Has wage inequality stopped growing?

Monthly Labor Review

Volumn 120, Issue 12, 1997, Pages 3-16

  Bernstein, Jared ^a   Mishel, Lawrence ^b  

^a ECONOMIC POLICY INSTITUTE   (United States)

^b NONE

Author keywords

[No Author keywords available]

Indexed keywords

EID: 0002104011     PISSN: 00981818     EISSN: None     Source Type: Journal    
DOI: None     Document Type: Article

References (28)

1. See Frank Levy and Richard Mumane, "U.S. Earnings Levels and Earnings Inequality: A Review of Recent Trends and Proposed Explanations," Journal of Economic Literature (September 1992), pp. 1333-81.
2. The Gini coefficient, or Gini index, is a conventional measure of inequality that ranges from zero when all earnings are shared equally to unity when one person has all the earnings in an economy. Along with other scalars, such as the log variance, the Gini index and relative wage percentiles are probably the most common techniques for measuring wage inequality. John Dinardo, Nicole Fortin, and Thomas Lemieux have introduced an innovative technique for examining changes in inequality through the use of kernel density estimators. (See John Dinardo, Nicole Fortin, and Thomas Lemieux, "Labor Market Institutions and the Distribution of Wages, 1973-1992: A Semi-Parametric Approach," Econometrica [September 1996], pp. 1001-44.)
3. For an example of this type of truncation, see Lynn A. Karoly and Gary Burtless, "Demographic Change, Rising Earnings Inequality, and the Distribution of Personal Well-Being, 1959-1989," Demography, vol. 32, no. 3, 1995, pp. 379-405.
4. Truncating the distribution above the necessary cutoff will lead to a downward bias in the growth of the Gini index in a period when the true earnings of workers with very high wages are rising relative to the wages of others. We find, for example, that the series of Gini coefficients for the bottom 98 percent of the earnings distribution has risen more quickly than that of the bottom 95 percent, particularly in the 1990s.
5. Actually, we calculate Pareto-imputed means by gender, they are available from the authors.
6. In our analysis of wages in this article, we exclude outlier cases, defined as cases wherein the hourly wage is less than $0.50 or greater than $100.00 in 1989 dollars.
7. This statement is strictly true only for those workers with identical demographic weights. Our analysis in this article applies the CPS demographic weights.
8. In an earlier version of this article, we used the bottom 98 percent of the hourly wage distribution. Here, our sample for this column comprises the bottom 98 percent of the annual earnings distribution.
9. Also, top codes were raised over this period, but we impose the same top code ($99,999) as in earlier years for the sake of consistency.
10. See Paul Ryscavage, "A Surge in Growing Income Inequality?" Monthly Labor Review, August 1995, pp. 51-61.
11. Ibid., p. 57.
12. Ibid., p. 60. Ryscavage finds an increase in the share of high earners in the 1993 March data with allocated earnings. Because there is a positive correlation between nonreporting and income levels, an increase in allocations among high earners should lead to a downward bias in inequality growth. (The Census Bureau allocates earnings using a "hot-deck" procedure that assigns the reported earnings value from a record with characteristics similar to those of the case with the missing value.) The impact of the survey design change also was analyzed by Anne E. Polivka. (See her "Data Watch: The Redesigned Current Population Survey," Journal of Economic Perspectives, vol. 10, no. 3, pp. 169-80.) Although Polivka does not examine the effect of the design change on inequality specifically, she shows that the change had no significant effect on the overall unemployment rate.
13. Ibid., p. 60. Ryscavage finds an increase in the share of high earners in the 1993 March data with allocated earnings. Because there is a positive correlation between nonreporting and income levels, an increase in allocations among high earners should lead to a downward bias in inequality growth. (The Census Bureau allocates earnings using a "hot-deck" procedure that assigns the reported earnings value from a record with characteristics similar to those of the case with the missing value.) The impact of the survey design change also was analyzed by Anne E. Polivka. (See her "Data Watch: The Redesigned Current Population Survey," Journal of Economic Perspectives, vol. 10, no. 3, pp. 169-80.) Although Polivka does not examine the effect of the design change on inequality specifically, she shows that the change had no significant effect on the overall unemployment rate.
14. Note that the question on usual earnings for the Outgoing Rotation Group does not specify a period. The survey is, however, solely focused on the respondent's employment situation of the previous week. Thus, we are confident that responses to the usual-earnings question are correctly interpreted.
15. Note that we do not use the Pareto procedure to impute the mean value of earnings above the top code in the data from the Outgoing Rotation Group, because, in this series, the earnings cutoff changes only once, in 1986, when it almost doubles. In that case, Pareto imputations would lead to implausible jumps in the year-to-year Gini index around the period of the change. This problem does not appear in the March CPS data, for which top codes are more frequently updated.
16. See, for example, Willard L. Rodgers, Charles Brown, and Greg J. Duncan, "Errors in Survey Reports of Earnings, Hours Worked and Hourly Wages," Journal of the American Statistical Association (December 1993), pp. 1208-18. As we point out later, this critique also holds for the Survey of Income and Program Participation, in which workers are asked how many hours per week they usually worked at their job over the past month.
17. Under the assumed pattern of bias, observed inequality could grow either because the bias worsened or because there was simply a larger relative share of high earners. Only in the former case, however, would the rising trend in inequality be misrepresentative. In their comparison of annual earnings and earnings from the usual pay period, Rodgers, Brown, and Duncan found that low earners tend to overreport, while high earners tend to underreport, their earnings. As the snare of high earners rises, this pattern would have the effect of dampening inequality growth in data from the Outgoing Rotation Group survey relative to March CPS data.
18. Rodgers, Brown, and Duncan, "Errors in Survey Reports," p. 1217.
19. See David Webster, "Wage Analysis Computations," Appendix B, in Lawrence Mishel, Jared Bernstein, and John Schmitt, The State of Working America, 1996-97 (Annonk, NY, M. E. Sharpe, 1997). Prior to 1979, these data are derived from the May CPS.
20. The Gini coefficient is more sensitive to transfers from the middle of the wage scale than are other scalars, such as the log variance, which is more sensitive to transfers from the bottom of the distribution. This may partially explain the similar acceleration in the Gini index and the 90-50 cutoff ratio in the 1990s.
21. The ratio between the median female and the median male hourly wage was 0.63 in 1979 and 0.77 in 1995. (See Mishel, Bernstein, and Schmitt, The State of Working America.) Eighty-two percent of the narrowing of the gap over that period was due to a decline in the median male hourly wage.
22. Note that, because women are paid less than men, aggregating the two groups increases the proportion of low-wage workers and, in turn, increases the variance of the distribution at a point in time. This effect can lead to increases in measures of overall inequality over time, due to growth in the labor force participation of women. A countervailing factor, however, is the compression of the gender gap.
23. For example, Robert I. Lerman, Reassessing Trends in Earnings Inequality in the U.S. (Washington, DC, The Urban Institute, 1997), uses this survey as a series of cross sections.
24. This question gets at the issue of underreporting, which may be systematic. Such comparisons are typically made for the CPS by the Census Bureau. (See, for example, Current Population Report, Series P-60, no. 172, Money Income of Households, Families, and Persons in the United States: 1988 and 1989 [U.S. Department of Commerce, Bureau of the Census, 1991, pp. 385-91].)
25. The preceding calculations use family Gini coefficients from table F-4 of the Census Bureau's World Wide Web site and take the Gini coefficient for earnings from table 1, column one, of the text of this article. Both of these statistics are derived from the March CPS.
26. See Rebecca M. Blank and Alan S. Blinder, "Macroeconomics, Income Distribution, and Poverty," Sheldon Danziger and Daniel Weinberg, eds., in Fighting Poverty: What Works and What Doesn't (Cambridge, MA, Harvard University Press, 1986), for an example of work of this nature and for historical citations. A recent paper by Blank and David Card ("Poverty, Income Distribution, and Growth: Are They Still Connected?" Brookings Papers on Economic Activity, no. 2 [Washington, DC, Brookings Institution, 1993]), found that a 1-point fall in unemployment led to a gain in family earnings in the bottom quintile that was twice as large (in percentage terms) as the gain for the highest fifth.
27. See Rebecca M. Blank and Alan S. Blinder, "Macroeconomics, Income Distribution, and Poverty," Sheldon Danziger and Daniel Weinberg, eds., in Fighting Poverty: What Works and What Doesn't (Cambridge, MA, Harvard University Press, 1986), for an example of work of this nature and for historical citations. A recent paper by Blank and David Card ("Poverty, Income Distribution, and Growth: Are They Still Connected?" Brookings Papers on Economic Activity, no. 2 [Washington, DC, Brookings Institution, 1993]), found that a 1-point fall in unemployment led to a gain in family earnings in the bottom quintile that was twice as large (in percentage terms) as the gain for the highest fifth.
28. For evidence of this claim, see tables 3.24 and 3.25 in Mishel, Bernstein, and Schmitt, The State of Working America, in which growth in wage inequality is decomposed into within- and between-group components.