The demographic and income data in this report are derived from the Current Population Survey’s Annual Social and Economic Supplements (ASEC), which are conducted in March of every year. The specific files used in this report are from March 1971 to March 2021 which contain data on the annual income of households from 1970 to 2020. Conducted jointly by the U.S. Census Bureau and the Bureau of Labor Statistics, the CPS is a monthly survey of about 60,000 households and is the source of the nation’s official statistics on unemployment. The ASEC survey in March typically features a larger sample size (about 75,000 in recent years) and is the basis for the annual Census Bureau report on income and poverty

Nonresponse bias has been on the increase in the CPS in recent years and increased substantially during the coronavirus pandemic. The Census Bureau has released updated survey weights, known as entropy balanced weights, to correct for nonresponse bias in the CPS. In this report, estimates from the March 2019 to March 2021 ASEC files are derived using the entropy balanced weights.

Methodological and other revisions to the CPS may also have an impact on estimated trends. For example, the 2015 ASEC introduced a redesigned set of income questions, and definitions of key socioeconomic categories, such as race and educational attainment, have changed over time.  

The sample for this analysis is restricted to adults living in households. The CPS microdata used are the Integrated Public Use Microdata Series (IPUMS) provided by the University of Minnesota. The IPUMS assigns uniform codes, to the extent possible, to data collected in the CPS over the years. More information about the IPUMS, including variable definition and sampling error, is available at IPUMS CPS

Longitudinal analysis

The CPS data have the characteristic that, in principle, up to half the people interviewed in one year are also interviewed in the next year. Some of the analysis in this report exploits this feature to study the transitions of adults across income tiers from one year to the next over the period from 2000 to 2021.

More specifically, respondents who were in their fourth month of interviewing in the March CPS ASEC one year (Month-in-sample or MIS = 4) appear again in the March CPS ASEC the next year for their eighth and final interview (MIS = 8). Thus, respondents with MIS = 4 in March 2000 are matched to respondents with MIS = 8 in March 2001, respondents with MIS = 4 in March 2001 are matched to respondents with MIS = 8 in March 2002, and so on.

The creation and the analysis of the matched sample of adults was facilitated by the linking variables available in the IPUMS-CPS database. Additional edits were imposed to ensure that respondents who were matched from year to year were of the same sex, race and nativity. Age was allowed to advance by up to two years. Households were dropped if the identity of the household head had changed from one year to the next. See the guidance provided by the National Bureau of Economic Research (NBER) for more on the matching of CPS data over time.

The boundaries that define the income tiers in the longitudinal analysis were based on the median income of the full ASEC cross-section sample of households in each year. Thus, these boundaries vary from year to year, as in the cross-section analysis. Adults in their first year in the longitudinal sample were sorted into income tiers based on their household income in the first year and were resorted into income tiers in the second year based on their household income for that year. 

It should be noted that, except for the computation of the boundaries of income tiers, the longitudinal analysis does not use the entropy balanced weights released by the Census Bureau as those pertain only to the full sample of survey respondents in a single year. Instead, the analysis uses the longitudinal weights in the IPUMS-CPS database (LNKFW1YWT). Also, longitudinal analyses may be subject to attrition bias, the result of respondents exiting the sample prior to their second or subsequent rounds of interviews, perhaps because of an address change.

Income

Household income is the sum of incomes earned by all members of the household in the calendar year preceding the date of the survey. The CPS collects data on money income received (exclusive of certain money receipts, such as capital gains) before payments for such things as personal income taxes, Social Security, union dues and Medicare deductions. Non-cash transfers, such as food stamps, health benefits, subsidized housing and energy assistance, are not included. More detail on the definition of income in the CPS is available in the documentation of the data. It should be noted that income data in the CPS public-use microdata files are top-coded to prevent the identification of a few individuals who might report very high levels of income. 

The data on income and wealth are adjusted for inflation with the Consumer Price Index Research Series (CPI-U-RS) of the Bureau of Labor Statistics (BLS) as published in the Census Bureau’s income and poverty report. This is the price index series used by the Census Bureau to deflate the data it publishes on household income. Since 1978, this is the CPI-U-RS index as published by the BLS. For years prior to 1978, the Census Bureau made its own adjustment to the CPI-U to approximate the trend in the CPI-U-RS. 

The choice of a price index does not affect the allocation of households into lower-, middle- or upper-income categories at a point in time. That is because the same price index applies to the incomes of all households and does not affect their income-based rank. However, the choice of a price index does affect measures of absolute progress over time. For example, from 1970 to 2020, the price level rose either 567% (CPI-U) or 497% (CPI-U-RS). This means that someone earning $10,000 per year in 1970 would be just as well off in 2020 earning either $66,700 (using the CPI-U) or $59,700 (using the CPI-U-RS). 

When examining trends in economic indicators over time, it is generally desirable to avoid comparisons across different points of the business cycle. The income comparisons in this study are principally based on income data pertaining to 1970, 2010 and 2020. These dates encompass or are very near periods of recessions (December 1969 to November 1970, December 2007 to June 2009, and February 2020 to April 2020).  

Households and families in Census data

The Census Bureau defines a household as the entire group of persons who live in a single dwelling unit. A household may consist of several persons living together or one person living alone. It includes the household head and all of their relatives living in the dwelling unit and also any lodgers, live-in housekeepers, nannies and other residents not related to the head of the household. 

By contrast, a family is composed of all related individuals in the same housing unit. Single people living alone or with two or more adult roommates are not considered families according to the Census Bureau approach. In the vast majority of cases, each housing unit contains either a single family or single person living alone. In the case of roommates, one person is designated as the “householder” (usually whoever owns the unit or in whose name the lease is held), and the other person or persons are designated as secondary individuals. In a few cases, there are households with families in which neither adult is the householder. These families are designated as either related or unrelated subfamilies, depending on whether one of the adults is related to the householder. 

Race, ethnicity, educational attainment, and marital status

In this report, White, Black and Asian adults are non-Hispanic. Hispanics are of any race. Asians include Native Hawaiian and Other Pacific Islanders.

“High school graduate” refers to those who have a high school diploma or its equivalent, such as a General Education Development (GED) certificate, and those who had completed 12th grade, but their diploma status was unclear (those who had finished 12th grade but not received a diploma are excluded). Adults with “some college” include those with an associate degree and those who attended college but did not obtain a degree. In the estimates for 1971, adults with a bachelor’s degree or higher level of education are those who completed at least four years of college.

“Unmarried” includes married (spouse absent), never married, divorced, separated and widowed. “Married” refers to opposite-sex couples only in 1971 but includes same-sex couples in 2021.

Adjusting income for household size

Household income data reported in this study are adjusted for the number of people in a household. That is done because a four-person household with an income of, say, $50,000 faces a tighter budget constraint than a two-person household with the same income. In addition to comparisons across households at a given point in time, this adjustment is useful for measuring changes in the income of households over time. That is because average household size in the U.S. decreased from 3.1 persons in 1970 to 2.5 persons in 2020, a drop of about 20%. Ignoring this demographic change would mean ignoring a commensurate loosening of the household budget constraint. 

At its simplest, adjusting for household size could mean converting household income into per capita income. Thus, a two-person household with an income of $50,000 would have a per capita income of $25,000, double the per capita income of a four-person household with the same total income. 

A more sophisticated framework for household size adjustment recognizes that there are economies of scale in consumer expenditures. For example, a two-bedroom apartment may not cost twice as much to rent as a one-bedroom apartment. Two household members could carpool to work for the same cost as a single household member, and so on. For that reason, many researchers make adjustments for household size using the method of “equivalence scales.” 

A common equivalence-scale adjustment is defined as follows: 

Adjusted household income = Household income / (Household size)N

By this method, household income is divided by household size exponentiated by “N,” where N is a number between 0 and 1. 

Note that if N = 0, the denominator equals 1. In that case, no adjustment is made for household size. If N = 1, the denominator equals household size, and that is the same as converting household income into per capita income. The usual approach is to let N be some number between 0 and 1. Following other researchers, this study uses N = 0.5. In practical terms, this means that household income is divided by the square root of household size – 1.41 for a two-person household, 1.73 for a three-person household, 2.00 for a four-person household and so on. 

One issue with adjusting for household size is that while demographic data on household composition pertain to the survey date, income data typically pertain to the preceding year. Because household composition can change over time, for example, through marriage, divorce or death, the household size that is measured at the survey date may not be the same as that at the time the income was earned and spent. 

Once household incomes have been converted to a “uniform” household size, they can be scaled to reflect any household size. The income data reported in this study are computed for three-person households, the closest whole number to the average size of a U.S. household since 1970. That is done as follows: 

Three-person household income = Adjusted household income * [(3)0.5

Adjusting for household size does have an effect on trends in income since 1970. However, it is important to note that once the adjustment has been made, it is immaterial whether one scales incomes to one-, two-, three- or four-person households. Regardless of the choice of household size, the same results would emerge with respect to the trends in the wellbeing of lower-, middle- and upper-income groups.

The work-experience unemployment rate

The U.S. Bureau of Labor Statistics (BLS) defines the work-experience unemployment rate as “the number of persons unemployed at some time during the year as a proportion of the number of persons who worked or looked for work during the year.” This estimate is based on data collected in the March CPS ASEC surveys on the work activities of respondents during the previous calendar year. Thus, the 2021 ASEC has data on people’s work activities in 2020.

The widely reported unemployment rate has a reference period of only one week, i.e., it is based on the work activities of respondents during the reference week of the basic monthly CPS. Because the March CPS ASEC has a one-year reference period, the number of persons it measures as having some employment or unemployment is greater than the number measured in a typical monthly CPS. For example, according to the BLS, the work-experience estimate shows 26.4 million people had a spell of unemployment in 2020. But the annual average of the monthly estimates of unemployment from the basic CPS was 12.9 million.

This analysis follows the BLS definition for estimating the work-experience unemployment except for two differences. First, the sample in this report is adults living in households. The BLS sample consists of people ages 16 and older, including those in group quarters. Second, in this report people who worked 50 or more weeks in a year are counted as not having experienced unemployment. The BLS counts anyone who worked less than 52 weeks during the year as unemployed as long as they also reported looking for work or being on layoff during the year.

Statistical significance

Comparisons between estimates are tested for statistical significance using the replicate weights in the CPS ASEC data. For 2019 through 2021, the replicate weights for a respondent are adjusted by the ratio of the entropy balanced weight for that respondent to the unadjusted March supplement weight. All tests for statistical significance are conducted using 95% confidence intervals.