to changes taking place in the labor market and to the composition of the Nation's households. 16

Another way to look at the growth in inequality over time is to compare income at selected positions in the income distribution. As table B shows, the household at the 95th percentile in 1995 had $113,000 in income, 7.8 times that of the household at the 20th percentile ($14,400). In 1967, the first year the U.S. Census Bureau began reporting on the income distribution of households, the household at the 95th percentile had about 6.3 times the income of the household at the 20th percentile.

An alternative way to look at growth in inequality is to examine the change in average real household income for each quintile (see table B). The average income of households in the top quintile grew 45 percent, from $75,344 in 1967 to $109,411 in 1995. During the 1967-to-1995 period, the average income in the bottom quintile grew by only 22 percent, from $6,827 to $8,350. Consequently, the ratio of the average income of the top 20 percent of households to the average income of the bottom 20 percent increased from 11.0 in 1967 to 13.1 in 1995.

State Income Data

Table C shows information on median household income by State. These data are being made available in response to numerous requests from data users for subnational income estimates. The CPS is designed to collect reliable data primarily at the national level and secondarily at the regional level. State estimates of income are considered less reliable and, therefore, caution should be used when interpreting these results.17 To reduce the chances of misinterpreting annual changes in State income estimates, the U.S. Census Bureau recommends evaluating changes based on 2-year moving averages.

Median income of households for States and their respective standard errors for 1993, 1994, and 1995 (in 1995 constant dollars) are shown in table C. This table also includes two 2-year averages of median household income, for 1993-1994 and 1994-1995, along with the numerical differences and percentage changes between these 2-year averages.

Based on the two 2-year averages, real median household income increased significantly for 11 States-Colorado, Illinois, Iowa, Kentucky, Maine, Mississippi, Missouri, Pennsylvania, Tennessee, Texas, and Wisconsin. Median household income did not change significantly for any of the remaining States or the District of Columbia.

16Long-run changes in living arrangements have taken place that tend to exacerbate differences in household income. Among these changes are the shift away from married-couple households and toward singleparent and nonfamily households which typically have lower incomes. See Daniel H. Weinberg, A Brief Look at Postwar U.S. Income Inequality, Current Population Reports, Series P60-191, June 1996.

17The U.S. Census Bureau expects to release more reliable biennial State estimates of household median and per capita income based on statistically modeled data beginning this fall for 1993.

When comparing the relative ranking of States, the U.S. Census Bureau recommends using 3-year averages. Use of 3-year averages reduces the chances of misinterpreting the results. Comparing income among the States using the 3-year average of 1993-1995 shows that median household income for Alaska, although not statistically different from that of Hawaii, was higher than that of the remaining 48 States and the District of Columbia. Conversely, the median household income for West Virginia, although not statistically different from the median for Mississippi and Arkansas, was lower than that of the remaining 47 States and the District of Columbia. The relative standing of the remaining States and the District of Columbia is less clear because of sampling variability surrounding the estimates.

VALUATION OF NONCASH BENEFITS

Traditionally, income data presented in Census Bureau reports have been based on the amount of money income received during a calendar year before taxes and excluding capital gains. This definition of income is narrow and does not provide a completely satisfactory measure of the distribution of income. The omission of data on taxes, capital gains, and the value of noncash benefits affects comparisons over time and between population subgroups. In the early 1980's, the U.S. Census Bureau embarked on a research program to examine the effects of noncash transfer benefits and of taxes on income distributional measures.

Estimates of tax data in this report are based on modeled data. Four types of taxes were simulated: 1) Federal individual income taxes, 2) State individual income taxes, 3) property taxes on owner-occupied housing, and 4) payroll taxes. Also, the valuation of noncash benefits such as food stamps, school lunches, housing subsidies, medicare, medicaid, employer contributions to health insurance, and net imputed return on equity in own home are presented in this section, 18

Table D shows the year-to-year changes for median household income in 1994 and 1995 for the 15 definitions of income. There were statistically significant increases in real income between 1994 and 1995 under each of the 15 definitions of income shown in this report. When comparisons are made between the official definition of income and the 14 other definitions of income, the percentage change in the official definition of income is not significantly different from the percentage changes in any of the other definitions.

Distribution Effects

Taxes, government transfers, and other benefits have substantial effects on the level of income as well as the distribution of income. These effects can be seen by

18 For more information on the methodology and procedures used to estimate taxes and to value noncash benefits see P60-186RD, Measuring the Effect of Benefits and Taxes on Income and Poverty: 1992.

examining distributional changes under the 15 different definitions of income used in this section. Text tables D, E, and F, and table 12 show data on the distribution of income under these alternative definitions.

Definition 1 is the official definition of income. It is based on money income before taxes and includes government cash transfers. Under definition 1, 3.7 percent of aggregate household income was received by the lowest quintile, 9.1 percent by the second quintile, 15.2 percent by the third quintile, 23.4 percent by the fourth quintile, and 48.6

percent by the highest quintile (see table E). In 1995, the Gini index for all households under definition 1 was .444.19

19Two methods are used in this report to estimate shares of aggregate income received by each quintile and the Gini index. The first method incorporates the use of actual sorted household data resulting in a Gini index of .450 and quintile shares of 3.7, 9.1, 15.2, 23.3, and 48.7. The second method uses grouped data and employs several interpolation routines resulting in a Gini index of .444 and quintile shares of 3.7, 9.1, 15.2, 23.4, and 48.6. The grouped data method was used throughout this report for calculating Gini indexes as they appear with other income

Definition 4 shows the effect on the income distribution when government cash transfers are deducted and capital gains and employee health benefits are added to the official income definition. This reveals a distribution of income generated by the private sector which was more unequal than the distribution under the official definition of income. (See table E.) Under definition 4, shares of income received by the lowest two quintiles of households declined from that of definition 1, while shares of income received by the two highest quintiles increased. The Gini index under this definition of income, .509, was 14.6 percent higher than the index under the official income definition (.444).

The effect of taxes on the distribution of income is shown in definition 8. Comparing estimates using definitions 8 and 4 shows the net effect of deducting Social Security payroll taxes, Federal individual income taxes, State individual income taxes, and adding the earned income tax credit (EITC). The combined effect of taxes on the Gini index was to reduce it from that of definition 4 by 5.5 percent, from .509 to .481.

Nonmeans-tested transfers had a substantial effect on reducing income inequality. The net effect of nonmeanstested cash and noncash government transfers on the

distribution of income is evident by comparing definition 11 estimates to definition 8 estimates. Including these benefits increased the share of income going to the lowest quintile, and lowered the share of income going to the highest quintile. These transfers also had a significant effect on the Gini index, lowering it by 14.3 percent, from .481 to .412.

Definition 14 shows the net effect of adding meanstested transfers by comparing it to definition 11. The share of income in the lowest quintile increased, and the share of income going to the highest quintile decreased. The Gini index declined 4.4 percent from .412 to .394.

An important finding of the U.S. Census Bureau's tax and benefit research is that government transfers have a significantly greater impact on lowering income inequality than the tax system. In 1995, subtracting taxes lowered the Gini index by 5.5 percent (from .509 to .481) while including transfers lowered the Gini index by 18.1 percent (from .481 to .394).

Table E also shows the effect of including net imputed return on home equity (definition 15). The inclusion of net imputed return on home equity had a minimal effect on the Gini index.

summary measures in the detailed tables as well as for share estimates under the alternative definitions of income. The grouped data approach is used as a more efficient (though not as exact) way to calculate the Gini

Effects on Selected Population Groups

Different income definitions result in quite different income

shown by the 1995 income data presented in table F, taxes and transfers affect income comparisons between population subgroups to varying degrees.

Under the official income definition, the ratio of Black median household income ($22,393) to White median household income ($35,766) was .63. Subtracting cash transfers and adding capital gains and health insurance supplements (definition 4) reduced the ratio to .61. The subtraction of Federal and State income taxes and payroll taxes (definition 8) results in an increase in the ratio to .66, and the addition of cash (definition 11) and noncash transfers (definition 14) results in a further increase in the Black-to-White income ratio to .69.

Based on the official income definition, the ratio of median income of Hispanic-origin households ($22,860) to White households ($35,766) was .64. A definition of income that has been broadened to include the effects of taxes and transfers (definition 14), resulted in a ratio of .71.

Differences in income ratios by income definition can be observed across different household compositions. Under the official income definition, the ratio of median income of households with a female householder, no husband present, with children, to that of married-couple households with children was .36. Based on a definition of income that includes the effect of taxes and transfers (definition 14), the ratio increased to .48.

The importance of income definitions to income differences between population subgroups is particularly evident for households with children and elderly members. Under definition 1, median money income was $40,954 for households with children under 18 years of age, and $20,503 for households with members 65 years old and over. Thus, in 1995 the ratio of median income of the latter group to the former group was .50. Subtracting cash transfers and adding capital gains and employer-provided health insurance (definition 4) lowered the ratio to .20. The subtraction of taxes (definition 8) raised the ratio slightly, to .23. The addition of cash (definition 11) and noncash transfers (definition 14) more than doubled the ratio, bringing it to .67, and adding the return on home equity (definition 15) resulted in a further increase in the ratio to .74.

CHANGES IN SURVEY METHODOLOGY

The income data in this report for calendar year 1995 are derived from the March 1996 Current Population Survey (CPS). They are the first estimates based entirely on households selected from the 1990 census-based sample design. The March 1996 sample incorporates the geographic definitions (officially released in June 1993) of metropolitan and nonmetropolitan residence from the 1990

decennial census.20 The March 1995 metropolitan/nonmetropolitan estimates shown in this report are also based on the 1990 census definitions.

Beginning with the January 1996 survey, the CPS sample was reduced by approximately 7,000 housing units for budgetary reasons. This sample reduction took place in seven States (Illinois, Massachusetts, Michigan, New Jersey, North Carolina, Ohio, and Pennsylvania), New York City, and the Los Angeles-Long Beach metropolitan area. The sample reduction affected the reliability of estimates at the National level and at the State and substate levels for those areas where sample was reduced. The reduction did not affect the reliability of those States not involved in the reduction.21

A revised edit and allocation procedure for race information was also introduced in January 1996. This new procedure assigns respondents reporting "Other (unspecified) race" in the race question to one of the four major race categories: White; Black; American Indian, Eskimo, and Aleut; and Asian or Pacific Islander. This new edit and allocation procedure was in response to the rising proportion of the CPS population reporting their race as "Other."

COMMENTS FROM DATA USERS

For the past 2 years, the data on income, poverty, and the valuation of noncash benefits were published in a single report entitled, Income, Poverty, and Valuation of Noncash Benefits. Beginning this year, income and poverty data will be published in separate reports with each report containing a section discussing the effects of the valuation of noncash benefits.

Comments received from data users related to the contents of last year's report resulted in the restoration of tables showing the following characteristics: household income by relationship, age, and sex of persons in the households; income of families by presence of related children under 18 years old and by work experience of the husband and wife; earnings of persons by work experience; and mean income of persons by source of income. Unpublished 1995 versions of tabulations previously published in the Money Income of Households, Families, and Persons in the United States reports are available at the address below for the cost of photocopying. Selected historical time-series tables are also available on the Internet (http://www.census.gov).

We are interested in your reaction to the usefulness of the information provided in this report and welcome your recommendations for improving our products. If you have

20 For detailed information on the 1990 sample design, see the Department of Labor, Bureau of Labor Statistics report, Employment and Earnings, Volume 41 Number 5, May 1994.

21 For detailed information on the sample reduction, see the Department of Labor, Bureau of Labor Statistics report, Employment and Earnings, Volume 43 Number 2, February 1996.