Political debate on income in the United States often has been
characterized as competition between two schools of economic
thought: one that focuses on the long-term increase in general
prosperity and one that focuses on the equalization of existing
incomes. Proponents of the first approach have much to hearten
them; the long-term increase in economic well-being in the United
States has been enormous. Today, the standard of living for the
average American is nearly seven times higher than it was 100 years
ago, after adjusting for inflation. The large gains in
prosperity have affected all Americans, including low-income
groups. At present, workers earning the minimum wage comprise the
lowest-paid 2 percent of all employees. Yet today's minimum wage
worker earns more, in real terms, in a single day than a
low-skilled worker earned in an entire six-day workweek at the turn
of the century. In other words, today's minimum wage worker earns
more in eight hours than a low-skilled worker earned in 70 or more
hours a century ago.
The Census analysis appears lucid and
straightforward. However, the Census data are marred by four
problems that lead to an overstatement of the level of economic
inequality. These problems are:
STAGE 3: Adjusting Quintiles to Contain
Equal Numbers of Persons.
The largest flaw in the Census income distribution data is that
its income "quintiles" do not contain equal fifths of the U.S.
population, but are in fact unequal in size. Indeed, in reality the top
Census "quintile" contains not 20 percent of the population but
24.3 percent, while the bottom quintile contains only 14.8 percent
of the population. The top quintile has 65 percent more persons
than does the bottom quintile. With conventional Census figures,
the bottom "quintile" is hollow, representing far less than
one-fifth of society; by contrast, the top "quintile" is
overpopulated, containing far more than one-fifth of persons,
workers, and work effort. Naturally, the demographic imbalance
between the quintiles has a considerable effect on the apparent
income imbalance between them.
Stage 3 uses the comprehensive post-tax
income data developed in Stage 2 and then makes a demographic
adjustment so that each income quintile in fact contains one-fifth
of the population. This adjustment ensures
that the economic status of each individual in the population is
treated as having equal value or importance. By contrast,
individuals are not treated equally in the current Census methods;
in general, individuals in married couple families are
underrepresented by the Census data and treated as less significant
than single persons or people in single-parent families.
The effects of the Stage 3 demographic
corrections are shown in Chart 2. The share of income of the
adjusted bottom quintile rises to 9.4 percent, while the income of
the top quintile falls to 39.7 percent. The adjustment of the
underreporting of income received by the lowest quintile of the
population is particularly important. With 9.4 percent of total
income, the actual share of income for this quintile is nearly
three times higher than the conventional Census figures show.
STAGE 4: Explaining the Remaining
Variance--Hypothetical Equalization of Work Performed.
Even after the quintiles are adjusted to contain equal numbers of
persons in Stage 3, there remains an enormous difference in the
amount of work performed within each corrected quintile. The annual
number of hours of employed labor in the top quintile is still
nearly twice that in the bottom quintile. This imbalance in work
certainly can be expected to contribute to an imbalance in
income.
Stage 4 analyzes the effects of the
imbalance of work on the distribution of income. It incorporates changes
from Stage 2 and Stage 3 and then makes a hypothetical adjustment
so that working age adults (ages 18 to 64) in each quintile are
assumed to all perform the same average number of hours of paid
work.
This adjustment naturally reduces the work performed and earnings
in higher quintiles and increases work and earnings in the lower
quintiles. Chart 3 shows the hypothetical distribution of income
that would occur if working age adults in each quintile performed
the same average number of hours of annual paid labor. The share
of income for the bottom quintile rises from 9.4 percent to 12
percent, while the share of the top quintile falls from 39.7
percent to 36.7 percent.
Comparison of the Top and Bottom
Quintiles.
These adjustments make a great difference in the measure of
apparent income inequality. For example, under conventional Census
figures (Stage 1), the top "quintile" accounts for some $2.5
trillion in income in 1997, while the bottom quintile has only $181
billion. Thus, the top quintile is shown as receiving $13.86 in
income for every $1.00 in the bottom. However, once incomes are
more completely counted and taxes are considered (in Stage 2), the
ratio drops considerably--to $8.05 for every $1.00 of income.
But
even this lower ratio continues to reflect the fact that the Census
data's top "quintile" is seriously overpopulated, while the bottom
is underpopulated. Once the quintiles are adjusted to contain equal
numbers of persons, the ratio of incomes of the top to the bottom
quintile drops to $4.23 to $1.00 (as shown in Chart
4). Moreover, even this difference is due in large part to the
fact that working age adults in the top quintile work twice as many
hours as those in the bottom. If such adults worked the same number
of hours, the income ratio would fall to around $3.07 to $1.00.
Comparison of the Top and Bottom
Halves.
Chart 5 shows similar data for
the top and bottom halves of the population. According to
conventional Census measurement methods, the top half of society
received $4.1 trillion, or 81 percent of total income, in 1997. The
bottom half of society, by contrast, received $973 billion, or 19
percent of the total. As stated previously, the Census figures
exclude major types of income and compensation and ignore taxes.
Even more critically, under Census procedures, the top "half"
contains not 50 percent of the population but 57.8 percent. The
Census Bureau's top "half" contains 63 percent of working age
adults who, in turn, perform 71 percent of the paid labor in the
economy.
With
a more accurate count of post-tax incomes and an adjustment so that
the top half contains 50 percent of the population, the annual
income received by the top half falls to $3.2 trillion while the
share of the bottom half rises to $1.4 trillion. Thus, the
conventional Census figures overrepresent the income available to
the more affluent half of society by nearly $1 trillion. The share
of total income received by the top half falls from 81 percent to
70 percent.
As
Chart 6 shows, the Census Bureau
represents the top half of society receiving $4.24 in income for
every $1.00 received by the bottom half. In reality, the correct
figure is $2.28 for every $1.00. The real level of inequality in
the economy is effectively half that represented by the
conventional Census figures.
The
revised level of income equalization in the United States is quite
surprising. Even after the Stage 3 population adjustments, the more
affluent half of the population still provides 59.5 percent of the
hours of work in the overall economy. Moreover, the top half
contains the bulk of the most skilled and productive laborers and
provides most of the vital investment in plant equipment, which is
necessary to sustain the prosperity of all Americans. Given these
realities, the 70 percent share of post-tax income going to the
most affluent half of society seems remarkably low; it is striking
evidence of the high degree of income equalization already
occurring in American society.
DETAILED
ANALYSIS:
UNDERREPORTING OF INCOME AND OMISSION OF TAXES
The
conventional Census income distribution data are based on the
concepts of "money income." Money income includes earnings,
interest, dividends, rents, Social Security retirement benefits,
pension or retirement income, survivors benefits, disability
benefits, veterans benefits, workers' compensation, alimony, and
some cash welfare benefits. Despite this list, it is now widely
acknowledged that the Census Bureau's money income figures grossly
underreport the economic resources available to Americans. For
example, the aggregate "money income" figures reported by the
Census Bureau in 1996 equaled only 70 percent of the comparable
personal income figures reported in the U.S. Department of
Commerce's National Income and Production Accounts (NIPA) that
serve as the basis for measuring the gross national product.
The
Census Bureau's annual Current Population Survey (CPS), which
serves as the basis for its income distribution data, collects data
on the receipt of many additional types of income beyond those
included under "money income." These additional income data,
however, are excluded from Census's official income distribution
figures, which are based on money income only. The Census Bureau
does publish data using expanded concepts of income in technical
tables in some publications; however, these tables, which offer 17
alternative definitions of income, are bewildering even to
professionals in the field. Yet in its texts describing inequality,
and in briefing materials given to the press, the Census Bureau
continues to promote figures based on limited "money income." As a
result, nearly all discussions of income inequality in the popular
media and among policymakers and government officials rely on data
that can be misleading.
Fortunately, the additional income data
collected in the Current Population Survey are made available to
researchers in electronic form, and we have used these data as the
basis for the analyses provided in this report.
Table
1 shows the effects of incorporating a more complete count of
income and taxes. (This is the same as the Stage 2 adjustment made
earlier, except that the adjustments are shown in greater detail.)
First, capital gains and losses are added (Stage 2A). This
adjustment raises total annual income by some $200 billion and
increases income inequality. Next, employee health benefits and
government transfers are added (Stage 2B). Government transfers
include the earned income tax credit, food stamps, school lunch
programs, public housing, Medicaid, and Medicare. Medicaid and
Medicare benefits are counted at their insurance or market value,
which equals the average government expenditures on benefits to
individuals in specific age and risk categories. These adjustments
add nearly $500 billion to the total annual income and decrease
income inequality. Finally, the effects of federal income tax,
state income tax, property taxes, and Social Security taxes are
shown in Stage 2C. This adjustment reduces annual total income by
some $1.2 trillion and markedly decreases inequality. We have
termed these figures in Stage 2C "comprehensive post-tax income."
They are the same as the completed Stage 2 figures presented in
Chart 2 and elsewhere in this report.
Accompanying Table
DETAILED
ANALYSIS:
POPULATION AND INCOME DISTRIBUTION
When
decisionmakers, journalists, and the public view the government's
official income distribution figures, there is a common and
implicit assumption that the quintiles contain equal shares of the
population. After all, the notion that we should measure
"inequality" by comparing the aggregate incomes of groups that are,
themselves, unequal in size is at best confusing. However, as
noted, the official Census income "quintiles" do not contain equal
shares of the population, and this fact skews the Census Bureau's
measure of income distribution.
No
one would think it valid to measure inequality between New York
State and Delaware by simply comparing the aggregate incomes in the
two states. In such a comparison, income differences would mainly
reflect vast differences in state populations. But the Census
Bureau makes precisely this sort of unbalanced comparison whenever
it compares quintiles of unequal size.
Chart
7 shows the percent of the population contained within each
Census "quintile." While the middle quintile does contain roughly
one-fifth of the population, the others do not. The high disparity
in population between the highest-income and lowest-income
quintiles is of particular interest. While the top quintile
contains 24.3 percent of the population, the bottom quintile
contains only 14.8 percent. In raw numbers, there are 64.2 million
persons in the top quintile, compared with 39.2 million in the
bottom quintile. Thus, for every person in the lowest quintile,
there are 1.64 persons in the top. This imbalance in population is
a major factor contributing to the apparent levels of inequality in
Census Bureau figures.
The
Census Bureau quintiles are unequal in size because they are based
on a count of households rather than persons. A household is
defined as a person or group of persons living in a single housing
unit. In the United States, high-income households tend to be
married couples with many members and earners. Low-income
households tend to be single persons with little or no earnings. It
should be no surprise, then, that the average household in the
Census Bureau's top quintile contains 3.1 persons, while the
average household in the bottom quintile contains 1.9 people.
Overall, 54.9 percent of the households in the bottom quintile have
only one person compared with 7 percent in the top quintile.
Although the disparity in the population
sizes of the Census quintiles is striking, an analysis of the types
of individuals in each quintile reveals even greater disparity. Chart
8 shows the number of people in each official quintile divided
into age categories: children (under 18), elderly (over age 64),
and working age adults (ages 18 to 64). The elderly comprise about
one-tenth of the total population. Elderly persons are generally
retired and thus tend to have lower incomes than families headed by
working adults. It should be no surprise, then, that the lowest
three official quintiles contain the bulk of elderly persons.
Children, by contrast, are more abundant in the higher-income
quintiles. For example, the top two quintiles contain some 34
million children, compared with 24 million in the bottom two
quintiles.
However, the greatest differences occur
among working age adults. The highest official quintile has 2.4
working age adults for each such adult in the bottom quintile. In
fact, the 44.1 million working age adults in the top quintile by
themselves outnumber the entire population (adults, elderly, and
children combined) of the bottom quintile. The number of working
age adults in the top quintile alone is greater than the number of
such adults in the lower two quintiles combined.
The
high-income and low-income quintiles constructed by Census differ
radically in population and age as well as family structure, which
significantly affects the amount of income in each quintile. The
Census practice of measuring inequality by comparing aggregate
incomes between "quintiles" that contain widely differing numbers
of persons can be extremely misleading. A far clearer picture of
income inequality can be obtained by adjusting the quintiles so
that each actually contains 20 percent of the population. The large
effects of equalizing the number of persons within each quintile
(in Stage 3) are shown in Chart 9. Natural differences
between the quintiles still exist; the bottom quintile has more
elderly persons and fewer working age adults than the other
quintiles. But these differences are quite modest compared with
those shown in Chart 8.
It
appears obvious that the quintiles shown in Chart 9 offer a fairer
basis for comparing income equality that the official unbalanced
"quintiles" in Chart 8. To a large degree, the relative poverty of
the Census Bureau's official bottom quintile shown in Chart 8
results from the simple lack of people within the quintile rather
than from economic factors. By contrast, differences in incomes
between the quintiles in Chart 9 will be the result mainly of
economic factors rather than of mere differences in the size of the
quintiles.
Rich
and Poor, Married and Unmarried. One frequently overlooked
dimension of the gap between the "rich" and the "poor" is how much
it is affected by marital status. As Chart 10 shows, only about 30
percent of all persons in Census's bottom quintile live in married
couple families; the rest either live in single-parent families or
reside alone as single individuals. In the top quintile, the
situation is reversed: Some 90 percent of persons live in married
couple families. In this case, equalizing the numbers of persons
within the quintiles makes little difference; even after each
quintile is adjusted to contain the same number of persons, 85
percent of persons in the top quintile continue to live in married
couple families compared with one-third in the bottom.
The
prevalence of marriage in the higher quintiles and its near absence
in the bottom quintile should not be a surprise. Marriage provides
the opportunity to bring two incomes into the home. Equally
important, married parents tend to have higher levels of ability
and skill than do non-married parents. This is particularly true in
the case of never-married mothers. Today, one child in three is
born out of wedlock to mothers who have, on average, very low
levels of math and verbal ability. The collapse of marriage among
the less capable members of society has tended to magnify
pre-existing tendencies toward inequality. Research by Robert I.
Lerman of the Urban Institute has shown that half the increase in
income inequality in recent years is a product of the growth of
single parenthood.
DETAILED
ANALYSIS:
INEQUALITY OF INCOME AND INEQUALITY OF WORK
As
noted, the official Census Bureau income quintiles contain unequal
shares of the population. However, even greater inequality results
from the amount of work performed within each quintile. Chart 11 displays the official
Census quintiles again. It shows both the percentage of working age
adults (ages 18-64) in each quintile and the percentage of total
hours of work performed by the quintile. The bottom official
quintile contains only 11.5 percent of working age adults and only
5.6 percent of all hours of work performed in the economy in 1997.
By contrast, the top quintile contains 27.6 percent of working age
adults and nearly one-third of all the hours of labor performed.
There are nearly five hours of paid work performed in the Census
top quintile for every hour of work performed in the bottom
quintile.
Thus, not only do the lower-income quintiles have
fewer working age adults, but each adult on average performs
significantly fewer hours of work than his counterparts do in the
higher quintiles. Chart 12 shows the average
number of hours of work per week per working age adult for each
quintile. While there are 14.4 hours of work performance for each
working age adult in the bottom quintile, the comparable number in
the top quintile is 34.6 hours. On average, non-elderly adults in
the Census Bureau's top quintile tend to perform almost three times
as much labor as those in the bottom quintile.
Chart 13 shows similar data
after the quintiles are adjusted to contain equal numbers of
persons (Stage 3). The share of working age adults in the bottom
quintile rises dramatically from 11.5 percent to 18.5 percent. The
share of work performed in the bottom quintile more than doubles,
rising from 5.6 percent to 13.1 percent. These large changes
underscore the degree to which apparent inequality is a direct
result of the arbitrary population imbalance between Census
quintiles.
Of
course, even after the quintiles are adjusted to contain equal
numbers of persons, large differences in the amount of work
performed remain. As Chart 13 shows, the amount of hours of work in
the top quintile is nearly twice that in the bottom quintile. This
is, in part, a result of the fact that the top quintile still
contains roughly one-fifth more working age adults than does the
bottom quintile, even after the Stage 3 demographic adjustment.
Even more important, however, is the continuing difference in the
average number of hours worked by adults. After the Stage 3
adjustment, non-elderly adults (ages 18-64) in the top quintile
work, on average, 34 hours per week compared with 21 hours in the
bottom quintile (see Chart 14).
In addition, the workers in the top quintile tend to be more
highly skilled and better paid. The average education level of
working age householders in the top quintile is four years greater
than those in the bottom quintile. Thus, income inequality in the
United States is intensified by the fact that more highly skilled
and more productive workers tend to work more while low skilled
workers work less.
CONCLUSION
An
accurate measurement of income distribution should meet three
criteria:
-
It should utilize the most accurate and
complete income data available.
-
It should take into account the effects of
taxes.
- It should treat all persons as having
equal value and importance within the system of measurement.
The
conventional Census Bureau measurement of income distribution fails
on all three tests of accuracy.
Of
particular importance is the fact that Census does not treat all
persons equally, but "weights" its data to give far greater
significance to some persons than it does to others. When
decisionmakers, journalists, and the public view Census income
distribution figures, most will assume implicitly that the
so-called quintiles contain equal shares of the population. After
all, the idea that we should measure "inequality" by comparing the
total incomes of groups that are themselves substantially unequal
in size is, at best, perplexing. But the Census quintiles do not
contain equal numbers of persons. The lowest income quintile is
significantly underpopulated while the top quintile is
overpopulated. This fact dramatically skews the apparent
distribution of income, making it appear less equal in the United
States than it actually is. Moreover, the critical fact that the
quintiles do not contain equal numbers of persons is not revealed
in Census reports.
The
limitations in the Census measurement of income distribution lead
to a considerable exaggeration of income inequality. According to
normal Census data, the top quintile of society in 1997 had $13.86
of income for every $1.00 received by the bottom quintile. However,
if incomes and taxes are counted more completely, and if the
quintiles are adjusted to contain equal numbers of persons, then
the ratio of the incomes of the top to the bottom quintile drops to
$4.23 to $1.00. Moreover, the remaining difference is due in a
large part to the fact that working age adults in the top quintile
work almost twice as many hours, on average, as those in the bottom
quintile. If such adults worked the same number of hours, the ratio
of incomes would fall to around $3.18 to $1.00.
Differences in income in the United States
are the natural result of vast differences in ability and behavior
between individuals. In general, those persons at high income
levels tend to be married, to work large numbers of hours per year,
to have high levels of skill and productivity, and to provide
higher levels of savings and investment necessary to sustain the
overall prosperity of the economy. By contrast, individuals in the
lowest income quintile tend generally to be non-married, to work
little, and to have lower levels of skill and productivity. Despite
these factors, the average per capita income within the bottom
quintile remains over $8,000 per year, which is slightly higher, in
inflation-adjusted terms, than the average per capita income in the
whole society at the beginning of World War II.
Robert E.
Rector is Senior Research Fellow at The Heritage
Foundation. Rea S. Hederman,
Jr. is a Policy Analyst in the Center for Data Analysis at
The Heritage Foundation.
Methodological
Appendix
This
paper examines the distribution of income and income inequality
with data extracted from the March Current Population Survey of
1998. Like the Census Bureau, this report studies income at the
household level. Group quarters are not included in this survey.
The authors also used data extracted from the Internal Revenue
Service's Public Use File for 1995 (IRS SOI), containing a sample
of actual tax returns designed to replicate the total tax returns
received by the IRS. In general, this study does not account for
the underreporting of income to the Census Bureau.
Post-Tax Income.
Comprehensive post-tax income includes money income plus realized
capital gains, the earned income tax credit, employer-provided
health insurance, school lunch benefits, food stamp benefits,
government rent subsidies, and Medicaid and Medicare benefits.
Federal and state income taxes, payroll taxes, and property taxes
are subtracted. Income variables that were only given at the family
or person level were aggregated to the household level. Thus, all
FICA taxes paid by a household were added together by person and
then subtracted from the household's final income. Food stamps and
other family-level income data were treated in the same manner.
"Comprehensive post-tax income" is very
similar to the Census income "definition 14," as described in the
Census Bureau's Current Population Reports, Income, Poverty and
Valuation on Noncash Benefits, except that it employs the basic
market or insurance value for Medicaid and Medicare without the
"fungible" adjustment. The insurance value of
Medicaid and Medicare (also called the market value) equals the
average net government outlay for persons of a specific risk class
within a given state. The risk classes used are elderly, disabled
persons, non-disabled adult, and non-disabled child. Under this
approach, the value of Medicaid or Medicare equals the average cost
to the government of medical services provided to a given class of
persons; it does not report specific medical expenditures for
particular individuals.
The
fungible method of valuing Medicare and Medicaid begins with the
insurance value of benefits but then alters the values based on the
family's income class. The full insurance value is assigned to
benefits received by the middle class, but a lower value or zero
value is assigned when the same benefits are received by a
low-income household. The fungible adjustment was devised for the
measurement of poverty, not income distribution. In measuring
poverty, it is used to determine whether a household's income
should be considered above the poverty threshold. However, the
fungible adjustment, which deliberately reduces the value of
benefits received by low-income groups, is not appropriate for the
measure of income equality that seeks to compare the economic
resources of one household relative to others. The fungible
adjustment results in a substantial undercounting of government
transfers to low-income groups.
Top Coding.
An adjustment was made to compensate for the Census Bureau's "top
coding" restriction. Top coding limits the maximum value of capital
gains reported in the CPS to $99,999. With normal CPS data, capital
gains values that exceed this limit are simply reported as $99,999.
In order to obtain a more thorough estimate of high levels of
capital gains income, we have replaced those capital gains values
subject to the top coding restriction with higher values taken from
Internal Revenue Service data. This adjustment was made in the
following manner: The 1995 Statistics of Income file of the IRS was
used to determine the mean amount of capital gains income for those
returns which reported capital gains income above $99,999. This
value was adjusted to 1997 dollars and substituted for each of the
CPS capital gains values subject to the top code restriction. These
adjustments mainly increase reported incomes in the top quintile.
There also are some other top coding problems, notably limits on
the amounts of earnings and taxes reported. However, no other top
coding adjustments were made in this study.
Ranking.
Adjustments to income in Stage 2 were performed at the level
of individual households. After each adjustment, the households
were re-ranked based on their new income figures. Households then
were weighted according to the CPS household weight variable. The
Stage 4 adjustments are more general; earnings were adjusted at the
quintile level based the aggregate earnings and average labor data
within the quintile.
Additional Missing Income.
Although the comprehensive income figures shown in Table
1 are a substantial improvement over conventional Census money
income data, they still fall short of real income in the United
States economy. This shortfall is due to serious underreporting of
incomes in the basic annual Census survey instrument, the Current
Population Survey (CPS). Even the most comprehensive measure of
pre-tax income from the CPS, which reaches $5.77 trillion (in Stage
2B), still falls short of personal income figures in the Commerce
Department's National Income and Product Accounts (NIPA) by some
$1.5 trillion.
Clearly, the incorporation of this
additional unreported income could skew the measure of income
distribution significantly. Correction for this additional
underreporting is beyond the scope of the current analysis but will
be the subject of future research at The Heritage Foundation. At
present, we can only list the types and magnitudes of unreported
income and offer tentative suggestions on their impact on income
distribution. The largest amount of income unreported in the CPS is
some $900 billion in interest, dividends, and rent, which would
accrue disproportionately to the higher-income and middle-income
quintiles. However, some $300 billion in government transfers and
benefits is also unreported; these funds would be concentrated in
the lower two quintiles. Some $300 billion in self-employment
income is unreported; this shortfall is mainly income in the
informal service sector and would accrue largely to the lower half
of the population. Finally, there is over $150 million in pension
and retirement income that is reported to the IRS but does not
appear in the CPS; this would accrue largely to the lower and
middle quintiles. Clearly, there is substantial unreported income
in both the top and bottom halves of the income distribution. If
all this income were reported accurately in the Current Population
Survey, it is uncertain whether this would significantly raise or
lower the levels of inequality reported in this paper.
Accompanying
Tables
For accompanying
tables, please click here for the PDF
file.
David Mulhausen provided valuable
assistance in the preparation of this report.