Are the rich getting richer and the poor getting poorer?
Historical Census Bureau income statistics can shed some light on this debate. Although the
Census Bureau has been measuring income for a half-century, and a large number of factors
have been identified as contributing to changes in inequality, the causes are still not
entirely understood.
The Current Population Survey (CPS) is a rich source of data on income inequality.
During the past 50 years, the annual demographic supplement to the March CPS has provided
researchers with a wealth of data on the income distribution. Since 1947, the Census Bureau
has employed a commonly used measure, the Gini coefficient (also known as the index of income
concentration),/1/ to measure family income inequality. With two exceptions, the Gini
coefficient decreased between 1947 and 1968. During this period, the Gini for families
indicated a decrease in income inequality of 7.5 (+/- 2.1) percent.2 Since 1968, however,
this trend has reversed. Income inequality for families, measured by the Gini coefficient,
increased between 1968 and 1998 (see Figure 1). The net effect over the entire 1947-1998
period is an increase in family income inequality./3/
Changes in the earnings distribution have an effect on overall income inequality.
Studying the earnings distribution of people can provide some clues to the underlying causes
of overall household income inequality. Earnings, which are an important part of a person's
total money income, provide a good indication of how labor markets allocate income to
individuals. This is particularly important if changes in income inequality are due to
structural changes in the economy, which can translate into differences in wage premiums paid
to workers with certain skills./5/
Figure 2 depicts how earnings inequality has changed between 1967 and 1998 for both men and
women who were full-time, year- round workers, as measured by the Gini coefficient. The
earnings distribution for men remained stable, with a few exceptions, between 1967 and 1980.
This changed between 1980 and 1989; the Gini coefficient for men's earnings (presented in
Table 1) increased from 0.315 to 0.361-a 14.6 (+/- 1.5) percent rise.
Changes in the women's earnings distribution occurred quite differently. Earnings inequality
among women who worked full-time, year- round did not increase from 1967 to 1986. In fact, the
Gini coefficient indicates that from 1967 to 1980- a period of relative stability for the men's distribution-women's earnings inequality fell by 0.033 (+/- 0.01) points. By 1986, the Gini
coefficient for women's earnings had returned to its 1967 level. In 1989, however, the Gini
coefficient for women's earnings was 17.0 (+/- 1.9) percent higher than in 1980/6/ and
4.0 (+/- 2.3) percent higher than its 1967 level.
Over the 1967-1998 period, earnings inequality for both men and women who were full-time,
year-round workers grew consistent with rising income inequality.
Households are now the main demographic unit of analysis.
Living conditions have changed considerably in the last 50 years. Today, a smaller percentage
of people live in families (two or more people living together who are related by blood,
marriage, or adoption) than was the case in
the 1940s. As a result, the Census Bureau began collecting and reporting data on the income
distribution of households,/7/ a more comprehensive unit of analysis, beginning in 1967. Over
time, the importance of household data has increased.
A period of rising household income inequality: 1967 to 1992
Changes in data collection methodology between 1992 and 1993 affected the measurement of income
inequality. As a result of these changes and an inability to accurately measure their effects,
comparisons of income inequality that bridge the years 1992 and 1993 are avoided in the remainder
of the report. The timing of this methodological change was convenient; it appears that the
growth of household income inequality has slowed post-1992.
Between 1967 (when income data for households first became available) and 1992, the shape of the
household income distribution changed dramatically. This 25-year period was one of increasing
household income inequality--as evidenced by several measures. These changes, however, took place
during a relatively short period.
Household income inequality was generally stable between 1967 and 1980.
Measures of income inequality traditionally used to study the income distribution of the United
States suggest that the 1967-1980 period was one of relatively stable inequality. The Gini
coefficient for households in 1967 stood at 0.399 (+/- 0.01) (see Table 4). In 1980, the Gini
coefficient was 0.403 (+/- 0.01), not statistically different from its 1967 level.
Comparing the aggregate shares of household income received by each fifth of the income
distribution (presented in Table 2), another common method of examining income inequality, shows
growing income equality during this period (see Figure 3A). For example, the aggregate share of
income held by the households in the lowest fifth grew by 7.5 (+/- 4.3) percent from 1967 to
1980. At the same time, households in the top 5 percent of the distribution experienced a decline
in their share of aggregate income from 17.5 (+/- 0.90) percent in 1967 to 15.8 (+/- 0.61) percent
in 1980, a 9.7 (+/- 5.8) percent decline. From 1967 to 1980, there was no change in the share of
aggregate income held by households in the middle 60 percent and the top fifth of the income
distribution.
The choice of measurement method does make a difference.
The Gini coefficient and aggregate shares of income indicate that household income inequality was
relatively stable and may have decreased between 1967 and 1980. Examination of selected
percentiles of the household income distribution tells a different story. Traditionally, the
Census Bureau has employed a number of selected percentile limits and ratios to study changes in
household income inequality. These include the ratio of income for the household at the 95th
percentile to the household at the 20th percentile (95/20); the 95th percentile to the median
(95/50); and the 20th percentile to the median (20/50).
In contrast to the shares and Gini measures, these percentile measures (as presented in Table 3)
suggest that household income inequality increased from 1967 to 1980. The 95/20 ratio was 6.33
(+/- 0.04) in 1967 and grew to 6.82 (+/- 0.04) by 1980-a 7.7 (+/- 0.76) percent increase. The
income of the household at the 95th percentile also increased relative to the median; the 95/50
ratio increased from 2.66 (+/- 0.03) to 2.91 (+/- 0.02)./8/ The ratio of the household's income
at the 20th percentile to the median was unchanged from 1967 to 1980.
Derivatives of these selected percentiles are also quite prominent in income (and earnings)
inequality literature. Some researchers choose to employ alternatives such as the ratio of the
90th percentile to the 10th percentile (90/10) and the median to the 10th percentile (50/10),/9/
partly because these measures are less affected by top-coding procedures.
Figure 4 shows that the 95/20 ratio and 95/50 ratio increased from 1967 to 1980, while the 90/10
ratio and 50/10 ratio both declined. Choice of which percentile ratio to use makes a difference.
The 90/10 ratio declined slightly from 9.22 (+/- 0.03) to 9.09 (+/- 0.01) during this time.
The 50/10 ratio also fell, indicating that the household income at the lowest decile grew
relative to the median.
Summary measures of inequality can provide additional information about the household income
distribution.
Summary measures are a convenient way to examine the distribution
of income. They provide a single statistic that summarizes the proper-ties of a given income
distribution. Once computed, a summary measure can be used as the focus of research or as a
variable in a statistical model. Several of these measures exist; as noted above, one of the
most popular is the Gini coefficient. Another popular measure is the mean logarithmic deviation
of income (MLD)./10/ Like the Gini, the MLD indicates that household income inequality did not
increase from 1967 to 1980 (see Figure 5).
The Atkinson measure of income inequality is another summary measure that researchers sometimes
use in income inequality research./11/ The Atkinson index is unique relative to other measures
of income inequality in that it allows the researcher to specify the social welfare function
underlying the research. The social welfare function for most measures of income inequality,
including the Gini and MLD, is predetermined by the measure's weighting scheme. The weighting
scheme is what deter-mines a measure's sensitivity to changes in different portions of the income
distribution. For example, the Gini's weighting scheme is such that it is most sensitive to
changes in the middle of the income distribution.
By setting the social welfare function for the Atkinson index, the researcher may choose to
emphasize the lower, middle, or upper end of the income distribution. The Atkinson index's
social welfare function, which may also be interpreted as the level of inequality aversion, is
set by a parameter bounded by the limits of 0 and 1 (see the Technical Appendix). As the
parameter approaches its lower limit (i.e., as aversion declines), the Atkinson gives more weight
to the upper end of the income distribution. As the parameter approaches its upper limit, the
Atkinson measure gives more weight to the lower end of the income distribution.
Figure 6 shows the percentage change for the Atkinson index relative to 1967, calculated at
three different levels of inequality aversion (0.25, 0.50, and 0.75). Although each Atkinson
index displays a similar inequality growth pattern over time, the results in Table 4 show that
the level of observed inequality differs for each calculation. From 1967 to 1980, the Atkinson
index computed emphasizing higher incomes (e = 0.25) decreased by 2.8 (+/- 2.3) percent. The
Atkinson for median (e = 0.50) and high (e = 0.75) aversion, however, were statistically unchanged
from 1967 to 1980.
When did household income inequality increase?
Whereas the data on household in-come inequality between 1967 and 1980 are ambiguous, it is clear
that the household income distribution became increasingly unequal beginning in 1981. Although
between 1980 and 1981 the only summary measures to increase significantly were the Atkinson
(e = 0.75) and the MLD, these changes signified the beginning of a period marked by rising
household income inequality.
The 1980s have been widely characterized as a period of rising income inequality. While true,
some of the measures presented here suggest that the rise in inequality started earlier--in the
mid-1970s. While the Gini coefficient was unchanged from 1973 to 1980, the MLD index showed
substantial growth-it rose 5.6 (+/- 2.5) percent between those 2 years. From 1980 to 1986, both
the MLD and Gini measured an increase in income inequality. The Gini coefficient rose 5.5
(+/- 1.9) percent and the MLD increased by 10.9 (+/- 2.5) percent during the same period. The
Gini coefficient also increased from 1986 to 1992.
Overall, the period between 1973 and 1992 was one in which income inequality grew, with the
cumulative rise in the Gini coefficient at 9.3 (+/- 1.2) percent. The MLD grew a total of
17.2 (+/- 2.6) percent over the same period.
The aggregate shares approach also indicates growing household income inequality from 1980 to
1992. Figure 3B illustrates the net percentage change in the aggregate share of household income
received by each fifth of the income distribution. As this illustration shows, households in
the top fifth of the distribution (particularly those in the top 5 percent) increased their share
of aggregate income, while those in the bottom four-fifths lost ground. Households in the top
fifth of the distribution increased their share of aggregate income by 7.3 (+/- 3.1) percent from
1980 to 1992. During the same period, households in the lowest two-fifths experienced a sharp
decline in their share of aggregate income. The bottom fifth's share of aggregate income declined
by 11.6 (+/- 2.8) percent. Households in the second fifth lost 8.7 (+/- 2.6) percent of their
share of aggregate income, not significantly different from the loss experienced by the bottom
fifth. These changes highlight the growing gap between the country's richest and poorest
households./12/
Figure 4 also depicts growing household income inequality during the 1980s. The 90/10 ratio
increased by 9.9 (+/- 2.9) percent from 1980 to 1989. This indicates that the gap between the
richest and poorest households in the United States had increased. The 50/10 ratio increased
by 1.8 (+/- 0.85) percent over the same period, which indicates growing inequality in the
bottom half of the income distribution. The differential growth rates between the 90/10 and
50/10 ratios also suggest that the spread between the top and bottom deciles increased more
than the spread between the middle and bottom deciles. In addition, the 90/50 ratio indicates
that there was an increase in the gap between the household at the median and the household at
the top decile of 7.8 (+/- 3.7) percent from 1980 to 1989./13/
What has happened to the income distribution since 1993?
Data collected since 1993 indicate that the trend of increasing income inequality, which
characterized the 1980s, has slowed or disappeared. The share of aggregate money income received
by households in the top quintile has not experienced a significant increase since 1993.
Households in each of the lower quintiles (i.e., those below the top quintile) had roughly the
same share of aggregate income in 1998 as in 1993.
Since 1993, the Gini coefficient has not experienced a single statistically significant
year-to-year increase. Nor was the change in the Gini coefficient over the entire 1993-1998
period statistically significant. As Figure 3C shows, there was no change in the aggregate
shares either. Only one measure, the MLD, suggests that household income inequality has increased
since 1993. The MLD indicates that income in-equality grew by 4.5 (+/- 2.2) percent from 1993 to
1998.
How do taxes affect income inequality?
The Census Bureau bases official estimates of money income from the March CPS on gross, or
pre-tax, income. The Census Bureau does produce a number of experimental definitions of income
to help researchers better understand the economic status of households in the United States.
/14/ Among the experimental measures of income is post-tax household income. The Census Bureau
defines post-tax household income as total household cash income (including realized capital gains),
less taxes. We compute post-tax household income both with and without the addition of the earned
income tax credit (EITC).
The ability to measure household income inequality both pre-tax and post-tax has important public
policy implications. First, it allows researchers to examine how, if at all, taxes affect the
distribution of household income. Second, it can provide insight as to whether or not tax
changes, such as a change in the EITC, affect observed household income inequality. To measure
differences in the pre-tax and post- tax income distribution, we computed Gini coefficients on
total pre-tax and post-tax household income.
Figure 7 displays Gini coefficients for both pre-tax and post-tax income for the 1993-1998
period. Not surprisingly, the results show that post-tax household income is distributed more
equally than pre-tax household income. In 1998, the post-tax Gini coefficient for households
was 0.430 (+/- 0.01), compared with 0.456 (+/- 0.01) for total pre-tax household income. This
difference notwithstanding, the Gini indexes calculated on the post-tax household distribution
have not experienced a statistically significant year-to-year change since 1993.
What drives changes in income inequality?/15/
Researchers have tied the long-run increase in income inequality to changes in the U.S. labor
market and household composition. More highly-skilled, trained, and educated workers at the top
are experiencing real wage gains, while those at the bottom are experiencing real wage losses
making the wage distribution considerably more unequal. Changes in the labor market in the
1980s included a shift from goods-producing industries (that had disproportionately provided
high-wage opportunities for low-skilled workers) to technical service industries (that
disproportionately employ college graduates) and low-wage industries, such as retail trade.
But within-industry shifts in labor demand away from less-educated workers are, perhaps, a
more important explanation of eroding wages than the shift out of manufacturing. Other factors
related to the downward trend in wages of less-educated workers include intensifying global
competition and immigration, the decline of the proportion of workers belonging to unions, the
decline in the real value of the minimum wage, the increasing need for computer skills, and
the increasing use of temporary workers.
At the same time, changes in living arrangements have occurred that tend to exacerbate
differences in household incomes. For example, increases in divorces and separations, increases
in births out of wedlock, and the increasing age at first marriage may have all led to a shift
away from traditionally higher- income married-couple households and toward typically
lower-income single-parent and nonfamily households. Also, the increasing tendency for men with higher-than-average earnings to marry women with higher-than-average earnings may have
contributed to widening the gap between high-income and low- income households.
Whether the trend toward increasing income inequality the country has seen in the 1970s and
1980s will continue, or whether it has stopped or even reversed itself, remains to be seen.
Accuracy of Estimates
Statistics from surveys are subject to sampling and nonsampling error. All comparisons presented
in this report have taken sampling error into ac-count and meet the Census Bureau's standards
for statistical significance. Nonsampling errors in surveys may be attributed to a variety of
sources, such as how the survey was designed, how respondents interpret questions, how able
and willing respondents are to provide correct answers, and how accurately the answers are
coded and classified. The Census Bureau employs quality control procedures throughout the
production process-including the overall design of surveys, the wording of questions, review
of the work of interviewers and coders, and statistical review of reports.
The Current Population Survey employs ratio estimation, whereby sample estimates are adjusted
to independent estimates of the national population by age, race, sex, and Hispanic origin.
/16/ This weighting partially corrects for bias due to undercoverage, but how it affects
different variables in the survey is not precisely known. Moreover, biases may also be present
when people who are missed in the survey differ from those interviewed in ways other than the
categories used in weighting (age, race, sex, and Hispanic origin). All of these considerations
affect comparisons across different surveys or data sources.
Contact Martha Jones, Demographic Statistical Methods Division, dsmd_s&a@ccmail.census.gov for
the information on the source of the data, the accuracy of the estimates, the use of standard
errors, and the computation of standard errors.
Comments From
Data Users
The Census Bureau welcomes the comments and advice of data users. If you have suggestions
or comments, please write to:
Daniel Weinberg
Chief, Housing and Household
Economic Statistics Division
U.S. Census Bureau
Washington, DC 20233-8500
daniel.h.weinberg@census.gov
or contact:
Income Statistics Branch
Arthur F. Jones Jr.
301-763-3227
Arthur.Jones.Jr@census.gov
Technical Appendix
This technical appendix contains an explanation of the calculations of the income inequality
measures used herein.
Desired Properties of Summary Measures of Income Inequality
Summary measures of income in-equality should possess two important properties: scale
invariance and the principle of transfers. A measure is said to be scale in variant if a
constant applied to all incomes in a distribution does not affect the degree of inequality.
The principle of transfers, another desired characteristic of inequality measures, dictates
that a measure of income inequality rises (falls) when one transfers income from a poorer
(richer) person to a richer (poorer) person. The summary measures included in this report
are scale invariant and adhere to the principle of transfers.
The Gini Coefficient
The Gini coefficient incorporates detailed shares data into a single statistic, which
summarizes the dispersion of income across the entire income distribution. The Gini
coefficient ranges from 0, indicating perfect equality (where everyone receives an equal
share), to 1, perfect inequality (where only one recipient or group of recipients receives
all the income). Although the Gini is based on the difference between the Lorenz curve
(the observed cumulative income distribution) and the notion of a perfectly equal income
distribution, this approach can be complex to compute. A more computationally convenient
equivalent may be used.
GINI = [(2/un**2)EPSILON(iXi)] - (n+1)/n
where u is the population mean, n is the weighted number of observations, and Xi is the
weighted income of individual i, which is also weighted by individual i's rank in the
in-come distribution. The functional form is based on the work of Partha Dasgupta,
Amartya Sen, and David Starrett, "Notes on the Measurement of Income Inequality," Journal
of Economic Theory 6 (1973), pp. 180-87.
The Mean Logarithmic Deviation of Income
The mean logarithmic deviation of income (MLD) is a member of the generalized entropy
family of in-come inequality measures. Among the attributes that make the MLD an attractive
measure is its ability to measure inequality both within and between groups. In addition,
the MLD has one of the most computationally convenient functional forms of all summary
measures discussed here.
MLD = (1/n)EPSILON[log(u/Xi)]
where Xi is the weighted income of individual i and 1 is the mean income of the selected
population. See Martin A. Asher and Robert H. DeFina, "The Impact of Changing Union
Density on Earnings Inequality: Evidence From the Private and Public Sectors," Journal
of Labor Research, 18, No. 3, (Summer 1997), pp. 425-437, for an applied look at the
MLD's decomposition.
The Atkinson Index
The distinguishing feature of the Atkinson index is its ability to gauge movements in
different segments of the income distribution. Researchers can place greater weight on
changes in a given portion of the income distribution by setting the e parameter
(referred to as the level of "inequality aversion"). The Atkinson index's functional
form is:
ATKINSON = 1 - (n/1)[EPSILON[(Xi/u)^1-e]^1/(1-e)
where Xi is the weighted income of individual i and 1 is the mean income of the selected
population. The e parameter, which is bound by the limits of 0 and 1, determines the
level of inequality aversion. The Atkinson becomes more sensitive
to changes at the lower end of the income distribution as e approaches its limit of 1.
Conversely, as the level of inequality aversion falls (that is, as e approaches 0) the
Atkinson becomes more sensitive to changes in the upper end of the income distribution.
Paul D. Allison, "Measures of Inequality," American Sociological Review, 43 (December
1978), pp. 865-880, presents a technical discussion of the Atkinson measure's properties.
FOOTNOTES
/1/ The Gini index ranges from 0.0, when all families (households) have equal shares of
income, to 1.0, when one family (household) has all the income and the rest none.
/2/ Some estimates are followed by a number in parentheses which can be added and
subtracted from the estimate to calculate the upper and lower bounds of the 90-percent
confidence interval.
/3/ Part of the increase from 1992 to 1993 is due to changes in survey methodology;
see box: A New Mode of Data Collection. See U.S. Census Bureau,
Measuring 50 Years of Economic Change, Using the March Current Population Survey, P60-203
for the historical series of family income Gini coefficients.
/4/ The Census Bureau introduced computer-assisted personal interviewing (CAPI) in January
1994 to the Current Population Survey. The March 1994 supplement permitted households to
report up to $1 million in earnings, up from $300,000, and we made parallel increases in
the reporting limits for selected other income sources. Both of these changes affected
the data. One analysis of the 1993 inequality statistics suggests that the increase in the
maximum amounts that could be reported accounts for about 1.8 percentage points, or about
one-third, of the 1992-1993 increase of 5.2 percent. The contribution of the change to
CAPI to the increase in measured inequality cannot be determined, but may well bring the
share of survey methods- related changes in inequality to over one-half of the 5.2
percentage point apparent increase. See Paul Ryscavage, "A Surge in Growing Income
Inequality?," Monthly Labor Review, August 1995, pp. 51-61.
/5/ Based on the findings of Barry Bluestone, "The Impact of Schooling and Industrial
Restructuring on Recent Trends in Wage Inequality in the United States," American Economic
Review, Papers and Proceedings, May 1990, pp. 303-307, and Kevin Murphy and Finis Welch,
"Industrial Change and the Rising Importance of Skill," in Sheldon Danziger and Peter
Gottschalk (eds.) Uneven Tides: Rising Inequality in America, New York: Russell Sage
Foundation, 1993.
/6/ The difference in the percentage change in the Gini coefficients for men and women
between 1980 and 1989 is not statistically significant.
/7/ A household consists of all people who occupy a housing unit. This includes related
family members and all unrelated people. The Census Bureau also counts as households
people living alone or unrelated people sharing a housing unit as partners. People
living in group quarters are excluded.
/8/ The increase in the 95/20 ratio was not statistically different from the increase
in the 95/50 ratio.
/9/ Jared Bernstein and Lawrence Mishel, "Has Wage Inequality Stopped Growing?" Monthly
Labor Review, December 1997, pp. 3-15, is one example.
/10/ An additional summary measure of income inequality that is sometimes used in
inequality research is the Theil entropy measure, which is based on Henri Theil's
Economics and Information Theory, Chicago: Rand McNally, 1967. The Theil, like the
mean logarithmic deviation of income (MLD), is a generalized entropy measure of income
inequality. We examined the Theil entropy measure and found its results to be similar
to that of the Gini coefficient and the MLD. Table 4 presents the results of the Theil
index's computation, as well as
the results using the variance of the natural logarithm of income (VLOG); another measure
sometimes used in inequality research. For the sake of brevity, we do not formally
analyze the findings from either method in this report.
/11/ See Technical Appendix (pages 10-11) for a description of the Gini, MLD, and Atkinson
measures of income inequality.
/12/ Between 1980 and 1992, those households in the middle 60 percent of the income
distribution experienced a 5.2 percent decline in their aggregate share of household income.
/13/ There is not a significant difference between the growth rate in the 90/10 ratio and
the 90/50 ratio.
/14/ P60-200 contains a technical discussion of how the 15 experimental measures of income
are constructed. See U.S. Census Bureau, Current Population Reports, P60-200, Money Income
in the United States: 1997 (With Separate Data on Valuation of Noncash Benefits), U.S.
Government Printing Office, Washington, DC, 1998.
/15/ This section is based on Paul Ryscavage and Peter Henle, "Earnings Inequality
Accelerates in the 1980s," Monthly Labor Review, December 1990; Sheldon Danziger and Peter
Gottschalk (eds.) Uneven Tides: Rising Inequality in America, New York: Russell Sage
Foundation, 1993; Lynn A. Karoly and Gary Burtless, "Demographic Change, Rising Earnings
Inequality, and the Distribution of Personal Well-Being, 1959-89," Demography, 32, No. 3
(August 1995), pp. 379-405; U.S. Council of Economic Advisors, Economic Report of the
President, Washington, DC: U.S. Government Printing Office, February 1992, Chapter 4; U.S.
Council of Economic Advisors, Economic Report of the President, Washington, DC: U.S.
Government Printing Office, February 1995, Chapter 5; and U.S. Council of Economic Advisors,
Economic Report of the President, Washington, DC: U.S. Government Printing Office, February
2000, Chapter 1.
/16/ Hispanics may be of any race.
