erlingfig2.png
VoxEU Column Industrial organisation Poverty and Income Inequality

Wage inequality: The spatial dimension

Income inequality has risen throughout the advanced world. Various explanations have been suggested for this, but these tend to focus on who you are. This column shifts the focus to where you work. Data from the US reveal that over the period 1992-2007, two-thirds of the rise in earnings dispersion was due to increased variation across establishments. Moreover, almost 80% of the increase in earnings dispersion among workers who remained at the same establishment from year to year was due to a widening of wages across establishments rather than within establishments.

Income inequality has risen throughout the advanced world.  Among major countries the US is the leader in inequality (Piketty and Saez 2014), but in the past decade or so it has risen more in Sweden than in any other OECD country (OECD 2015). Thanks to the Wall Street occupiers, the increase in the top 1% has received most of the attention (Atkinson et al. 2011).  Explanations are plenty, such as the effects of globalisation and technological change (Goldin and Katz 2008, Acemoglu and Autor 2012) or the weakening of trade unions (Machin 2016). With very few exceptions, however, they all take the same departure and focus on who you are – your skills, whether you are a union member or not, your gender, ethnicity, and so on.

The spatial dimension of wage inequality

Our new work shifts the focus – it depends on where you work (Barth et al. 2016).  Figure 1 illustrates the point with data for the US. The upper line shows the development of the variance in log weekly earnings from 1977 to 2010, calculated from the March Current Population Survey (CPS) conducted every year. The bottom line shows the development of the variance in the residual from a log earnings regression, conditioning on years of education, Mincer experience and its square, and race, all interacted with gender.

Considering the widening of the earnings distribution, it looks like everything is in the residual.1  So, what is in the residual?  Our analysis says what is in the residual is the distribution of average earnings across establishments, which we measure as the wage bill per employee from the US Census Bureau’s Longitudinal Business Database (LBD).  The middle long-dashed line in the figure shows that inequality in the average earnings among establishments increased at the same pace as the residual and is of sufficient magnitude to explain the bulk of inequality and the trend of inequality among observationally similar workers. Put differently, the widening of the residual earnings distribution results from a widening of dispersion across establishments, rather than from a widening across individuals within establishments.  Because the data in Figure 1 come from different sources, however, we needed additional data from a single source, with information on individual workers and the establishment for which they work, to perform an analysis of variance that nails down this claim.

Figure 1 Variance of individual earnings, residual earnings and establishment earnings, 1977-2010

The data that allow a complete decomposition of inequality and its trend due to who you are and who you work for are the Longitudinal Employer Household Dynamics Data (LEHD) combined with establishment data, which contain establishment and person identifiers.  The results for the 1992-2007 period preceding the Great Recession show that two-thirds of the rise in earnings dispersion is due to increased variation across establishments, while one-third is due to increased variation across individuals within establishments. Among the group of workers who remain at the same establishment from year to year, almost 80% of the increase in earnings dispersion is due to a widening of wages across establishments rather than within establishments.  This latter observation rules out a major role for increased concentration of workers by skill within establishments due, say, to outsourcing or specialisation across establishments. By definition, year-to-year stayers do not shift among establishments.  Similarly, we estimated an establishment fixed effect every year, conditional on human capital characteristics,2 gender and race of workers within an establishment, and calculated the variance of this establishment component.  More than 86% of the widening of the wage distribution across establishments is due to increased variance in the establishment component of pay.3

Figure 2 Change in average individual and establishment earnings by percentile of the earnings distribution, 1992-2007

The widening of the wage distribution across establishments benefited the top earners the most. Figure 2 (from Barth et al. 2016) illustrates this point. The solid line shows the growth from 1992 to 2007 in the average earnings level of each percentile of the individual earnings distribution.  The dashed line shows the growth in the average establishment component of the individuals in each percentile. The patterns are very similar, again showing that the widening of the wage distribution across individuals arises from an underlying widening of the establishment distribution.

The widening of the wage distribution across establishments is not only a matter of certain industries taking off. It happens within industries as well, and even within establishments within the same firm. Freeman (2014) shows how it has been happening within the hotel industry in the US, and similarly, within the hotel industry in the Boston area. This is not only about industry and region, it is about establishments becoming further and further apart.

The evidence that the bulk of the increase in inequality in the US occurred by establishments moving apart in earnings space raises new questions about wage determination and the power of competitive forces to compress wages toward the single market-clearing level of the most basic economic theory. Rent-sharing models, efficiency wage models, and dynamic monopsony models of frictions in the labour market generate wage differentials across employers beyond those associated with individual worker productivity. Earlier work on wages (Dunlop 1957) and the balkanisation of labour markets (Kerr 1954) pointed out the dispersion of pay among workers with different skills, but did not have the appropriate data to examine changes over time.  Why does the US labour market, arguably one of the least institutionalised and sensitive to supply and demand, prevent establishment differentials widening so much?  Where is the equalising force to eliminate these establishment wage differentials?

How wages respond to the heterogeneity of employer conditions, and to changes in those conditions, matters greatly for the efficiency of the allocation of resources and for incentives to innovate and implement new technology (Moene and Wallerstein 1997, Freeman and Kleiner 2005).  On the one side, the widening dispersion of wages would seem to sabotage the creative destruction of dynamic markets.  Low and falling wages at the bottom of the wage distribution may allow low productivity firms to remain in the market instead of exiting, while at the top of the distribution increasing wages at the best firms would likely reduce their acquisition of more workers. 

Conclusions

The observed increase in inequality among establishments has implications for the direction of policies to reverse the trend and eventually reduce inequality. To the extent that the change in inequality among establishments reflects the weakening of unions to establish ‘level playing fields’ in a sector, and/or of worker mobility across firms to do the same, policymakers should shift some of their focus on individuals and their skills to strengthening employer policies and institutions that determine the differences in pay among individuals.  To make headway against inequality requires both blades of the market scissors – demand as well as supply – employers as well as workers and the institutions that influence pay differences among establishments and firms.

References

Acemoglu, D. and Autor, D. (2012) "What Does Human Capital Do? A Review of Goldin and Katz's The Race Between Education and Technology", Journal of Economic Literature, 50:2, 426-463

Atkinson, A. B., Piketty, T. and Saez, E. (2011) "Top Incomes in the Long Run of History", Journal of Economic Literature, 49, 1: 3-71

Barth, E., Bryson, A., Davis, J. C. and Freeman, R. (2016) "It's Where You Work: Increases in Earnings Dispersion Across Establishments and Individuals in the US", Journal of Labor Economics, 34, 2, 2: s67-s97 

Dunlop, J. (1957) "The Task of Contemporary Wage Theory." In George W. Taylor and Frank C.

Freeman, R. B., and Kleiner, M. M. (2005) "The last American shoe manufacturers: Decreasing productivity and increasing profits in the shift from piece rates to continuous flow production",  Industrial Relations, 44, 2:307–30

Machin, S. (2016) "Rising Wage Inequality, Real Wage Stagnation and Unions" in  Cappellari , l., Polachek , S. W. and Tatsiramos, K. (eds.) Inequality: Causes and Consequences (Research in Labor Economics, Volume 43) Emerald Group Publishing Limited, pp.329 - 354

OECD (2015) In It Together: Why Less Inequality Benefits All, Paris: OECD

Pierson, (eds.) New Concepts in Wage Determination. NY: McGraw-Hill, pp. 117-39.

Freeman, R. B. (2014) "The Subcontracted Labor Market", Perspectives on Work, 38-42, LERA

Goldin, C. and Katz, L. F. (2008) The Race between Education and Technology, Harvard University Press

Kerr, C. (1954) "The Balkanization of Labor Markets", in E. Wight Bakke (ed.) Labour Mobility and Economic Opportunity, New York: John Wiley and Sons, PP. 92-110

Moene, K. O. and Wallerstein, M. (1997) "Pay inequality", Journal of Labor Economics 15, 3:403–30

Piketty, T. and Saez, E. (2014) "Inequality in the long run", Science, Vol. 344, 6186: 838-843

Endnotes

[1]This is not entirely true; this picture does not show the growth in dispersion due to the well known widening of educational differences (Goldin and Katz 2008) because this widening is counteracted by a narrowing of the distribution of earnings across gender over the same period, producing a relatively stable situation in the overall variance of the predicted earnings over time.

[2]We have obtained a measure of education by merging it in from the Decennial Censuses to the Longitudinal Employer Household data base for an 18% sample of the population.

[3]Increased worker-worker assortative matching explains nothing, whereas an increase in the assortative matching of workers to establishments explains some (about 11%) of the widening. Note, however, that this latter sort of matching between workers and establishments would not matter if there were no establishment component to pay in the first place.

3,150 Reads