Keywords

JEL Classifications

Introduction

Study of the structure of wages has been a preoccupation of economists for a long time and dates back at least as far as Adam Smith. Until the early 1990s economists commented on the remarkable stability of the wage structure in the post-war period. But then many empirical studies (for example, Bound and Johnson 1992) noticed that wage inequality in America had risen dramatically since the late 1970s. Related empirical research (notably by Goldin and Katz 1999, 2001) went back further in time uncovering other periods of changing wage structures in American history. Other countries, notably the United Kingdom, also saw a significant increase in wage inequality at about the same time as the recent US changes (Machin 1996). These observations kick-started what has become a huge empirical and theoretical literature seeking to measure and explain changes in wage inequality (see the survey of Katz and Autor 1999). Since wages are a major part of people’s income and economic well-being, the increase in wage inequality feeds through to income, consumption and poverty rates. So understanding the patterns of wage inequality is important from a normative as well as a positive perspective.

In this article we examine what has happened to the wage distribution since the 1960s, looking principally at the United States, where the bulk of the economic research has focused, but where possible also examining other countries. Section “What Has Happened to the Wage Distribution?” describes the observed changes in the structure of wages (although we fully acknowledge there are some contentious, and as of yet unresolved, issues about the observed patterns of change). Section “Explanations of Changes in Wage Inequality” looks at the main explanations of the observed changes that have emerged from the large body of work in this area. Section “Conclusions” offers some conclusions.

What Has Happened to the Wage Distribution?

Overall Trends in US Wage Inequality

To set the scene, Fig. 1 plots out the salient features of the US full-time weekly wage distribution from 1963 through to 2003. At least three things stand out from Fig. 1. First, educational wage differentials – measured as the gap in pay between college and high school educated workers – have risen consistently since 1979 (after falling somewhat in the 1970s and rising somewhat in the 1960s). The rate of increase was more rapid in the 1980s than after 1992. (This ongoing secular rise in educational wage premia is also seen in the hourly wage series from March outgoing rotation group of the Current Population Survey, CPS; see Lemieux 2006.) Second, the 90–10 wage differential – defined as the difference in weekly pay for those at the 90th and 10th percentiles of the overall wage distribution – has been rising since 1976 (and maybe even earlier). Third, the ‘residual’ 90–10 wage differential – the difference between those at the 90th and 10th percentiles of the overall wage distribution after controlling for education, experience and gender – has risen consistently since 1967, especially after the mid-1970s (see Juhn et al. 1993). This increase in ‘within group’ wage inequality has also generated much excitement and interest from theorists, but is particularly hard to interpret in the light of compositional changes (Lemieux 2006).

Wage Inequality, Changes In, Fig. 1
figure 233figure 233

Changes in US wage inequality, 1963–2003. Note: based on full-time weekly earnings for all workers in the March Current Population Survey (Source: Autor et al. (2005))

Full size image

Even though different data-sets show some differences and there are some variations in inequality measures across data sources, the overall picture is one of a dramatic increase in American wage inequality since 1979.

Comparing the United Kingdom with the United States

The United Kingdom is another country where wage inequality has risen dramatically. Comparison of the United States and United Kingdom is useful to pin down certain issues to do with the rise in wage inequality. One important point is that since 1980 there are marked decadal differences in the opening up of the wage structure. Analysis of US and UK micro-data uncovers a clear picture for the 1980s in both countries: wage growth was more pronounced at higher points of the distribution, and faster wage growth higher up the distribution is almost monotonic in both counties, leading to large increases in wage inequality. An important difference, however, is that in the United Kingdom there was positive wage growth throughout the distribution whereas in the United States workers in the bottom quartile actually experienced zero or negative wage growth.

The picture becomes more complex post-1990. In both countries the 90–50 continues to diverge (‘upper tail inequality’) whereas the 50–10 (‘lower tail inequality’) in the United States actually shrinks, indicating some wage compression. In the United Kingdom the 50–10 is stable (increasing a bit in the 1990s and shrinking a bit in the 2000s). Overall then, the increase in wage inequality has been stronger in the upper tail than the lower tail taking the period as a whole, and has been more pronounced in the 1980s than post-1990.

A marked and important similarity between the two countries is the continuous and rapid growth of wages at the very top of the distribution. In the United Kingdom, wage growth at the 95th percentile (and above) is greater than at other percentiles of the wage distribution in the 1980s, 1990s and 2000s. This is also true for the United States (except for the 10th percentile in the 1990s). So within the picture of overall rising inequality the very rich have done particularly well.

The other key feature of the changing wage distributions in the United Kingdom and the United States (and elsewhere) has been the polarization of work into ‘good jobs’ and ‘bad jobs’ (defined as high-wage and low-wage jobs). While there has been significant growth in well-paid ‘good jobs’ at the upper tail of the distribution (like lawyers, senior managers and consultants) there has been an increase in low-paid ‘bad jobs’ in the lower tail of the distribution (like cleaners, hairdressers, shop assistants and burger flippers). In the 1990s especially it seems that the middle of the distribution seemed to do somewhat worse than those at the top or bottom. These findings have been reported on in the United States (Autor et al. 2006), United Kingdom (Goos and Manning 2007) and Germany (Spitz-Oener 2006).

The Experience of Other Countries

There is less systematic evidence for the evolution of the wage distribution outside of the United States and the United Kingdom, especially for more recent years. Table 1 uses OECD data to show 90–10 male wage ratio for a range of countries between 1980 and 2000. Broadly speaking, the 1980s rise in inequality was seen only in the United Kingdom and the United States and in specific countries where particular episodes to move to a much more market-oriented economy occurred (notably New Zealand). Elsewhere wage inequality did not alter much. The 1990s is a little different, with evidence of widening wage structures starting to occur in places previously characterized by stable wage structures – Germany is a very good example of this. Moreover, as we discuss below, the Continental European countries did experience a larger increase in unemployment, which may be due to the same underlying forces that have pushed up wage inequality in Britain and America.

Wage Inequality, Changes In, Table 1 Male 90–10 wage ratios across countries, 1980–2000
Full size table

Explanations of Changes in Wage Inequality

A natural place to begin to analyse the observed changes in the wage structure is to consider a model of changes in supply and demand. We then need to incorporate institutional features (such as minimum wages and trade unions) into the model.

Sources of Skill Premia: Supply and Demand

Rising wage inequality has been accompanied by an increase in the relative demand for skilled or educated workers. This is evident since, despite the increase in the relative supply of more skilled workers in many countries, their relative wage has also gone up, suggesting that relative demand for skilled workers has been rising faster than relative supply. In Table 2, for example, the proportion of graduates grew from 20.8 per cent of the population in 1980 to 34.2 per cent in 2004 in the United States. (In the United States the graduate measure is having a bachelor’s degree or higher – that is, excluding people with some college who do not get a degree.) The equivalent figures from the United Kingdom were even more dramatic – the growth in graduates was from 5 per cent to 21 per cent over the same time period. However, at the same time the relative wages of graduates compared with those of non-graduates increased. In a competitive model of the labour market with skilled and unskilled workers, these facts can be reconciled by an increase in the relative demand for skilled workers.

Wage Inequality, Changes In, Table 2 Aggregate trends in graduate/non-graduate employment and wages, UK and USA, 1980–2004
Full size table

A simple way to formalize this, following Katz and Murphy (1992), is in the context of a constant elasticity of substitution (CES) production function with two labour inputs:

$$ {Q}_t={\left[{\alpha}_t{\left({a}_t{N}_s\right)}_t^{\rho }+\left(1-{\alpha}_t\right){\left({b}_t{N}_u\right)}_t^{\rho}\right]}^{1/\rho } $$
(1)

In Eq. (1) aggregate output is Q and is produced with college educated-equivalent skilled labour (Ns) and high school-educated equivalent unskilled labour (Nu) in period t. The parameters a and b represent skilled and unskilled augmenting technical change, α indexes the share of work activities of skilled labour and ρ is a parameter that determines the elasticity of substitution between skilled and unskilled labour \( \left(\sigma =\frac{1}{1-\rho}\right) \). Skill-biased technological changes involve increases in a/b or α.

Assuming college and high school equivalents are paid their marginal product we can use Eq. (1) to solve for the ratio of marginal products of the two types of labour, Ws/Wu, and relative supplies of labour, Ns/Nu, in year t as:

$$ \ln {\left({W}_s/{W}_u\right)}_t=\left(1/\sigma \right)\left[{D}_t-\ln {\left({N}_s/{N}_u\right)}_t\right] $$
(2)

where

$$ {D}_t=\sigma \left[\ln \left({\alpha}_t/\left(1-{\alpha}_t\right)\right)+\rho \ln {\left(a/b\right)}_t\right] $$
(3)

is a relative demand index of shifts favouring college equivalents and is measured in log quantity units. The impact of changes in relative skill supplies (Ns/Nu) depends on the elasticity of substitution, σ. The larger this parameter is, the bigger will be the effects of supply changes on relative wages. Eq. (3) shows that changes in D can arise from (disembodied) skill-biased technical change, non-neutral changes in relative prices or quantities of non-labour inputs and shifts in product demand.

Katz and Murphy (1992) implemented an empirical version of Eq. (2), replacing D with a linear time trend (‘trend’) for US data between 1963 and 1987. They estimate:

$$ \ln {\left({W}_s/{W}_u\right)}_t={\upgamma}_0+{\upgamma}_1 trend+{\upgamma}_2\ln {\left({N}_s/{N}_u\right)}_t+{v}_t $$
(4)

finding \( {\widehat{\gamma}}_2 \) to be significantly negative (equal to − 0.709), implying an elasticity of substitution of about 1.4 (σ = −1/\( {\widehat{\gamma}}_2 \) = 1:41), with a significant trend increase in the college premium of 3.3 per cent per annum (\( {\widehat{\gamma}}_1 \)= :033). In the literature that has followed the estimates of the elasticity of substitution are typically in the 1.4–1.6 range (see, for example, the study of Autor et al. 2005). The main point to take away from these estimates is that there appears to be a systematic demand shift towards more skilled workers throughout the four last decades of the 20th century.

This is not to suggest that supply side changes are unimportant. Deviations of relative skill supplies from the trend are negatively associated with deviations of the relative wage from trend as suggested by γ2 < 0. The slowdown of the growth of education in more recent cohorts is certainly one factor accounting for the increase in inequality as shown by Card and Lemieux (2001). But the most important factor over the longer run in accounting for the growth in educational wage differentials appears to be the trend demand shift towards the more skilled. The critical question then becomes: what could account for this change?

The Cause of Relative Demand Shifts: Technology or Trade?

To date, the two main explanations for the demand shift towards the more skilled are skill-biased technological change (SBTC) and increased international trade. We examine each of these in turn.

Skill-Biased Technological Change

Equation (3) above directly relates the change in the skill premia to SBTC. The idea is that new technologies such as information and communication technologies (ICTs) are complementary with the skills of more-educated workers. More-educated workers may find it easier to cope with the uncertainty surrounding new technologies in general, or may have a particular advantage in using ICT effectively. Rapid falls in the quality-adjusted prices of ICT or a more rapid investment in new technologies (for example, from higher R&D intensities) could therefore have shifted demand towards more-skilled workers.

There is now abundant empirical evidence that suggests that SBTC is an important and international phenomenon (for example, see the survey in Bond and Van Reenen 2007). A typical analysis estimates the following cost share equation (usually for industries or workplaces):

$$ \Delta SHARE={\beta}_1 TECH+{\beta}_2\Delta \ln \left(K/Y\right)+{\beta}_3\ln \left({W}_s/{W}_u\right)+e $$
(5)

where SHARE is the wage bill share of skilled workers, TECH is a measure of technical change, K is the capital stock, Y is value added, Ws/Wu is relative wages, Δ the difference operator and e an error term. This relationship can be derived from the stochastic form of a translog short-run variable cost function with labour as the two variables and physical capital and technological capital as the two fixed factors (for example, Berman et al. 1994). The test of skill-biased technical change is whether β1 > 0, and the overwhelming preponderance of econometric evidence supports this finding.

An example of the genre is Machin and Van Reenen (1998), who examine this relationship using manufacturing data across many industries in seven OECD countries (the United States, the United Kingdom, France, Japan, Germany, Denmark and Sweden) in the 1970s and 1980s. In all of the countries examined they found that demand was shifting more quickly towards skilled workers in the more technologically advanced industries (that is, β1 > 0 in Eq. (5)). This was robust to using either occupation or education as a measure of skills, using either R&D intensity or computer use as a measure of technology, and instrumenting own R&D with frontier (US) R&D. In most countries they also found evidence of capital–skill complementarity (β2 > 0). Estimating versions of Eq. (5) in other countries, in non-manufacturing sectors (for example, Autor et al. 1998) and on more disaggregated plant-level data (for example, Doms et al. 1997) also appears to uncover evidence of SBTC.

There are several other sources of evidence on SBTC. Berman et al. (1998) report evidence of faster skill demand shifts occurring in the same sorts of industries in different countries, and one may view this as informing the SBTC hypothesis (to the extent that similar industries in different countries utilize similar technologies). A less-used alternative to test for SBTC is to regress the adoption of technologies on skill prices (that is, when skilled workers’ wages rise relative to those of unskilled workers this should depress the incentive to adopt new technologies) or skilled labour supply – some evidence for this method is in Caroli and Van Reenen (2001) and Doms and Lewis (2006), and also supports SBTC. A third method is to directly estimate the production function or the cost function underlying the factor demand Eq. (5). This has also tended to uncover evidence of skill–technology complementarity (for example, Bresnahan et al. 2002). Finally, some authors have directly regressed individual wages on computer use or controlling for other factors (for example, Krueger 1993). In our view, this is a rather unsatisfactory test of SBTC, however, as computers are likely to be allocated to more productive workers, as has been found by several studies (Chennells and Van Reenen 1997; DiNardo and Pischke 1997).

Although we have stated the SBTC hypothesis in quite a blunt fashion, the influence of technical change almost certainly acts in a more subtle ways to affect outcomes as detailed case studies suggest (Blanchard 2004). For example, some econometric studies suggest that technical change operates through organizational changes (for example, through decentralizing or delayering hierarchies) that are typically associated with increased demand for skilled workers (Caroli and Van Reenen 2001; Bresnahan et al. 2002). Moreover, computerization does not simply involve increasing all skill demand, but it substitutes for different types of tasks. Autor et al. (2003) offer a more nuanced version of the SBTC hypothesis, arguing that computerization reduces the demand for routine tasks (for manual and non-manual workers) but results in an increase in demand for analytic or non-routine skills. Thus, routine non-manual tasks (for example, clerical work) may be replaced by computers, whilst some non-routine tasks done by manual workers (like cleaning) are largely unaffected by IT. The evidence on polarization of work referred to above where the ‘middle’ of the wage distribution has suffered at the expense of the bottom as well as the top is in line with this. Building on upon these empirical observations, Autor et al. (2006) develop a model where IT replaces routine tasks to rationalize the experience of the 1990s when polarization of jobs occurred in the United States.

Overall then, there is strong support for the importance of SBTC. Some critics (most strongly expressed in Card and DiNardo 2002) argue that SBTC cannot be the reason for increased inequality because technical change is continuous whereas the change in wage inequality is episodic. Regardless of whether one agrees with the characterization of technical change, this misses the point that SBTC is meant to account for the longer-run pressure to increase skill demand (the D in Eq. (2)) and not necessarily the ‘twist’ in the wage structure in the 1980s. Similarly, the fact that inequality growth slowed down post-1995 whereas productivity growth accelerated does not disprove the SBTC argument, as the speed of technical change is not the same as the bias of technical change.

Increased International Trade

At first glance, the simple Heckscher–Ohlin model of trade offers a seemingly cogent explanation of why unskilled workers have faired badly in recent decades. Less-developed countries such as China and India have become integrated into the global economy as trade barriers and transportation and communication costs have fallen. Unskilled workers in the OECD counties now have to compete not only with workers at home but also with a large number of workers overseas. The influx of cheap goods produced with low-skill labour puts downward pressure on the wages and employment opportunities of unskilled workers in the West, and is responsible for the observed shifts in relative labour demand.

To model this we explicitly consider two regions: ‘North,’ which is skill-abundant and ‘South’ which is unskilled-abundant. There are four industries: tradable high-skill intensive, tradable low-skill intensive, non-tradable high-skill intensive and non-tradable low-skill intensive. The Stolper–Samuelson theorem establishes that relative wages in each country will depend on relative output prices of the tradable industries: the higher the relative price of the skill-intensive good, the higher the relative wage of the skilled workers. What happens when a small open economy in the North moves from autarky to free trade? The removal of trade barriers increases the relative price of the skill-intensive good and this means the skill premium rises in the North.

Although this model is coherent, it also offers several other predictions which turn out to be at odds with the data (see Desjonqueres et al. 1999, for extensive discussion of these predictions). First, the increasing specialization of the North in skill-intensive goods under free trade means that employment should shift between industries to skill-intensive industries. But because relative skill prices have risen we should expect to see that employment within industries shifts towards (the cheaper) unskilled workers. Decompositions of the increase in the aggregate employment share of skilled workers, however, almost all show that within industries there has been a strong shift towards skilled workers. This might be because the level of aggregation of industries is too high, but more disaggregated industries and even firm-level studies suggest that a sizable proportion is ‘within’. Even more convincingly, Desjonqueres et al. (1999) show that non-traded sectors – such as hotels and wholesale outlets – also show a shift towards skilled workers (and an increase in the educational wage premium). This pattern of within-industry shifts is consistent with general SBTC, but inconsistent with the basic trade theory.

Second, we should observe that relative prices of the unskilled-intensive sectors should fall rapidly in the North. There is some evidence for this in the United States but there is no significant relationship for any other country (at least until the mid-1990s). Even in the United States the evidence from Krueger (1996) suggests that this relationship was only apparent after 1989, when wage inequality grew slowly. Finally, naive regressions that include import penetration and other trade variables in Eq. (5) generally find no role for these trade variables (for example, Machin and Van Reenen 1998). This does not take into account the general equilibrium effects underlying the Heckscher–Ohlin model, of course.

Overall there is little support for the trade-based explanation of demand shifts. There are two caveats to this conclusion. First, most of these studies were based on data prior to the early 1990s when China started to become more of a major exporter. Second, trade might induce some of the skill biased technological change discussed in the previous section as suggested by Acemoglu (2002).

Labour Market Institutions

Research trying to reconcile cross-country differences in change in wage inequality has emphasized the role of labour market institutions that affect wages differently in different places. There are several features of this work, ranging from studies that look in detail across countries to those that focus on the role played by particular labour market institutions like minimum wages or trade unions.

Cross-Country Evidence

As discussed in Section “What Has Happened to the Wage Distribution?”, there has been considerable heterogeneity in the evolution of relative wages across OECD countries since the 1970s. The rise in inequality was much stronger in the Anglo–Saxon countries (for example, the United States and the United Kingdom) than elsewhere (for example, France, Germany and Japan). Although the technology and/or trade shocks discussed in the previous subsections should be global events, the Continental European and Japanese economies have experienced a much greater increase in unemployment than the United States since the late 1970s. One view is that European unemployment and American inequality are ‘two sides of the same coin’ – institutional rigidities (and perhaps generous welfare benefits) placed a floor under the wages of unskilled workers in Continental Europe, resulting in increased unemployment rather than greater inequality. (There is a also a new, growing body of work arguing that tastes and social norms are important for explaining cross-country patterns of change; see, amongst others, Bénabou and Tirole 2006.)

This is probably too crude. Nickell and Bell (1995) have shown that relative unemployment rates between skilled and unskilled workers did not rise by as much as would be expected in this simple model. Similarly, the cross-country correlation between the growth in unemployment and earnings inequality is not very strong (for example, Burniaux et al. 2006). Finally, European countries may have been better at keeping up the growth of supply of the quantity and quality of skills than in the United States and United Kingdom (although Table 2 shows that skill expansion in the United Kingdom was very rapid).

At the very least, the fact that wage inequality has not risen in the countries where minimum wages and/or union power remained strong suggests that institutions do have an important role to play.

Minimum Wages

There is much evidence that minimum wages compress wage differentials (DiNardo et al. 1996). In the United States the real value of the Federal minimum wage fell significantly during the 1980s, and some authors argue that this can account for all of the change in wage inequality (for example, Lee 1999). By the same token the uprating of the minimum wage in the 1990s helps explain the slowdown in wage inequality. As Card and DiNardo (2002) emphasize, the time series pattern is very strong – see Fig. 2.

Wage Inequality, Changes In, Fig. 2
figure 234figure 234

The time series relationship between the US federal minimum wage and wage inequality (Source: Autor et al. (2005))

Full size image

A problem with the ‘purely institutional’ argument, however, is that it seems highly unlikely that the minimum wage can explain what is happening in the top half of the wage distribution. Analysis of the minimum wage suggests that the impact on workers above median wages is close to zero. Nevertheless, the most striking finding of the analysis in Section “What Has Happened to the Wage Distribution?” was that there appeared to be a secular increase in the 90–50 wage ratio since the late 1970s in the United States (and the United Kingdom). It is hard to reconcile these facts with the minimum wage-explains-all story. Similarly, when Autor et al. (2005) add the minimum wage to Eq. (4), although it has the expected negative sign it does little to reduce the long-run unexplained relative demand shift towards higher education wage differentials.

Where the institutional story does better is in accounting for the dramatic increase in residual wage inequality in the bottom half of the wage distribution in the 1980s. This residual wage change was more episodic, and most of the change is plausibly accounted for by the minimum wage (and compositional effects – see below).

Another problem with the pure minimum wage explanation is that wage floors changed much less in other countries where wage inequality also rose. For example in the United Kingdom, the minimum wage system that operated at the time when wage inequality rose (the ‘Wage Councils’) only covered a relatively small proportion of the workforce (around 12 per cent at the time of abolition in 1993). Furthermore, during the 1993–9 time period when all non-agricultural minimum wages were abolished in the United Kingdom, wage inequality at the lower end actually started to stabilize (Dickens et al. 1999; Machin and Manning 1994).

Trade Unions and Imperfect Competition

As with minimum wages there is robust evidence that unions act to compress wage differentials (for example, Freeman 1980; Card 1996). Since unions have declined in the United States and the United Kingdom, this may be another institutional mechanism putting upwards pressure on wage inequality. Unionization rates fell from 25 per cent to 15 per cent between 1979 and 1998 in the United States, and from 53 per cent to 31 per cent in the United Kingdom over the same period. Gosling and Lemieux (2004) argue that union decline can account for over a third of the increase in male wage inequality in both countries over the 1983–98 period.

As with the minimum wage explanation, it is rather difficult to evaluate these statistical decompositions as they are not based on an underlying economic model. But it does seem rather implausible that unions could be the major explanation in the United States for the ongoing increase in the 90–50 ratio since (a) they comprise such a small part of the workforce and (b) their membership is mainly drawn from the bottom half of the wage distribution.

An alternative set of theories has emerged that emphasizes rents derived from imperfect competition (albeit from a different source from unions). This approach has frictions in the labour market that generate heterogeneous wages even for identical workers. Some more productive or technologically advanced firms may share quasi-rents to workers who are matched with them (for example, Van Reenen 1996). If the dispersion of these wage premia has increased over time, this could lie behind the increased wage inequality. For example, in Caselli (1999) firms experiment with the uncertain new technology, and some of those that are successful obtain higher productivity, resulting in higher wages for the workers with whom they are matched. To date, there is little hard empirical evidence on these theories, although Faggio et al. (2006) offer some evidence that firm productivity heterogeneity has increased and this is linked to firm wage inequality as Caselli’s model would suggest.

Conclusions

There has been a dramatic increase in wage inequality since the late 1970s in the United States, the United Kingdom and other anglophone countries. A significant part of this is due to the growth of wage differentials between educational groups. We have argued that the fundamental reason for this is a long-run growth in the relative demand for skills driven by technology change (rather than trade). Changes in skill supply and institutional changes have affected the timing of how skill-biased technical change impacts upon the wage structure. The increase in inequality in the United States and the United Kingdom slowed down after 1990, but has continued to grow in the upper tail of the wage distribution, and wage inequality has started to rise in places previously characterized by stable wage structures (like Germany), indicating that explaining changing patterns of wage inequality remains high on the research agenda of empirical economists.

See Also