Department of Labor Logo United States Department of Labor
Dot gov

The .gov means it's official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you're on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Consumer Expenditure Surveys
November 2023
CONSUMER EXPENDITURE SURVEYS
PROGRAM REPORT SERIES

Distribution of U.S. Personal Consumption Expenditures Using Consumer Expenditure Surveys Data: Methods and Supplementary Results

By Thesia I. Garner, Robert S. Martin, Brett Matsumoto, and Scott Curtin1

1. Overview

The distribution of economic well-being has been the focus of researchers and policy makers for many years, with economic well-being most often defined in terms of income, expenditures or consumption, and wealth. This note outlines a method developed by the U.S. Bureau of Labor Statistics (BLS) to distribute personal consumption expenditures (PCE) as a counterpart to the BEA distribution of Personal Income (PI) product by Fixler et al. (2017, 2020) and Gindelsky (2021). This document also includes some analysis that supplements the tables posted to the BLS webpage. For an earlier version of the analysis with more details and background information, see Garner, et. al. (2022).2 All results in this report are based on PCE data released on September 28, 2023.

The overall strategy is to use Consumer Expenditure Surveys (CE) microdata to describe the distribution of PCE across households in the U.S., or in the case of the CE, consumer units. Doing this requires not only matching comparable categories across statistical series, but also adjusting less-comparable CE categories of expenditures so that they better match PCE definitions, deleting expenditure items that are out-of-scope for PCE, and imputing expenditures for items that are out-of-scope for CE. The base of the analysis is the quarterly CE Interview Survey, which covers about 95% of CE spending (based on 2019 data). For some categories, the CE Diary is either the sole (e.g., postage stamps, non-prescription drugs) or more reliable (e.g., clothing and footwear) source of expenditure reports. A statistical matching procedure imputes these Diary expenditures to the CE Interview sample. Once a set of comparable and augmented CE data are assembled, PCE spending shares by deciles of equivalized total expenditures are computed as well as distributional statistics (such as the ratio of the 90th to the 10th percentile and Gini coefficient) for total expenditures and total equivalized expenditures. Equivalized expenditures are calculated by dividing consumer unit total expenditures by the square root of the number of people within the consumer unit.

The process of combining CE and PCE data to create distributions of PCE is summarized as follows.

  1. Map CE to PCE product categories.
  2. Match CE Diary to Interview.
  3. Impute expenditures which are out-of-scope to CE (e.g. some health care).
  4. Restrict CE sample to minimum of two interviews.
  5. Annualize CE survey expenditures.
  6. Apply Pareto adjustment to top 5% of total spending distribution to mitigate understatement of inequality.
  7. Scale CE estimates upward to match PCE by major product.
  8. Create adult-equivalized PCE and compute deciles and other statistics.

The following section describes many of the inherent issues and challenges in further detail.

2. Data and Methods

A first step in producing the distributional statistics is to adjust data from both the CE and PCE. This is necessary as there are differences in the CE and PCE that are based on the purpose of each and thereby the populations covered.3 The CE is designed to collect out-of-pocket spending on goods and services by consumer units living in the U. S. in noninstitutional setting (with the exception of students living in college or university housing). In contrast, the PCE reflects purchases of goods and services made by and on behalf of households and by non-profit institutions serving households (NPISHs). For this study we distribute PCE, as defined by Major Type of Product (NIPA Table 2.3.5) across consumer units using data from the CE. We only explicitly distribute the components of the PCE that exclude expenditures by NPISHs to more closely match the types of expenditures and population covered by the CE. 4 However, to match the PCE tables, we publish topline PCE estimates (including means, medians, and shares) which have NPISH spending included. To get these aggregates, the NPISH spending is assumed to be distributed the same as the household consumption expenditure portion of PCE, so as to not affect the overall PCE distribution. While this assumption is strong, the CE data do not include data on receipts of such benefits. 

The PCE covers a broader set of goods and services than does the CE, with the focus on expenditures for what is produced within a year in line with the National Income and Product Accounts (NIPA). The CE reflects consumer unit purchases from private and public establishments, and in addition, purchases from other consumer units through household-to-household transactions (e.g., for used cars). In contrast, the PCE reflects purchases from private businesses, costs incurred by NPISHs in providing goods and services on behalf of households, and purchases financed by third-party payers on behalf of households. Third-party payer expenditures include those for employer-paid health insurance, medical care financed through government programs, and financial services (such as banking services) that benefit households but for which they do not pay directly. 5 However, in contrast to the CE, household-to-household transactions are excluded from the PCE.

2.A. CE Sample Selection

The CE comprises two surveys, an Interview and Diary, each with its own sample of consumer units. The Interview Survey has a three-month recall period and is designed to cover larger expenditure purchases (e.g., major appliances) and recurring items (e.g., rent, utilities). Consumer units are interviewed once every three months for up-to four consecutive quarters on a rolling basis. In contrast, the Diary is designed to cover smaller expenditure, such as prescription drugs, and frequently purchased items, such as detailed food. For these goods and services, data are collected over two consecutive weeks using one-week diaries.

The Interview and Diary overlap in coverage for some categories, though sometimes at differing levels of aggregation or frequency. As consumer unit-level expenditures are the building blocks of this analysis, we generally choose the Interview as the source when both surveys have coverage. We estimate total expenditures from the CE that are as close to PCE definitions as possible, and out of these we find about 95% can be represented by the Interview in 2019, and expect for similar results for other years. This results in a remainder of about 5% for which the Diary is the only source or the most-reliable source by our judgment.6

For a given calendar year, we start with the set of quarterly CE interviews that were collected from the first quarter through the first quarter of the next calendar year. The staggered timing of CE interviews, combined with missing data, means a relatively small sample give annual expenditures for an exact calendar year. To form a larger sample, we include all consumer units whose expenditure reference periods started as early as November of the prior year or ended as late as February of the following year, provided they completed at least two quarterly interviews. This yields a sample of 8,236 consumer units in 2017, 7,717 in 2018, 7,171 in 2019, 6,734 in 2020, and 6,726 in 2021. For consumer units who completed fewer than four quarters, we scale their expenditures up to represent one year of expenditures

We also recalibrate the sampling weights to match calendar year counts and average demographic characteristics from the Current Population Survey (CPS). The BLS usually calibrates CE data to 35 controls or known totals of people and households in the civilian non-institutional U.S. population; the known totals are from the CPS. The controls are counts based on certain geographic and demographic variables, with some being people counts and others being household counts. These include, for 2020, 14 age/race groups, 9 Census divisions, 9 urban Census divisions, the total number of Hispanics, the total number of owners, and the total number of households. The latter two controls are household counts; the BLS converts these household counts to CE consumer unit counts by factoring in a relatively small adjustment.

This project requires modifications to the normal CE calibration process. The CE normally treats each consumer unit independently in a quarterly weighting process, with a consumer unit defined as having participated in any one of its four quarterly interviews. However, we use interviews from an expanded period as noted earlier (the first quarter of the calendar year through the first quarter of the subsequent calendar year), with a consumer unit defined as having participated during that period in either 2 of its 4 quarterly interviews, 3 of its 4 quarterly interviews, or all 4 of its quarterly interviews. This special way of defining a consumer unit means that we generate only one set of calibrated weights for each consumer unit, no matter if it completed two, three, or four quarterly interviews. Our calibrations averaged the 35 controls from the CPS data over the period that the data were collected, rather than over three months. Lastly, we adjusted the weights for the set of quarterly CE interviews used in this project since the data are a subsample from five quarters of CE’s collected data. This adjustment ensures the sum of the weights match the number of consumer units in the population in the intended calendar year.

We start with the monthly expenditure (MTAB) files from the CE and map spending to PCE categories, as described in the following subsection. After mapping CE expenditures to the PCE categories and imputing some missing categories, we scale expenditures so that population totals by major product type match PCE estimates from NIPA Table 2.3.5.

2.B. CE-PCE Concordance

As stated, we match the CE microdata to the definitions of the PCE categories to the greatest extent possible. We start from a mapping of CE spending categories, called Universal Classification Codes (UCC) to PCE categories or product types. BLS maintains the mapping for comparison purposes (Bureau of Labor Statistics, 2019). Some details on this mapping and imputations are introduced in the subsections below. The procedure used to distribute motor vehicle maintenance and repair services, financial services furnished without payment, income loss insurance, and worker's compensation insurance is similar to McCully (2014) in that some of the PCE amount are distributed using a noncomparable CE series (e.g., motor vehicle insurance premiums) as an indicator. They differ from McCully (2014) in that our methods use only published PCE amounts and only CE microdata (McCully also used wage data from the Current Population Survey Annual Social and Economic Supplement).

2.B.1. Automobiles

The CE published estimates for new and used automobiles subtract trade-in allowances from sales prices. To better reflect PCE aggregates, we add back in the trade-in values for new vehicle purchases. We then subtract these trade-in allowances from used vehicle spending under the assumption that vehicles traded in are eventually resold to the household sector, and as such, more closely align with the definition of new and used automobiles in the PCE. For this analysis, CE used vehicle expenditures do not include those that reflect household to household transactions for consistency with the PCE. Section 3.C discusses the impact of alternative treatments of used vehicles on the analysis.

2.B.2 Financial Services Furnished Without Payment

PCE includes financial services—such as free checking accounts and online record-keeping—which benefit consumer units, but CE does not capture because they are not purchased explicitly. We distribute the PCE amount for these services assuming they are proportional to the value of balances held at commercial banks and other depository institutions, as well as pension funds. The CE only collects this information in the fourth interview, which is missing for some consumer units due to attrition. To impute values for these missing units, we use a statistical matching procedure based on other observations in the same income quartile.

2.B.3. Health care

Table 2.1 summarizes the imputation procedures and data sources for health care.7 The comparison of CE and PCE health care expenditures is complicated by two issues. First, the scopes of the CE and PCE differ. Before the CE can be mapped to the PCE, expenditures that are out of scope in the CE must be imputed to consumer units. Second, the health care expenditure categories in the PCE are defined differently than in the CE, and adjustments must be made to the CE expenditures as some items that are classified as health expenditures in CE get mapped to non-health categories in the PCE.

Table 2.1: Imputations of Health Care Expenditures
Type For Whom Value of Imputation Source of data

Public

Medicare

Assigned to consumer unit (CU) members enrolled in Medicare National average benefits scaled by state level spending differences CMS 2020 Medicare Trustees Report (https://www.cms.gov/files/document/2020-medicare-trustees-report.pdf)

Medicaid

Assigned based on the number of CU members identified as participating National average expenditures per enrollee scaled by state level spending differences CMS National Health Expenditures (NHE)

CHIP

Assigned based on the number of CU members identified as participating National average expenditures per enrollee CMS National Health Expenditures (NHE)

Other public (VA, Tricare, and other military, and IHS)

Assigned based on the number of CU members identified as participating National average expenditures on private care Agency Budgets

Private

Employer provided

Assigned to each CU reporting employer coverage based on the number of individuals covered: self only (number covered = 1), self plus one (number covered = 2), family (number covered >2) Census region average premium paid by employer for health insurance by policy type (self only, self plus one, family) MEPS-IC (https://datatools.ahrq.gov/meps-ic)

Individual

Assigned to each CU reporting individual coverage and receiving a subsidy State average premium tax credit among those who receive a credit CMS (https://www.cms.gov/files/document/2016-2021-1h-effectuated-enrollment-tables.xlsx)

Notes: For private insurance, the imputed amounts are added to out-of-pocket premiums reported in the CE. For other public plans, only care provided in non-government facilities is in scope for the PCE. The VA and IHS budgets include these amounts as a separate line item. The Department of Defense submits a separate budget for care purchased from private providers for Tricare.

The scope of the health care categories differs greatly between the CE and PCE. The CE only includes out-of-pocket spending for health insurance and health care goods and services. In contrast, the PCE in addition includes expenditures by employers and the government on behalf of consumers which we impute using external sources.8 For employer contributions to health insurance premiums, we use Medical Expenditure Panel Survey – Insurance Component (MEPS-IC) data to compute average employer contributions by plan type (self, plus one, and family) and Census region. For plans purchased in the individual market, we add the average tax credit by state to the out-of-pocket premiums for individuals who report receiving a subsidy. No adjustments are made for those who purchase unsubsidized plans in the individual market. For Medicaid and Child Health Insurance Program (CHIP), we use the average expenditures per enrollee from the National Health Expenditure (NHE) Table 21. For Medicaid, we apply state-level factors based on data from Center for Medicare and Medicaid Services (CMS). For Medicare, we compute the average expenditure per enrollee for traditional Medicare (parts A and/or B), Medicare Advantage (part C), and prescription drug coverage (part D) using Boards of Trustees (2020). We use state-level Medicare spending data to scale these values by state and add these amounts to the out-of-pocket premiums included in the CE data for part C which capture the out-of-pocket premium in excess of the part B premium. Other government programs (Indian Health Service, CHAMP-VA, and Tricare) provide much of the care at government owned and operated facilities, which are out of scope for the PCE. Care provided in non-government owned facilities through these programs are in scope, so we use the average expenditure for private providers per enrollee for these programs.9 The per enrollee amounts are multiplied by the number of covered members in the consumer unit.

Once the total premium expenditures are imputed for the CE, some adjustment must be made to CE health expenditures to match the category definitions in the PCE. With the imputations for employer and government contributions to health insurance, CE health insurance expenditures include the total premiums and are categorized as health expenditures along with out-of-pocket spending on medical goods and services. Health insurance in the PCE measures net premiums (premiums minus benefits) and is categorized as a financial service. As defined for the PCE, spending on medical goods and services includes the total amount paid to the providers (out-of-pocket payments plus any insurance reimbursement), while CE only includes out-of-pocket spending. The medical care category in the CE includes medical goods and medical services (including insurance). In contrast, in the PCE (Table 2.4.5 which includes the underlying detail for Table 2.3.5), some medical goods (e.g., therapeutic appliances and equipment) are categorized as durables, others as non-durables (e.g., pharmaceutical products), and other as health care services. The health care product category in the PCE as presented in Table 2.3.5 includes non-insurance medical services only.

Before the CE can be mapped to PCE, health insurance premiums that go towards benefits need to be reassigned to medical goods and medical services. The remaining premium amount reflects the net premiums and can be mapped to the health insurance category in the PCE. Thus, total health insurance premiums in the CE are allocated across four different PCE categories: Therapeutic appliances and equipment (durables), pharmaceutical and other medical products (non-durables), health care services (health care), and net health insurance (financial services). Data from the National Health Expenditure Accounts (NHE) are used to reassign total premiums into these three categories based on the type of insurance.

2.B.4. Higher Education Services

The CE only collects information on out-of-pocket tuition payments by consumer units, while PCE is based on the receipts and outlays of colleges and universities and implicitly, therefore, includes scholarships, grants, and payments made on behalf of students. We distribute the college tuition PCE, minus the CE amount, proportionally to the number of college students in the household.

2.B.5. Motor Vehicle Repair Services

A significant portion of PCE is paid by insurance companies on behalf of households. We distribute the PCE amount less the CE amount proportionally to the household’s spending on motor vehicle insurance. To our knowledge, consistent and timely data is not available to allocate insurance premiums directly as we do with health insurance, though this is a potential area for future research.

2.B.6. Wage Loss and Workers Compensation Insurance

We distribute PCE amounts proportionally to using wages and salaries from the CE. These are collected in the first and fourth interviews, and BLS carries over values from the first interview to the second and third interview when producing the datafiles.

2.C. Remaining CE-PCE discrepancies

For this study, we were able to adjust CE to account for some differences, but not all. We redefined CE expenditure categories to follow those of the PCE as much as possible. For example, for comparability to the PCE, rental equivalence as collected in the CE is used as in the published comparisons between CE and PCE.10 This contrasts the CE published tables of means, which use out-of-pocket shelter expenditures.11 In addition, household-to-household transactions (e.g., for vehicles) are dropped from the CE expenditures. Our efforts to adjust CE data for education spending from NPISHs or the government are limited to the procedure described in Section 2.B.4. While the value of home-produced food on farms is included in PCE food expenditures, the value of this is not included in the CE. The adjustment procedure we used to scale up CE data to match PCE aggregates accounts for this under-representation.

We were only able to partially adjust for the differences due to CE and PCE population coverage. Since the BEA produces aggregates for households and NPISHs, we were able to focus on household expenditures only. In other words, we excluded final consumption expenditures identified as the net expenses of NPISHs in Table 2.3.5; these are defined as gross output of NPISHs minus the receipt of sales of goods and services by NPISHs. We also do not make any adjustments to PCE totals for residents traveling abroad or domestic spending by nonresident households. The PCE population includes among resident households, government employees and private employees living abroad; these households are not included in the CE population. However, included in the CE population but not the PCE are foreign students if they have a U.S. address or live in U.S. college or university housing.

2.D. Statistical Matching to Impute Diary Expenditures

We implement a statistical matching procedure based on Hobijn et al. (2009) to impute the remaining 5% of PCE not sourced from the Interview Survey. Similar donors from the Diary sample provide the missing data for each Interview consumer unit, where similarity is determined by a model of monthly expenditure as a function of demographic characteristics.12 The model provides a convenient method of weighting a relatively large number of characteristics by doing so according to which linear combination most strongly predicts expenditures. We then use the predicted values to form measures of distance between Interview consumer units and prospective Diary donors. The only characteristic guaranteed to match between donor and recipient is quintile of the annual before-tax income distribution.13 The matching procedure is many-to-one, as we draw four donor Diaries for each Interview in each month with replacement.14

First, we stratify both Interview and Diary consumer unit samples by quintiles of annual before-tax income. The rankings of income are done for each survey. For each expenditure reference month 𝑡 and quintile 𝑞, we use the Interview sample to estimate the regression Formula where yht is logged expenditure of consumer unit h, uht is an error term, and xht include Census region, urban/rural, age, race, sex, and education of the reference person, consumer unit size, and the prior year’s annual before-tax income.15 For the model, the measure of expenditure we use is the total monthly expenditure of the Interview household after mapping to PCE product definitions, but before imputations for employer or publicly provided health care.16 We use the least squares estimator weighted by the CE sampling weight, finlwt21.

Let formula be the slope estimate for quintile 𝑞 in month 𝑡. As household characteristics are available and comparably defined in both surveys, we calculate predicted values formula for each Diary and Interview observation. For a given Interview observation ℎ and Diary observation 𝑘, the distance metric is defined as  formula

Within each month and income quintile, we calculate δt  for all {ℎ,𝑘} pairs. Then for each Interview observation ℎ, we randomly select (with replacement) four 𝑘 from the twenty smallest δt out of all the Diary observations from the same month and income quintile. The random component is intended to ensure a more even distribution of matches across Diary observations. The detailed set of expenditures (after adjustments to match PCE definitions) of the donor Diary is then assigned to the recipient Interview. As one donor Diary is intended to represent one quarter of one month of expenditure, but Diaries correspond to a one-week reference period, the donor Diary expenditures are scaled by 13/12. This process is repeated for each Interview observation, for each month it is in the sample.

2.E. Upper Tail Adjustment

Even after imputations to capture spending which is out-of-scope to the CE, it is possible that the data understate inequality in consumption expenditures. To mitigate this, we follow the suggestion in Zwijnenburg, et al. (2022) by making draws from a type-I Pareto distribution. The distribution is applied to the top 5% of consumer units ranked by total spending after adjustments and imputations, but before scaling to match NIPA totals. We used a shape parameter of 2 based on Zwijnenburg, et al. (2022) and our judgement.

2.F. Scaling Estimates to Match PCE

After allocating CE to PCE categories and imputing missing items, we then sum expenditures for each consumer unit to the PCE major product level. CE aggregates differ from PCE aggregates for most categories. Table 3.1 describes the extent to which CE and PCE differ after adjustments to health-related allocations and imputations. Before adjusting for these discrepancies, we impose a lower bound of zero which affects a small number of observations with negative expenditure totals at the major product level.17

To allocate PCE expenditures across the distribution of equivalized total expenditures, as defined above, a next step is needed. CE expenditures are scaled up to consumer unit-level PCE estimates so that major product group totals match BEA’s published estimates. This scaling is referred to as a proportional adjustment and reflects the allocation of the gap between CE and PCE totals in proportion to the underlying micro data. For example, if aggregate PCE health care expenditures equal $2,458 billion and consumer unit X had 0.000005% of total CE health care expenditures (after imputations for health insurance and allocations to PCE categories), consumer unit X is assigned 0.000005%*$2,458 billion as its total health care expenditure. The scaling implicitly distributes the “missing” product-level expenditures identically to those captured by the CE. Ideally, we would aim to use supplementary data to make imputations to narrow the gaps as much as possible before scaling. Our efforts in this area thus far originally focused on healthcare, where government and employer provided benefits have a much more equal distribution than out-of-pocket spending captured in the CE. See Garner, et. al. (2022) for more details. Since then, we have added further imputation (see subsection 2.B), including for financial services furnished without payment.

2.G. Equivalence Scales and Ranking

To create the deciles across which to distribute the PCE expenditures, we rank consumer units based on their total expenditures equivalized by the square root of family size as the equivalence scale. In creating the ranking (as a precursor to creating quantile groupings), consumer units are weighted by their CE sampling weight (finlwt21).18 Total expenditures are defined as the sum of aggregate expenditures based on PCE major product type after all the processing described in Section 2.B to match PCE definitions and impute spending not covered by CE (e.g., health care), and after scaling up expenditures to PCE totals as described in Section 2.F.

3. Supplemental Results

3.A. CE/PCE Coverage Rates

Table 3.1 below shows the ratios of CE to PCE aggregates before the scaling to match PCE product type totals. See Garner, et. al. (2022) for more details on the impact of the allocation and imputation of health care spending on the PCE coverage rates. For example, for all categories in 2019, the CE/PCE ratio is 0.71. For Durable Goods, it is 0.55, for Non-Durable Goods, it is 0.65, and for Services it is 0.75. The ratio for Housing and Utilities is 1.12 in 2019. That this is greater than unity is largely due to owner equivalent rent. In the CE, this is measured using estimates reported by the consumer unit, while the BEA bases their measure on an imputation model based on actual rents. See Garner and Short (2009) for more on this topic.

Table 3.1: CE Coverage of PCE Total After Adjustments and Imputations
Category 2017 2018 2019 2020 2021

PCE less final consumption expenditures of nonprofit institutions serving households

0.70 0.80 0.79 0.80 0.77

Durable goods

0.66 0.64 0.63 0.67 0.60

Motor vehicles and parts

1.02 0.98 1.02 1.11 0.83

Furnishings and durable household equipment

0.56 0.55 0.58 0.59 0.55

Recreational goods and vehicles

0.36 0.34 0.27 0.34 0.45

Other durable goods

0.43 0.42 0.45 0.44 0.40

Nondurable goods

0.66 0.66 0.66 0.63 0.61

Food and beverages purchased for off-premises consumption

0.76 0.77 0.77 0.75 0.75

Clothing and footwear

0.46 0.47 0.48 0.42 0.37

Gasoline and other energy goods

0.88 0.84 0.86 0.89 0.83

Other nondurable goods

0.57 0.57 0.56 0.52 0.50

Household consumption expenditures (for services)

0.85 0.88 0.86 0.89 0.83

Housing and utilities

1.22 1.24 1.19 1.20 1.21

Health care

0.75 0.79 0.79 0.84 0.77

Transportation services

0.77 0.78 0.75 0.84 0.79

Recreation services

0.43 0.46 0.42 0.44 0.46

Food services and accommodations

0.54 0.55 0.56 0.48 0.56

Financial services and insurance

0.80 0.84 0.83 0.83 0.72

Other services

0.85 0.86 0.85 0.81 0.79

Note: Data represent CE divided by PCE as published in BEA NIPA Table 2.3.5 (September 28, 2023 release).

3.B. Impact of Weighting

Researchers have used various weighting options for the production of distributional statistics (e.g., percentile ratios like the 90/10 and aggregate indexes like the Gini). For example, de Queljoe et al. (forthcoming 2024) and Gindelsky (2020) weighted equivalized expenditures using household or consumer unit weights (finlwt21 for the CE) following the EG DNA guidelines.19 The reasoning for selecting this option is that households, not people, are the focus of income and consumption expenditure national accounts. In contrast, the OCED ICW expert group (OECD 2013) recommended that the use of person-weighting when producing distributional statistics based on an economic unit (e.g., household, consumer unit, family) or any other statistical unit that combines individuals. The assumption with this weighting option is that equivalized household income or expenditures are available to each person in the household. In the case of the CE, the person-weight would be defined as finlwt21 times consumer unit size. Another option is to weight by the number of “consumption units”; this is the number of equivalized adults times the household or consumer unit weight. The number of equivalent units is a function of the equivalence scale chosen. For example, with the equivalence scale being the square root of consumer unit, a 2-adult household is represented by 1.41 equivalent units (times the consumer unit population weight). When conducting distributional statistics, this latter weighting option ensures that the sum of equivalized expenditures within a decile, for example, equals the total expenditure for the decile.

In our estimates, we use finlwt21 (the CE sampling weight) for computing statistics concerning equivalized PCE and for the creation of decile and other quantile groups (e.g., the 10%-20% group).20 We study alternative weights—finlwt21 times family size and finlwt21 times the square root of family size. We find a minor quantitative impact, which is shown in Table 3.2. The effects of including family size in the weight are a slight decrease in the mean and median and a slight decrease in the share accounted for by the top decile and percentiles. Summary measures like the Gini index are constant (to two decimal places), while the 90/10 ratio is slightly lower (3.39 vs. 3.45). Garner, et. al. (2022), in Appendix A, show the impact of the weight change on the deciles of total equivalized expenditure.

Table 3.2: Impact of Weighting on Quantiles and Distributional Statistics, 2019
Statistic / Weight finlwt21 finlwt21*sqrt(fam_size) finlwt21*fam_size

Equivalized total mean

$73,118 $72,527 $71,241

Equivalized total median

$59,379 $58,876 $57,938

Equivalized total: 0-20% share

8.6% 8.6% 8.7%

Equivalized total: 20-40% share

12.8% 12.7% 12.8%

Equivalized total: 40-60% share

16.3% 16.3% 16.3%

Equivalized total: 60-80% share

21.0% 21.0% 21.0%

Equivalized total: 80-100% share

41.3% 41.4% 41.3%

Equivalized total: 80-99% share

32.6% 32.7% 32.6%

Equivalized total: Top 1% share

8.7% 8.7% 8.7%

Equivalized total: Top 5% share

19.1% 19.2% 19.1%

Equivalized total: 90/10

3.47 3.45 3.43

Equivalized total: Gini index

0.32 0.33 0.32

Equivalized total: Theil index

0.25 0.25 0.25

Equivalized total: Log-Deviance

0.18 0.18 0.18

Equivalized total: CV

0.73 0.75 0.75

Notes: Columns track changes in distributional statistics when different weights are used to both to create quantile groups and weight the statistics. Finlwt21 is the sampling weight used by CE, adjusted for our sample of interviews. “fam_size” refers to the number of members of the consumer unit.

3.C. Impact of Net Spending for Used Automobiles

Table 3.3 below presents results under alternative treatments of used motor vehicles compared against the baseline results in column (1). The baseline processing rules of excluding purchases which originated from another household and subtracting sales made by the consumer unit most closely match PCE definitions. Indeed, these rules produce CE aggregates which come closest to PCE totals (ratios close to unity). Subtracting sales has relatively minor effect on the distributional results, while excluding household-to-household purchases shifts the distribution of expenditures slightly toward the lower quantiles of equivalized PCE once recalculated.

Table 3.3: Motor Vehicles and Parts Results With Alternative Processing, 2019
(1)* (2) (3) (4)

Processing Rules

Exclude used purchases from households

Yes Yes No No

Subtract used sales

Yes No Yes No

Spending share by decile of equivalized expenditure

0-20%

1.1% 1.4% 1.9% 2.0%

20-40%

5.1% 4.8% 5.7% 5.8%

40-60%

7.8% 8.8% 11.0% 11.4%

60-80%

23.2% 21.4% 23.0% 23.5%

80-100%

62.7% 63.7% 58.4% 57.3%

CE/PCE Coverage before Scaling

Used Vehicles

1.00 1.15 1.32 1.48

Used Cars

1.25 1.44 1.78 1.97

Used Trucks

0.88 1.02 1.11 1.25

* Baseline

References

Agency for Healthcare Research and Quality. 2021. Medical Expenditure Panel Survey. Accessed April 13, 2022. https://www.meps.ahrq.gov/mepsweb/.

Blanchet, Thomas, Emmanuel Saez, and Gabriel Zucman. 2022. "Real-Time Inequality." Unpublished manuscript. https://eml.berkeley.edu/~saez/BSZ2022.pdf.

Boards Of Trustees, Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds. 2020. 2020 Annual Report of the Boards of Trustees of the Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds. Washington, DC: Center for Medicare and Medicaid Services. https://www.cms.gov/files/document/2020-medicare-trustees-report.pdf.

Bureau of Economic Analysis. 2021. "Chapter 5: Personal Consumption Expenditures." In Handbook of Methods, by Bureau of Economic Analysis, 1-69. Washington, DC: Bureau of Economic Analysis. https://www.bea.gov/resources/methodologies/nipa-handbook/pdf/chapter-05.pdf.

—. 2021. Distribution of Personal Income. December 15. Accessed May 14, 2021. https://www.bea.gov/data/special-topics/distribution-of-personal-income.

Bureau of Labor Statistics. 2019. Comparing expenditures from the Consumer Expenditure Surveys with the Personal Consumer Expenditures: Results of the CE/PCE Concordance. Accessed May 9, 2022. https://www.bls.gov/cex/cepceconcordance.htm.

Centers for Medicare and Medicaid Services. 2021. National Health Expenditure Data. Accessed July 29, 2021. https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/NationalHealthExpendData.

de Queljoe, Matthew, Joseph Grilli, and Jorrit Zwignenburg. 2024 (forthcoming). OECD Centralised Approach to Calculate Distributional Results in line with National Accounts Totals. Paris: OECD.

Fisher, Jonathan D., David S. Johnson, Timothy M. Smeeding, and Jeffrey P. Thompson. 2022. "Inequality in 3-D: Income, Consumption, and Wealth." Review of Income and Wealth 68 (1): 16-42. doi:10.1111/roiw.12509.

Fixler, Dennis, David Johnson, Andrew Craig, and Kevin Furlong. 2017. "A Consistent Data Series to Evaluate Growth and Inequality in the National Accounts." Review of Income and Wealth S437-S459. doi:https://doi.org/10.1111/roiw.12324.

Fixler, Dennis, Marina Gindelsky, and David Johnson. 2020. Measuring Inequality in the National Accounts. BEA Working Paper 2020-3, Washington, DC: Bureau of Economic Analysis. https://www.bea.gov/system/files/papers/measuring-inequality-in-the-national-accounts_0.pdf.

Garner, Thesia I., and Kathleen Short. 2009. "Accounting for owner-occupied dwelling services: Aggregates and distributions." Journal of Housing Economics 18: 233-248. doi:10.1016/j.jhe.2009.07.004.

Garner, Thesia I., Robert S. Martin, Brett Matsumoto, and Scott Curtin. 2022. Distribution of U.S. Personal Consumption Expenditures for 2019: A Prototype Based on Consumer Expenditure Survey Data. Working Paper 557, Washington DC: Bureau of Labor Statistics. https://www.bls.gov/osmr/research-papers/2022/pdf/ec220120.pdf.

Gindelsky, Marina. 2021. Technical Document: An Updated Methodology for. Washington, DC: Bureau of Economic Analysis. https://apps.bea.gov/data/special-topics/distribution-of-personal-income/technical_document.pdf.

—. 2020. United States: Household distributional results in line with national accounts, experimental statistics. Accessed July 28, 2022. https://www.oecd.org/sdd/na/household-distributional-results-in-line-with-national-accounts-experimental-statistics.htm.

Hobijn, Bart, Kristin Mayer, Carter Stennis, and Giorgio Topa. 2009. "Household Inflation Experiences in the U.S.: A Comprehensive Approach." Working Paper 2009-19. Federal Reserve Bank of San Francisco. http://www.frbsf.org/publications/economics/papers/2009/wp09-19bk.pdf.

Martin, Robert. 2022. "Democratic Aggregation: Issues and Implications for Consumer Price Indexes." Unpublished manuscript.

McCully, Clinton P. 2014. "Integration of Micro and Macro Data on Consumer Income and Expenditures." In Measuring Economic Stability and Progress, by Dale W. Jorgenson, J. Steven Landefeld and Paul Schreyer (Eds.). University of Chicago Press.

Organization for Economic Cooperation and Development. 2020. Distributional Information on Household Income, Consumption and Saving in Line with National Accounts Guidelines. Organization for Economic Cooperation and Development.

Passero, William, Thesia I. Garner, and Clinton McCully. 2014. "Understanding the Relationship: CE Survey and PCE." In Improving the Measurement of Consumer Expenditures, 181-203. University of Chicago Press. http://www.nber.org/chapters/c12659.

Zwijnenburg, Jorrit, Grilli, Joseph, and Engelbrecht, Pao. 2022. "Pareto Tail Estimation in the Presence of Missing Rich in Compiling Distributional National Accounts." Paper prepared for the 37th IARIW General Conference, August 22-26, 2022. https://iariw.org/wp-content/uploads/2022/07/Jorret-Joseph-Pao-IARIW-2022.pdf.

If you are deaf, hard of hearing, or have a speech disability, please dial 7-1-1 to access telecommunications relay services or the information voice phone at: (202) 691-5200. This article is in the public domain and may be reproduced without permission

  • Notes

1 Division of Price and Index Number Research (Garner, Martin, Matsumoto), Division of Consumer Expenditure Surveys (Curtin), Bureau of Labor Statistics, 2 Massachusetts Ave., NE, Washington, DC 20212, USA. Emails: Garner.Thesia@bls.gov, Martin.Robert@bls.gov, Matsumoto.Brett@bls.gov, Curtin.Scott@bls.gov.

2 See Garner, T. I., Martin, R.S., Matsumoto, B. and Curtin, S. 2022 “Distribution of U.S. Personal Consumption Expenditures for 2019: A Prototype Based on Consumer Expenditure Survey Data.” BLS Working Paper 557.

3 See Passero et al. (2014) for a description of differences in concepts, measurement, and populations and a BLS approach to support comparisons of aggregate expenditures from the CE and PCE. For CE to PCE ratios based on aggregate expenditures for 2020, overall CE to PCE ratio is 0.62, while for comparable items it is 0.73 (see https://stats.bls.gov/cex/cecomparison/pce_profile.htm)

4 While the PCE includes expenditures made by resident households who normally live in the U.S. but who are temporarily aboard (see earlier footnote for the definition of “resident”), we do not adjust the PCE aggregates to deduct expenditures by these households for this analysis due to data limitations.

5 Neither the CE nor PCE includes in expenditures the value of in-kind transfers of goods and services such as government low-income food assistance (e.g., Supplemental Nutrition Program for Women, Infants, and Children), energy assistance (e.g., Low Income Home Energy Assistance Program, LIHEAP), and rental assistance (e.g., Housing Choice Voucher Program Section 8, rent control, free rents).

6 These proportions are computed using all consumer units before restricting to the set with two or more interviews.

7 Garner, et. al. (2022) used only national averages for imputed health expenditures, while subsequent releases incorporate state-level information for Medicare, Medicaid, and private insurance.

8 In the OECD’s Expert Group on Distributional National Accounts (EG DNA) methodology, this type of adjustment is referred to as Method C where missing components are imputed according to exogenous data, e.g., sociodemographic data or in our case, reports of different types of health insurance coverage (see de Queljoe et al. 2022). However, in our case, Method A, proportional scaling, is applied after the health expenditure imputations.

9 Care purchased from private providers is identified as a separate line item in the budgets for these programs. We ignore any out-of-pocket premiums reported in the CE for these programs to avoid double counting, which also allows us to avoid the issue of determining how much of the premiums go towards out-of-scope care.

10 https://stats.bls.gov/cex/cecomparison/pce_profile.htm

11 See the following for the definition of shelter for owner occupied housing: https://stats.bls.gov/cex/csxgloss.htm#housing

12A similar procedure is being used in ongoing research to create household-weighted Consumer Price Indexes (Martin, 2022).

13 The Diary samples are small enough, particularly on a monthly basis, that conditioning on multiple characteristics quickly leads to empty cells. See Hobijn, et al. (2009) for more discussion.

14 This is in contrast with the one-to-one “optimal transport” method that Blanchet, et al. (2022) uses to match CPS observations with public-use tax data. A many-to-one match is convenient for our purposes because there are many more Interview observations than Diary observations in a given month.

15 These demographic variables technically pertain to the collection quarter or some other reference period. For instance, in the first interview, income represents the 12 months prior to the interview date, and this value is assigned to the second and third interview records. In the fourth interview, income is collected again for the 12 months prior to the interview date. We implicitly assume the demographic variables are representative of characteristics from the expenditure reference months.

16 Alternatively, it might be attractive to use the Diary sample to estimate Diary expenditures as a function of demographic characteristics, as we intend to impute these expenditures for the Interview sample. However, we find that characteristics explain relatively little variation in Diary expenditures, perhaps due to the short (week-long) record or recall period.

17 The constraint affected 0.05% of observations in 2019 and had an effect of increasing the ratio of total CE to PCE spending by about 0.2 percentage points.

18 In Garner, et. al. (2022), the weight used in ranking was finlwt21 times the consumer unit size.

19 https://www.oecd.org/sdd/na/OECD-EG-DNA-Guidelines.pdf

20 Quantile groups are created using PROC UNIVARIATE in SAS to compute the quantiles themselves, and then assigning group membership based on consumer unit-level equivalized PCE relative to these quantiles.

Last Modified Date: Friday, December 16, 2022