Pursuing that may be more a problem of empirical method for the insti­tutional side, and on the demand-and-supply side the problem may be more on the the­oretical side as long as institutions continue to be viewed as alien bodies. For both sides there is a perspective of work to do at the firm level. Matched employer-employee data (Cardoso, 2010; Lane, 2009) can help enlighten the role of both institutions and labor supply and demand (see, e.g., Andersson et al., 2006; Matano and Natichioni, 2011 for some interesting attempts). In addition, though much attention has been paid to data quality, a better grasp of the customary use of inequality measures seems desirable.

18.5.