Understanding Zip Code Data


When working with Zip Code level data, there are several factors to consider. This article, while not exhaustive, lists the main items to keep in mind when using Zip Code data for your geographical study Region.



Zip code files, like other IMPLAN files, can be combined to create a Region or used independently. Please keep in mind that while an individual zip code file will create Multipliers and have impacts, these impacts will be minimal. This is because zip codes may have little or no population and/or little to no employment. Also, many individual zip codes are not large enough to allow for local sourcing of materials needed for the indirect effect or to provide adequate services to create significant induced effect leading to much of the potential impact being lost in leakages.

Zip code regions may represent very small economic Regions and consequently, be extremely open to leakages. These leakages, even to nearby regions (in some instances this could literally be across the street) are lost from the Region resulting in very small Indirect and Induced Effects. Please take into consideration where indirect purchases will be located and where employees may be spending their labor income when customizing zip code regions for analysis.



Multi-Regional Input-Output (MRIO) analysis allows for purchases to be made between Regions (the zip code(s) and the linked regions) by tracking trade between these regions, potentially capturing many of the lost impacts. 



Not all zip codes listed by the U.S. Postal Service may be represented in the zip code package you receive. The USPS can open/close post offices or reorganize routes on an on-going basis, thereby changing zip-code demographics. Because of this, the County Business Patterns and Census demographic zip code representation may not be current.

Depending on the year, roughly 4% of zip-codes have neither County Business Patterns (CBP) employment nor Census (demographic) data. If your region includes any of these zip codes, they will not be available in the package you purchase, as there is inadequate information to create Multipliers for these regions. Depending on the year, there will be roughly 8,000 zip-code files for which there is only CBP data with no demographic data. Most of these are P.O. boxes and “unique” point codes. They serve business but do not represent residential population. With no household representation, all employment would be considered in-commuting and have no local induced effect.

Conversely, there are also some zip-code files with only demographic data and no CBP data. While CBP data does not exist for these zip-codes, they may still contain employment in some sectors since CBP data do not cover all IMPLAN sectors. Population is used as a distributor for most of these non-covered sectors. Farm counts by zip-code from the 2017 Census of Agriculture are used to distribute agricultural industry data in these cases.

There will be cases where you may find an Industry exists in an actual zip code region but does not show up in the zip code data (or even the county data). This occurs because of unreported sectors in the CBP and inconsistencies in data between CBP and BLS covered wages and employment. When this issue occurs, you will need to customize your Region to add the Industry to the Study Area. CBP data are primarily obtained from administrative records supplied by the IRS, Social Security, and other sources. CBP is tabulated on an establishment basis, and each business location is tabulated only once according to the primary business activity. The industry classifications of establishments in the CBP are self-reported in the vast majority of cases. BLS CEW data are obtained from quarterly tax reports submitted to State Employment Security Agencies.

546 Industry Scheme


The 2018 data is now the default year in IMPLAN! The data release this year is especially exciting because we have a brand new Industry scheme with 546 Industries based on the updated BEA Benchmark and Census of Agriculture.

The new Industry scheme obviously has some changes from the 536 Industry scheme. We only lost three Industries due to disaggregation. Notably, wholesale trade has been split into 10 different wholesale Industries. Insurance carriers is now split into direct life and all others. Real estate is now split into tenant-occupied and other real estate. Finally, at both the local and state levels, the non-education spending pattern will be split into hospitals and health services and other services. Details on the changes can be found in the article BEA Benchmark & The New 546 Industry Scheme.



Compare data across time (2001 – 2018) in IMPLAN, using the most up-to-date Industry Scheme!

When there are significant changes to the underlying data, there is no longer an apples to apples comparison. Therefore, you will not be able to compare data sets, Projects or Event Templates in the 536 Industry scheme (or another) non-546 Industry scheme) with the new 546 Industry scheme. 

For comparing across different Industry Schemes, we have a bridge that should help in most instances. 

The 2018 data and the 546 Industry Scheme is the default on your Regions screen. The Regions screen will show you data from 2001-2018 in the 546 Industry Scheme. 




To find the older data years (2012-2017), navigate from your home screen to the Projects screen.  On the Projects screen, click the NEW PROJECT button in the upper right.




In the pop-up box, give your new Project a title and then in Industry Set, choose “US – 536 Sectors.” Then ensure that your Household Set is “Set 1 – (2015 Datasets or later).” This will give you the choice on your Regions screen between 2015, 2016, and 2017 datasets.  To find data from 2012-2014, choose “Set 2 – (2014 Datasets or earlier).”




Click CREATE PROJECT and you will be taken to your Regions screen to set up as always!






2018 Data Release Notes

546 Industries, Conversions, Bridges, & Construction – 2018 Data

BEA Benchmark & The New 546 Industry Scheme

IMPLAN Sectoring & NAICS Correspondences


Sectors are a way of describing your Industry. All versions of the Sectoring schemes (except the 528) are based on NAICS codes but represent differing levels of NAICS code rollups. Generally speaking, manufacturing Sectors are 4-5 digit NAICS; whereas, agriculture and services are 3-4 digit NAICS. Choosing the correct Sector is vital to an accurate analysis.



The Sectoring schemes represent rollups of NAICS descriptions, and each Sector has its own spending pattern derived from the BEA expenditures patterns. To see what NAICS codes and descriptions are associated to a particular Sector, you can start typing the words in the Sector directly into the Specification box on the Impacts screen.

Type in a keyword to pull up associated descriptions. Note that subsets of words will also be displayed, so for example the word “wine” will also bring up twine. Curiously, plurals are not always recognized, so if you type in “books” but do not receive fields related to your search try “book“. Alternatively, in our downloads section are spreadsheets that provide NAICS code descriptions and IMPLAN Sector bridges.

One other key element to keep in mind when selecting a Sector is that the Sector also contains the representative ratios of Intermediate Expenditures:Value Added and the Value Added ratios for that Industry in that Study Area. For all these reasons it is very important to ensure that your Sector represents the Industry you are modeling. If you have a line-item budget and want to create your own Sector, please see the section on Analysis-by-Parts (ABP).

Please note also that in data sets prior to 2012 there are no NAICS 23* correspondences, and while descriptions are included in 2012 forward, there are no corresponding IMPLAN Sector Codes because our construction Sectors come from Census rather than NAICS. Please see related articles for this and other information about special Sectors in the IMPLAN Sectoring scheme.



Below is the current (2018 data) IMPLAN Sectoring scheme. Manual bridges can be found in the article 546 Sector Industries, Conversions, Bridges, & Construction – 2018 Data.  



Special Sector Definitions


Certain IMPLAN Sectors require additional explanation, either because they are not NAICS based or they have special properties. Below are the special Sector descriptions (sector numbers are based on the 546 sector scheme for 2018 IMPLAN data sets).1



IMPLAN construction Sectors are classified by structure type (Census definitions) rather than NAICs codes. For this reason, Sector searches for construction will not pull up corresponding IMPLAN Sectors. Thus, when working with Construction Sectors, the Definitions of IMPLAN’s 546 Construction Sectors can be helpful.



This Sector estimates what owner/occupants would pay in rent if they rented rather than owned their homes. This Sector creates an industry out of owning a home, and its production function represents repair and maintenance of that home. The Sector’s sole product (Output) is ownership and is purchased entirely by personal consumption expenditures (i.e., the household Sector).

There is no Employment or Employee Compensation for this industry. Taxes on production for this Sector are largely made up of property taxes paid by the homeowner, while Other Property Income is the difference between the rental value of the home and the costs of home ownership. Interest payments and mortgage payments are a transfer in the SAM and are not part of the production function for this Sector.

Sector 449 is included in the database to insure consistency in the flow of funds. It captures the expenses of home ownership such as repair and maintenance construction, various closing costs, and other expenditures related to the upkeep of the space in the same way expenses are captured for rental properties.



While not a true special Sector, there are often many questions regarding what Sector 525 produces. This sector covers live-in household staff: maids, butlers, chauffeurs, etc. 



IMPLAN Sectors 526-534 represent government agencies that cover a substantial portion of their operating costs by selling goods and services to the public. They operate much like private sector firms, hiring labor and purchasing other inputs to produce goods that are sold through markets. Other Federal\State\Local government enterprises (i.e., those other than postal, electric utility, and transportation services) include things such as government owned and operated liquor stores, airports, sewer and sanitation services, gas, and water supply2. This differs from Administrative Government sectors (components of consumption – i.e., final demand), because administrative do not respond to local market demands.



IMPLAN Sectors 535-538 are Commodities not produced intentionally by any US industry:

  • Used and secondhand goods are goods that are traded but were not produced during the current year. While used goods are not part of the current-period gross output of the economy, they are part of the supply available for consumption. They come from capital, government institutions, and households.
  • Scrap consists of commodities that are cast off as part of a production process and then resold. Examples include sales of used aluminum cans to recyclers and sales of scrapped vehicles to metal recyclers.
  • Rest of world adjustment  “The rest-of-the-world adjustment to final uses consists of values for exports and imports that have offsetting adjustments to personal consumption expenditures (PCE) and government… This adjustment is required in order to conform the commodity treatment of the I-O use table to the expenditure concepts used for final uses in the NIPAs. This is accomplished by making offsetting adjustments between PCE and gross exports and between Federal Government nondefense purchases and exports and imports…For example, foreigners traveling in the United States consume goods and services, such as accommodations, that are included in the source data for PCE. In order to put the PCE estimate on a NIPA basis, an adjustment is made to account for these purchases.”3
  • Non-comparable foreign imports are goods that are not available anywhere in the nation. They consist of three types of services: (1) services that are produced and consumed abroad, such as airport expenditures by U.S. airlines in foreign countries; (2) service imports that are unique, such as payments for the rights to patents, copyrights, or industrial processes; and (3) service imports that cannot be identified by type, such as payments by U.S. companies to their foreign affiliates for an undefined basket of services.



Administrative government activities (e.g., legislatures, police protection) are not subject to local market forces (i.e., not driven by local demand); as such, they are held exogenous to the multiplier model.

IMPLAN Sectors 539-546 represent the payroll/value added of these administrative government Sectors. This is necessary because, while the Commodity purchases of these government institutions are already represented in the SAM, there is no payroll Commodity; thus, these Sectors are included as a bookkeeping element to account for these institutions’ payrolls. By definition, these Sectors have no intermediate purchases and thus will not generate indirect effects. For these sectors, Employee Compensation or Employment should be used as Event values; entering the operational value of the government as an Output value will greatly overestimate the impact. When modeling government programs or budgets, you will need to use the appropriate spending pattern associated to the budget activity. 



Government Institutions in IMPLAN do not have Sector designations. Instead, they can be modeled from the Impacts screen using an Institutional Spending Pattern Event. For details on how to edit and use these, visit Editing Institutional Spending Pattern Events. The following governmental spending patterns are available.

  • Federal Government
    • NonDefense
    • Defense
    • Investment
  • State/Local Government
    • NonEducation
    • Education
    • Investment
  • Capital
  • Inventory Additions/Deletions

Sectoring Schemes


Sectoring schemes provide a means of classifying and aggregating Industry and Commodity data. Each database source can have its own unique format or scheme for presenting Industry data (e.g. IMPLAN scheme or the REA scheme). An Industrial classification scheme allows categorization according to the type of products or services produced by the Industry or Industries.

Employment and Value Added data used in IMPLAN originates from surveys of industry establishments. This establishment may be a small business with a single location, or it may be a branch location of a large firm. Each establishment in the defined region is counted separately on the covered (social security or unemployment) employment rolls. When the establishment submits a report or responds to a census or a survey, its data are collected and assigned an establishment code depending on the primary product produced by that establishment.

The industry classification scheme used for all federal government industry based data sets is the 6-digit North American Industrial Classification Scheme (NAICS), as described in the most current NAICS manual, published by the Office of Management and Budget.

This scheme was adopted in 1997 and replaced the previously used Standard Industrial Classification (SIC) codes. Unlike the SIC, NAICS was developed jointly by the United States, Mexico, and Canada to allow for comparability between all North American Industrial data.

The current NAICS scheme is 2017. NAICS reports five levels of Industry detail, ranging from the 2-digit detail (the most aggregate) to the 6-digit (the most detailed). To learn more about the history of NAICS click here. Certain IMPLAN Sectors – including the construction Sectors (50-62), Sector 449 – Owner-occupied dwellings, Sector 525 – Private households, and Sectors 526-546, do not follow a normal NAICS pattern. Read more information about these Specialty Sectors. Read more for additional information on the IMPLAN Sectoring scheme and for a listing of the current 546 scheme.




A major data source used to derive IMPLAN databases is the Bureau of Economic Analysis’ Regional Economic Accounts (REA – formerly known as REIS). At the state level, REA reports in 3-digit NAICS detail for employment and income. At the county level, income is reported at 3-digit NAICS but employment is provided at the 2-digit NAICS detail.


Data from the Bureau of Labor Statistics (BLS) is used for deflators and some output estimates. The BLS uses a different sectoring scheme, again based on the NAICS code system.


IMPLAN’s current 546-sector scheme is based on the Bureau of Economic Analysis’ latest Benchmark Input-Output Study. Since every five years the BEA updates their input-output accounts, it means that on those years, IMPLAN data sets also undergo important updates as well. In late 2018, the BEA released its 2012 industry statistics and benchmark make-use tables (also known as I-O tables) which include methodological improvements to more accurately reflect the ever-changing national economy.  For more information, visit BEA Benchmark & The New 546 Sectoring Scheme.


IMPLAN Database years Number of IMPLAN Sectors BEA Benchmarks
1996-2000 528 1987 and 1992
2001-2004, 2006 509 1997
2007-2012 440 2002
2013-2017 536 2007 with parts of 1997 and 2002
2018 546 2012


Prior to NAICS, this was the most common scheme as described in the 1987 Standard Industrial Classification Manual. This scheme had four levels of detail ranging from 1-digit detail as the most aggregate to 4-digit as the most detailed. IMPLAN datasets prior to 2001 are SIC-based.

How Commuter Employee Compensation is Estimated


While payroll taxes are paid in the county of employment, personal income taxes on that same income are paid in the county of residence, and these two places differ for commuters. Additionally, household demand is generated at the location of the household (that is, at the employee’s place of residence). Therefore, a proper measure of regional and inter-regional induced effects requires accounting for these inter-regional flows of employment-based income. IMPLAN derives region specific commuting flows as described below.  

It should be noted that while IMPLAN accounts for commuting precisely so that more spending is kicked off in the place of residence than in the place of work, where that spending ultimately occurs is based largely on IMPLAN’s trade flow model. For example, suppose that an employee in Mecklenburg County, NC lives in neighboring Rowan County and therefore takes his Employee Compensation less payroll taxes (unofficially termed “Commuter EC” for the purposes of this article) home with him to Rowan County, where he then pays personal income taxes on that income. Now suppose that this individual likes to go bowling, but there is no bowling alley in Rowan County.  These bowling expenditures occur back in Mecklenburg County. Thus, while the commuting data ensure that the employee’s demand originates in the place of residence, the fulfillment of that demand may occur in the place of work (or any number of other counties).



Initial estimates of Commuter EC between counties is derived from U.S. Census data. The Census Bureau’s Journey-To-Work (JTW) data provide information on commuting flows of people from county-of-residence to county-of-employment (including commuting to the same county as residence, or intra-commuting). IMPLAN combines the Census county-to-county commuting data with IMPLAN’s own annual estimates of county-level Commuter EC to estimate flows of compensation from the county in which compensation is earned back to the county of residence. Commuter EC is the remaining portion of Employee Compensation (EC) once payroll taxes and foreign commuting are removed. This adjustment needs to be made as payroll taxes are paid in the region in which compensation is earned and foreign commuting is treated as a leakage from the model. On the question of foreign commuting, IMPLAN uses U.S. level data on worker earnings that flow out of and into the country, from NIPA, distributed to states and counties based on EC and household income, respectively.

Unfortunately, the Census’s JTW data are lagged compared to IMPLAN’s annual Commuter EC estimates. Therefore, IMPLAN turns to the BEA REA data on earnings flows, which while only providing in- and out-flow data for a region and not its flow partner, are more up-to-date. As the JTW data include intra-flows and the REA gross flows data do not, IMPLAN also utilizes REA data on earnings by place of work to derive intra flows. These data (the gross in- and out-flows and the intra-flows) are used as controls in a matrix RAS of the Commuter EC previously derived from the JTW data. As the REA data are also lagged, IMPLAN does not control strictly to their values. Once completed, updated coefficients of commuting flows are derived, which are applied to annual county-level Commuter EC values. 

The resulting county level commuting flows are utilized in the generation of regional SAMs. Their inclusion allows for the calculation of the share of regionally generated compensation that leaks from the region (i.e., the region’s in-commuting rate). A region’s in-commuting rate is calculated as:

Commuter EC outflows / total EC generated in the region
    Commuter EC outflows are displayed in the IxC SAM as transfers of the EC column to
    Domestic Trade and Foreign Trade

The in-commuting rate is then used to determine leakages of EC in a region. For example, a region with a 25% in-commuting rate will see 25% of earned EC stemming from an impact analysis treated as leakage from the local economy.

The Commuter EC data are also utilized in Multi-Regional Input-Output analysis (MRIO); Commuter EC outflows from a region to linked model regions generate induced effects in the linked model regions. To continue with the prior example, if the entirety of the 25% in-commuting rate in the direct effect region represented commuters from the linked region, then the entirety of the 25% leaked Commuter EC in the direct effect region would be treated as additional household income in the linked region.

Note, in-commuting rates are region specific but not industry specific. If your analysis of an industry requires that you adjust the in-commuting rate, please see our article on adjusting compensation to account for a known in-commuting rate.



Flows of Commuter EC (EC less payroll taxes) between zip codes are calculated by distributing the flows of Commuter EC between the Counties to which those zip codes belong among those zip codes. The shares for out-flows of Commuter EC are based on total EC generated by each zip code in the county, while the shares for in-flows of Commuter EC are based on total receipts of EC less payroll taxes by households in each of the zip codes. Commuter EC flows for zip code-based regions like Congressional Districts or custom city models are simply sums of the component zip codes’ Commuter EC flows. 

Why is Personal Income for My Region so High?

Alternative Measures of Household Income

Robert Brown and Ann Dunbar of BEA also contributed to this paper.

This paper has been prepared for presentation to the Federal Economic Statistics Advisory Committee (FESAC) on December 14, 2004. It represents work in progress and does not represent any agency’s final positions on issues addressed. The FESAC is a Federal Advisory Committee sponsored jointly by the Bureau of Labor Statistics of the U.S. Department of Labor and by the Bureau of Economic Analysis (BEA) and the Bureau of the Census of the U.S. Department of Commerce.


Two of the most widely used measures of household income are BEA’s personal income and the Census Bureau’s money income. These two statistics spring from different traditions of measurement—personal income from national income accounting and money income from income distribution analysis. Yet, many of the conceptual difficulties in developing guidelines for income distribution statistics are the same or similar to the problems encountered in specifying guidelines for national income accounting.

This paper first considers briefly what is meant by the concept “income” and how the debate about the boundaries defining income has been framed. Then, personal income and money income are compared conceptually and empirically. This comparison highlights certain ways that the two measures differ—in the inclusion or exclusion of lump sum payments, of income of non-profit institutions serving households, and of in-kind payments; in the treatment of pension accruals versus disbursements; and in adjustments for underreporting.

Both personal income and money income are more limited concepts than the Haig-Simons-Hicks (HSH) theoretical concept of income as the maximum amount that can be consumed in a given time period while keeping real wealth unchanged. Both personal and money income, for example, do not capture income from capital gains. The Census Bureau has developed a set of alternative measures of money income designed to better measure economic well-being. These are briefly reviewed in the paper.

The final section of the paper discusses possible further extensions of the two income measures. Alternative measures of personal income are proposed that move away from an accrual and toward a disbursement approach to accounting for retirement income and that incorporate disbursements from a variety of tax-preferred assets. The alternative personal income measures address user needs to better measure the tax base or the capacity to spend.

Some of the principal points of the paper are the following:

  • BEA personal income is the income received by persons from participation in production, from government and business transfer payments, and from government interest.1 BEA estimates personal income largely from administrative data sources.
  • The Current Population Survey (CPS) Annual Social and Economic Supplement is the source of the Census Bureau’s official national estimates of poverty. CPS money income is defined as total pre-tax cash income earned by persons, excluding certain lump sum payments and excluding capital gains.
  • BEA estimates that personal income for the US was $8.678 trillion in 2001, as compared to a CPS money income estimate of $6.446 trillion.2 Over 64 percent of this $2.232 trillion gap—$1.429 trillion—can be accounted for by differences in the income types that are included in the two measures, including the $982 billion of property income that is counted in personal income but not in CPS money income.
  • Half of the remaining $804 billion money income gap can be accounted for by BEA adjustments to proprietors’ income and wages and salaries for underreporting in BEA source data.
  • The Census Bureau has developed a number of alternative measures of money income that may measure economic well-being better than CPS money income. These measures remove taxes, add in-kind transfers, add realized capital gains or losses, and add the imputed return on equity in own home. The Census Bureau has found that a broadened definition of income results in a more equal distribution of income and tends to reduce the gaps between the incomes of traditionally high-and low-income groups.
  • An important issue in measuring income is whether certain income types should be captured when accrued or when disbursed. BEA personal income includes employers’ contributions into pension plans, while CPS money income includes pension disbursements. The BEA approach measures payments to factors of production, but the CPS approach better measures current capacity to spend.
  • Alternative measures of personal income and disposable personal income are considered at the end of the paper. These alternative measures might better serve users who need measures of the current capacity to spend or of the tax base. These proposed definitions also move toward the theoretical HSH concept of income, capturing incomes when disbursed from all types of retirement schemes and capturing realized capital gains.

1 – “Persons” in BEA’s state personal income consist of individuals and quasi–individuals who serve or act on behalf of individuals. Quasi–individuals consist of nonprofit institutions that primarily serve individuals, private noninsured welfare funds, and private trust funds.
2 – As discussed in footnote 5, the BEA estimate reported here differs from the estimate in the National Income and Product Accounts because it is a national total of State Personal Income.

What is income?

A variety of definitions of household income have been advanced in the literature. Many of these spring from the Haig-Simons-Hicks (HSH) concept of income as the maximum amount that can be consumed in a given period while keeping real wealth unchanged (Eisner, 1989). This very general concept, cited in the System of National Accounts at section 8.15, has been applied differently by macro-analysts interested in measuring the income of the macro economy and by micro analysts interested in the distribution of income.

Income measures produced by different government agencies can in part be distinguished by definitional boundaries of income. The debate about these boundaries is well summarized in the report of the Canberra Expert Group on Household Income Statistics (The Canberra Group, 2001). According to the Group, the debate has centered on the following three questions:

  1. Should income include only receipts that are recurrent (i.e., exclude large and unexpected, typically one-time, receipts)?
  2. Should income only include those components that contribute to current economic well-being or extend also to those which contribute to future well-being?
    Components of income that contribute to future well-being include employer contributions to pension funds and social insurance, interest and dividends earned on retirement-based assets and capital gains.
  3. Should income allow for the maintenance of the value of net worth?

The Canberra group recognized that there are two traditions of measurement that have influenced the estimation of income. The macro approach has its roots in national income accounting and in particular in the System of National Accounts (SNA). This approach aims at estimating income for the macro economy as a whole or for other geographic aggregates. The BEA measure of personal income comes from this tradition. In contrast, the micro approach to income measurement has its roots in micro economics and in particular the study of poverty and income distribution. The Census Bureau’s estimates of money income arise from this approach. Not with standing the different traditions, the Canberra Group notes that many of the conceptual difficulties in developing guidelines for income distribution statistics are the same or similar to the problems encountered in specifying guidelines for national income accounting.

The micro and macro approaches differ in whether they stress the type of income or the means of payment (Harrison, 1999). The macro approach categorizes income according to the type of transaction giving rise to an income flow without regard to the means of payment. The types of transactions identified in the macro approach include income generated in the course of production, from the distribution of property income, or from current transfers.

The micro approach focuses on the means of payment, without regard to the how the income flow is generated. According to the Canberra Group, the definition of income in the micro approach is driven mainly by what the individual perceives to be an income receipt of direct benefit. Such an approach implies that it is current economic well-being, as opposed to future well-being, which is of interest to the microanalyst. The recipient may be scarcely aware of income components that contribute to future economic well-being (such as contributions to pension plans). Therefore, in addition to there being an issue as to whether these should be included in income, there is the practical difficulty of collecting such information from survey respondents.


Two of the most widely used measures of household income are BEA’s personal income and the Census Bureau’s money income. These two measures differ in the scope of individuals covered, in the income items included, in the sources of the data and in the extent of disaggregation of the estimates. This section will discuss the general definitions, sources and uses of these two measures, while the next section presents a reconciliation of aggregate income estimates as a means of indicating the nature and size of differences.3

Personal income and disposable personal income

Personal income is the income received by persons from participation in production, from government and business transfer payments, and from government interest. Personal income includes income received by non-profit institutions serving households, by private non-insured welfare funds, and by private trust funds.4 Income from production is generated both by the labor of individuals and by the capital that they own. Private income not earned in production, such as from capital gains or the sale of assets, is excluded. Personal income is calculated as the sum of wage and salary disbursements, employer contributions for employee pension and insurance funds, proprietors’ income, property income (personal interest, dividend and rental income), and transfer payments to individuals, less personal contributions for social insurance.

Disposable personal income is personal income less personal tax payments. While personal income does not include capital gains realized through the sale of assets, personal income taxes do include the taxes paid for these capital gains.

Personal income and disposable personal income are released by the BEA both as aggregate and as per capita estimates for differing geographic areas and time periods. Estimates are not available according to demographic characteristics of individuals.

Estimates of personal income are based primarily on data from administrative records and from censuses and similar surveys. The data from administrative records may originate either from the recipients of the income or from the source of the income. The most important sources of these data include the state unemployment insurance programs, the social insurance programs of the Center for Medicare and Medicaid Services and the Social Security Administration; the Federal income tax program of the Internal Revenue Service, veterans benefits programs, and military payroll systems of the U.S. Department of Defense.

The data from censuses are mainly collected from the recipients of the income. The most important sources of census data are the Census of Agriculture, which is now conducted by the U.S. Department of Agriculture (USDA), and the Census of Population and Housing, which is conducted by the Census Bureau. Some estimates are based on data from other sources. For example, the USDA’s national and state estimates of the income of all farms constitute the principal basis for BEA’s national and state estimates of farm proprietors’ income. The USDA uses sample surveys, along with census data and administrative–records data, to derive its estimates.

State personal income estimates are used widely in the public and private sectors to study economic trends for States and regions and to measure and track the levels and types of income that are received by the people who live or work in a State. Federal Government agencies use the estimates as a basis for allocating $167 billion and for determining matching grants. Federal agencies also use the estimates in econometric models, such as those used to project energy and water use. State governments use the estimates in econometric models to project tax revenues and the need for public services. Many states have set constitutional or statutory limits on State government revenues and spending that are tied to State personal income or to one of its components. The estimates are also used in market and economic research.

Census money income

The Census Bureau collects income data on several major surveys, including the Annual Social and Economic Supplement (ASEC) of the Current Population Survey (CPS), the Survey of Income and Program Participation (SIPP), the decennial Census and the American Community Survey (ACS). The CPS is the source of official national estimates of poverty and the most widely used source of annual national income estimates.

The CPS measure of money income is defined as total pre-tax cash income earned by persons, excluding certain lump sum payments and excluding capital gains. It includes money wages and salaries, self-employment income, property income (dividends, interest and rents), money transfer payments from a variety of government and private welfare and social insurance schemes (such as social security, unemployment and workers’ compensation, and public assistance), private and government retirement income, interpersonal transfers (such as alimony and child support) and other periodic income.

Unlike BEA’s measure of personal income, CPS money income excludes employer contributions to government employee retirement plans and to private health and pension funds, lumps-sum payments except those received as part of earnings, certain in-kind transfer payments—such as Medicare, Medicaid, and food stamps—and imputed income.5 Money income includes, but personal income excludes, personal contributions for social insurance, income from government employee retirement plans and from private pensions and annuities, and income from interpersonal transfers, such as child support.

The Census Bureau releases estimates of household money income as medians, percent distributions by income categories and on a per capita basis. Estimates are available by demographic characteristics of householders and by the composition of households.

Census money income estimates are based on the Annual Social and Economic Supplement (ASEC) of the CPS. Data are collected from a sample of households by means of a structured questionnaire. For each person in the sample age 15 years or older, the ASEC asks questions about the amount of money income received in the previous year from up to 50 different income sources. In 2002, survey responses were obtained for approximately 78,000 households. While data collectors attempt to collect data directly from each eligible household member, proxy reporting by other household members is approximately 50 percent. According to the Census Bureau, this may introduce non-sampling error because respondents may provide less accurate information on other members of the household than about themselves.

As mentioned previously, the CPS is the source of official U.S. poverty estimates and the income and poverty measures are widely used as barometers of economic well-being for the Nation. In addition to their importance to researchers and policymakers, income and poverty data from the CPS are also used in federal funding formulas that allocate billions of dollars annually to localities based on differences in economic well-being. For example, the State Children’s Health Insurance Program (SCHIP) allocates approximately $4 billion annually based on CPS-derived figures on the number of low-income uninsured children in each state. Also, the Title I Program uses CPS poverty data to allocate $14 billion dollars annually to school districts.

3 – A third widely used measure of income is IRS adjusted gross income (AGI). For a comparison of BEA personal income and IRS AGI, see Ledbetter (2004).
4 – Mead, McCully, and Reinsdorf (2003) identify the following 5 categories of nonprofit institutions serving households:

    1. Religious and welfare, including social services, grant-making foundations, political organizations, museums and libraries, and some civic and fraternal organizations
    2. Medical care;
    3. Education and research;
    4. Recreation, including cultural, athletic, and some civic and fraternal organizations; and,
    5. Personal business, including labor unions, legal aid, and professional associations.

5 – Imputed income is the market value of certain transactions that do not occur in the market economy or that are not observable in BEA data. BEA’s measure of imputed income includes pay-in-kind in the form of meals and lodging, the rental value of owner-occupied housing, the value of farm products consumed at home by the producers, the value of investment income earned on life insurance, and the value of services provided to persons by depository institutions without an explicit charge.


This section presents a reconciliation of aggregate estimates of BEA personal income and CPS money income. BEA’s national estimate of personal income derived from state personal income (SPI) estimates is converted to an “SPI-derived money income” estimate by adding and subtracting income types to bring personal income to the same scope as CPS money income.6

BEA estimates that state personal income for the US was $8.678 trillion in 2001, as compared to a CPS money income estimate of $6.446 trillion. Sixty-four percent of this $2.232 trillion gap—$1.429 trillion—can be accounted for by differences in the income types that are included in the two measures (see Table 1).

Personal income contained $2.241 trillion in 2001 that was not in CPS money income. Personal income exceeds money income in part because the former includes not only income received by individuals but also income received on behalf of individuals. In 2001, $982 billion in property income (dividends, interest and rents) was received on behalf of individuals by pension plans, nonprofit institutions serving households, and fiduciaries. Personal income also contains other income categories not in CPS money income. Most notably, personal income included $563 billion in employer contributions for employee pension and insurance funds and $593 billion in transfer payments, mostly non-cash, like Medicaid, food stamps, and energy assistance.

SPI-derived money income in 2001 included $813 billion not in personal income. Almost half (44 percent) of that—$360 billion—came from disbursements of retirement income benefits.7 Money income also included $372 billion in personal contributions to social insurance (largely social security) that was deducted from personal income.

While not affecting the total gap between income estimates, BEA and the Census Bureau categorize some types of income differently. The principal difference is the treatment of S corporation profits. Shareholders of S corporations report their share of company profits (whether distributed or not) on their individual tax returns. BEA classifies as dividends all S corporation profits distributed to shareholders, regardless of whether the shareholders are employees of the corporation. Census money income treats these profits as dividends when they are received by non-employee shareholders, but treats them as wage and salary income to shareholder-employees. $189 billion was reallocated from dividends to wages and salaries to make the personal and money income estimates comparable. Another difference occurs in the treatment of distributed earnings from money market accounts. These are classified as interest by BEA and dividends by the Census Bureau; therefore, $52 billion was reallocated from interest to dividends in this reconciliation.

The Money Income Gap by Type of Income for 2001

After adjusting for differences in income types included in the two measures, SPI-derived money income still exceeds CPS money income by $804 billion. What accounts for this “money income gap?” Some insights can be gleaned by comparing the gap by type of income as shown in Table 1, line 38. The gap occurs primarily in wages and salaries, proprietors’ income, personal dividends, personal interest, social security, and other retirement and disability income.

The income category experiencing the largest money income gap is proprietors’ income. BEA’s estimate of SPI-derived proprietors’ money income (that is, BEA’s estimate of proprietors’ income adjusted to include CPS money income categories) is $630 billion in 2001, as compared to a reported CPS money income estimate of $329 billion. The nearly $302 billion gap in these estimates can be fully accounted for by BEA misreporting adjustments.

BEA uses Internal Revenue Service (IRS) tabulations of sole proprietorship and partnership income tax returns as the primary source for nonfarm proprietors’ income estimates. IRS tax return data do not include the income of “nonfilers,” that is, those who are not required to file tax returns or those who illegally evade filing. Further, some filers underreport income. While the IRS can verify certain types of income reported on individual returns, such as wages, interest, and dividends, by matching tax return information with corresponding third party reports, document matching is ineffective for verifying business income.

BEA adjusts for income earned, but not reported on tax returns, by adding an estimate of “misreporting”. The adjustment is an extrapolation based primarily on the 1988 Taxpayer Compliance Measurement Program (TCMP) audit, 1999 exact match study, and current activity indicators, such as the Census Bureau’s value of new construction. Proprietors’ income has been consistently underreported to the IRS. The last TCMP audit estimated that proprietors’ actual income was more than double levels reported on tax returns (Landefeld and Fraumeni, p. 33). The 2001 proprietors’ income misreporting adjustment accounts for 42 percent of proprietors’ state personal income and 49 percent of SPI-derived proprietors’ money income in 2001.

Although the Census Bureau does not make a similar adjustment to money income estimates, BEA includes the misreporting adjustment in its derivation of SPI-derived money income in the belief that it is the best available approximation of actual unreported proprietors’ money income. However, respondents who underreport to the IRS may also underreport in a voluntary survey such as the CPS. At $308 billion in 2001, the proprietors’ income misreporting adjustment fully accounts for the $302 billion proprietors’ money income gap that year.8

The “other retirement and disability income” category constitutes another major source of the total money income gap. This income category consists primarily of retirement benefits from private, government, military, railroad, and individual funds. It also includes payments to beneficiaries of state temporary and disability insurance, black lung, pension benefit guarantee, and private accident insurance disability funds. It does not include either Social Security or workers’ compensation. Large both in percentage and dollar terms, at $360 billion SPI-derived money income in this category exceeds the CPS level of $253 billion by 42 percent.

SPI-derived money income significantly exceeds CPS money income in every government retirement income category. SPI-derived pension benefit figures are 49 percent higher than CPS money income for federal retirement and 91 percent higher for state and local government. BEA estimates in these categories are based on data from the Monthly Treasury Statement and the Census Bureau. Estimates of individual annuity benefits also vary widely. The BEA figure, based on data from the National Association of Insurance Commissioners, exceeds the CPS estimate by 481 percent.

CPS and SPI-derived wage and salary money income differ by only 3 percent, but this small percentage represents $158 billion. BEA includes a $104 billion adjustment for wage and salary income earned in the underground economy, which estimates cash wages from legal activities that are earned “off the books.”9 Although the CPS is designed to include these wages, as with proprietors’ income, individuals who don’t report or underrepresent income to the IRS or other agencies may be unlikely to fully report these wages on a voluntary survey such as the CPS, despite assurances of confidentiality.

Census Bureau research by Roemer (2002) comparing CPS wage data with administrative earnings records from the Social Security Administration’s Master Earnings File has shown that the CPS underestimates wages of part-year, part-time workers. Because the CPS does not survey military personnel living on a U.S. post without family, wages earned by military personnel from secondary jobs in the civilian sector would not be included. Underreporting by proxy reporters especially of secondary jobs may also be a factor. Finally, since the reference period for the CPS ASEC is the past calendar year, respondents may fail to recall small amounts and payments that are received infrequently. This might affect not only the reporting of wages for short duration jobs, but also the reporting of other small income components.

Within property income, CPS and SPI-derived money income differ substantially in the personal interest and dividend income categories. At $259 billion, SPI-derived personal monetary interest exceeds the CPS level of $188 billion by 38 percent. In 2001, taxable and tax-exempt interest reported on individual tax returns totaled $243 billion.10 Given the similarity between the BEA estimate and level of personal interest income reported to the IRS, the interest money income gap appears due to underreporting on the CPS survey. This may result in part from incomplete information provided by proxy reporters.

SPI-derived dividend income is $148 billion, 69 percent higher than the CPS dividend income level of $88 billion. Dividend income reported on individual tax returns for 2001 totaled $116 billion. The dividend money income gap occurs at least in part due to CPS underreporting, since the CPS level falls $28 billion below the IRS reported level. SPI-derived interest may be expected to exceed the IRS level since individual tax return data do not include the income of nonfilers, but it is unclear whether this fully explains the $32 billion by which the SPI derived dividend figure exceeds the data from individual income tax returns.

Within transfer payments, the major gap occurs in Social Security. CPS money income reports Social Security as $376 billion. At $425 billion, SPI-derived Social Security (based on data from the Social Security Administration) exceeds the CPS level by $49 billion and 13 percent.

6 – The reconciliation uses BEA’s national estimate constructed from state personal income (SPI) rather than the national estimate from the National Income and Product Accounts (NIPA’s). The main differences between the NIPA and SPI estimates of personal income stem from the treatment of the income of U.S. residents who are working abroad and the treatment of the income of foreign residents who are working in the United States. The national total of the state estimates of personal income consists of only the income earned by persons who live within the United States, including foreign residents working in the United States. This is closer to the scope of the CPS, though the CPS excludes certain individuals residing in the US, including military on US posts without family, the institutionalized, decedents in the reference year, and child workers under 15 (agricultural workers can legally be as young as 10).

7 – To produce SPI-derived retirement money income, estimates of lump-sum payments were removed from BEA’s national retirement benefit estimates. While lump sum payments (including withdrawals) constitute a negligible portion of public retirement payments, they appear to comprise over half of private retirement payments. BEA national private pension benefits are based primarily on Department of Labor (DOL) tabulations of Form 5500 reports filed by employers and data compiled by the American Council of Life Insurance (ACLI). BEA estimated private pension lump sum payments using the 1998 Form 5500 ratio of benefits from defined contribution plans to total private retirement benefits applied to the 2001 BEA national private pension benefit estimate. Although the unadjusted BEA national estimate of private pension benefits was substantially greater than the CPS figure, after the removal of lump sum payments the SPI-derived money income measure exceeded the CPS figure by only $6 billion or 6 percent.

8 – Given that the two primary studies on which the misreporting adjustment is based have not been conducted in recent years, the reliability of the 2001 misreporting adjustment may be questioned. The IRS has replaced the TCMP with the National Research Program (NRP), which has as part of its mandate the measurement of filing and reporting compliance. NRP audits were begun in 2002 and will provide a more accurate picture of current filing and reporting gaps when results become available. (U.S. Internal Revenue Service, 2002).

9 – For a fuller discussion of the underground economy see Carson (May and July 1984) and Parker (1984).

10 – See “Individual Income Tax Returns, Preliminary Data, 2001,” SOI Bulletin, Winter 2002-2003, p. 137.



The traditional money income concept is limited and does not provide a completely satisfactory measure of economic well-being. For example, money income (unlike BEA’s disposable income concept) does not include the effects of taxes and, therefore, does not reflect the effect of tax law changes on economic well-being. Similarly, the official measure of money income excludes the effect of noncash benefits (such as employment-related group health insurance and food stamps), which enhance economic well-being and are also included in BEA’s personal income. The Census Bureau has a fairly long history of producing estimates that address these shortcomings.

Since the early 1980s, the Census Bureau has published analysis showing the effect of using a broadened income definition on measures of economic well-being. Currently, annual Census Bureau reports on income and poverty show the effect of using an income measure that includes the effect of noncash benefits and taxes on the distribution of income, prevalence of poverty, and level of income inequality based on the 17 income definitions as summarized below:

Definition 1: official money income

Definition 1b: definition 1 plus capital gains/losses less taxes

Definition 2: definition 1 less government cash transfers

Definition 3: definition 2 plus capital gains/less capital losses

Definition 4: definition 3 plus the value of employment-related health benefits

Definition 5: definition 4 less Social Security payroll taxes

Definition 6: definition 5 less federal income taxes (excluding the Earned Income Tax Credit)

Definition 7: definition 6 plus the Earned Income Tax Credit

Definition 8: definition 7 less state income taxes

Definition 9: definition 8 plus non-means-tested government cash transfers

Definition 10: definition 9 plus the value of Medicare

Definition 11: definition 10 plus the value of regular-price school lunches

Definition 12: definition 11 plus means-tested cash transfers

Definition 13: definition 12 plus the value of Medicaid

Definition 14a: definition 13 plus the value of other means-tested government noncash transfers less Medicare and Medicaid

Definition 14: definition 13 plus the value of other means-tested government noncash transfers

Definition 15: definition 14 plus net imputed return on equity in own home

Obviously, the construction of 17 definitions of income was not based on the premise that each of these definitions represented a viable income concept. Rather, the construction of so many income definitions was to facilitate the analysis that examines which components of a broadened income measure are most responsible for the significant changes in income summary measures as one transitions from the money income concept to an expanded definition of well-being. That said, there are several expanded income definitions that the Census Bureau has found useful to track trends and differences between groups. For example, the 2002 CPS income report (U.S. Bureau of the Census, 2003) highlighted four definitions of income in addition to the traditional money income definition. These were definitions 1b, 14a, 14, and 15. It should be noted that in the 2002 income report for the first time these alternative income measures were featured in the main body of the report and presented along with the money income measures (in previous reports these figures were examined in supplemental report sections).


Clearly, an expanded definition of income has a significant effect on income and poverty summary measures. Looking at 2002 data for definition 15, for example, we see that while the median income is somewhat higher under the most comprehensive definition of income ($43,760 based on definition 15 vs. $42,409 based on money income), mean income under the most comprehensive definition is lower than money income and the distribution of income is substantially more equal under the expanded definition (see Table 1). The Gini index, for example was .400 under definition 15, 11 percent lower than the Gini index for money income.11 The percentage of aggregate income received by the top 20 percent of the income distribution was also lower (45.6 percent for definition 15 vs. 49.6 percent for money income). Census Bureau figures have consistently shown that government transfers have a much greater impact on lowering income inequality than the tax system. In 2002, for example, subtracting taxes and including the Earned Income Tax Credit lowered the Gini index by about 4 percent, while including transfers lowered the Gini index by around 17 percent.

As would be expected, the use of alternative income measures also has a significant effect on poverty measures. Using the same poverty thresholds as the official measure, the poverty rate based on the most broadened definition of income (definition 15) was 8.6 percent in 2002, 3.5 percentage points lower than the official poverty rate of 12.1 percent. Poverty rates increased between 2001 and 2002 based on both definitions of income.

It is also instructive to look at the effect of the use of alternative income definitions on the relationship of incomes between population subgroups. For example, under the money income definition, the median income in 2002 of households with householders that reported the single race of Black ($29,026) was 62 percent of the median of non- Hispanic White households in which the householder reported no other race. The comparable percentage under the broadest definition of income was 67 percent. Similarly, the use of a broadened definition of income reduces the gap between the median incomes of married-couple family households with children and households with a female householder, no husband present, with children (from 39 percent to 48 percent). Comparisons such as these show that the use of a broadened definition of income not only results in a more equal distribution of income, as might be expected it also tends to narrow the income differences between groups of households with traditionally high incomes and groups with lower incomes.

The Census Bureau plans on continuing to highlight alternative definitions of income because they offer a more comprehensive picture of economic well-being and are more sensitive to the effect of government tax and transfer policies than a money income concept. It should be noted that expanded definitions of income bring many complications, as noncash/tax values are not directly collected in the CPS and are therefore calculated. Thus, they are more prone to methodological changes that could conceivably make time series comparisons more problematic. For example, the Census Bureau’s goal is that the next release of after-tax income estimates (this fall) should incorporate a revised and improved tax model. But these improved estimates would be for calendar years 2002 and 2003 only. As the Census Bureau continues down the road of highlighting broadened income measures, the tradeoffs between the desire to continually improve methods and preserve the time series must be understood and factored into implementation decisions.

11 – The Gini index measures dispersion of income across an entire range and expresses it as a single statistic. At the extremes, 0 indicates perfect equality (everyone receives an equal share) and 1 indicates perfect inequality (one recipient or group receives all income).


As discussed previously, two important uses of BEA’s estimates of personal income, particularly at the state and local level, are to track spending capacity and to measure the tax base. There are alternatives to BEA’s data that can be used to address these needs. The IRS provides estimates of adjusted gross income and its major components at the state and county level, but these are available with a lag due in part to the need to wait for returns to be filed. The Census Bureau’s estimates of money income, and the alternative income measures in produces, are more timely, but the limited sample size of the CPS means that the Census Bureau only publishes two year moving averages by state.12 Thus, state and local users of income data often rely on BEA’s estimates of personal income, which are the most timely and comprehensive income estimates available at a detailed level of geography.

Alternative BEA measures of income may better meet user needs than does personal income. Personal income differs from a measure of the tax base, since it includes some nontaxable forms of income (e.g., employer contributions for pensions and health insurance) but excludes others (e.g., pension distributions and realized capital gains). Disposable personal income does not fully measure the capacity to spend, since it does not reflect either all money income flows available for spending or the accumulation of wealth that might be drawn down to support consumption. The following discusses how alternative measures of personal income and disposable personal income might be constructed so as to better meet user needs. This discussion is preliminary and will benefit from the input of the FESAC committee.

One alternative approach recognizes that the present scope of personal income is broader than households, since it also includes non-profit institutions serving households (NPISH’s). State and local area estimates of household income could be generated separately from NPISH’s, paralleling estimates that have been generated at the national level as the result of the latest comprehensive revision of the National Income and Product Accounts (NIPAs).13 The rationales for excluding NPISH’s are that their consumption patterns are different for those of households and that they are tax-exempt. Thus, state and local area estimates of BEA household income (and disposable household income) might better proxy for consumer spending capacity and the tax base.

In order to estimate household income at the state and local level, income and transfer payments from outside the personal sector would need to be split between households and NPISH’s. Further, income would need to reflect transfers between the household and NPISH sectors. Currently, transfers that NPISH’s receive from households – or make to them – are excluded from personal income because they are intrasector transfers in the consolidated accounts of households and NPISH’s. Estimates of household income would need to reflect transfers from NPISH’s.14

How different are estimates of personal income and household income? Data at the national level suggest that they are quite similar: $8.685 trillion for US personal income and $8.647 trillion for US household income in 2001 or a difference of less than 0.5 percent (Mead, McCully, and Reinsdorf, 2003). The similarity of the estimates stems from the fact that personal income in the form of property income and transfers that is attributable to NPISH’s is both relatively small and is largely offset by transfers from NPISH’s to households in the calculation of household income. However, it is conceivable that differences between personal and household income could be greater at the state and local area level, to the extent that donors to and recipients from NPISH’s are different individuals and live in different areas.

While the impact on estimates of income from excluding NPISH’s may be small, the previous reconciliation between BEA personal income and the Census Bureau’s money income indicates that there are other, more sizeable components of personal income that are not received directly by households. As previously noted, personal income includes employers’ payments into employee pension plans, but does not measure pension disbursements. Another alternative measure of personal income would remove the contribution items associated with pension plans and add back pension disbursements. Here “pension plans” refer to both defined benefit and defined contribution plans.

Constructing personal income based on pension disbursements rather than contributions can be rationalized in a variety of ways. In their paper on alternative measures of personal savings, Perozek and Reinsdorf (2002) note that pension funds are assigned to the personal sector as opposed to the business sector. Contributions from business into retirement accounts is compensation across sectors that is included in personal income, while disbursements from these accounts are transfers wholly within the personal sector that don’t increase personal income. Perozek and Reinsdorf argue that placing defined contribution plans in the personal sector is appropriate because they belong to employees. However, inclusion of defined benefit plans in the personal sector is more controversial, since employees are not entitled to all of the funds that accrue in defined benefit plans, but rather are entitled only to pensions based on a formula. If defined benefit plans were assigned to the business and government sectors, then personal income would be generated when pensions are disbursed, not when contributions are made into pension plans.

The Perozek and Reinsdorf argument applies only to defined benefit plans. A rationale for treating defined contribution plans in the same fashion must be sought elsewhere. There are two additional rationales that may apply to both defined benefit and defined contribution plans. First, if personal income (or, better, disposable personal income) is being used to measure current spending capacity and if there are liquidity constraints that limit borrowing against these plans, then pension plan disbursements measure increased capacity to spend better than pension plan contributions.15 Second, if personal income is being used to proxy for the tax base, then an income measure that includes taxable pension disbursements is preferred to a measure that includes non-taxed contributions.

BEA’s regional economics staff estimated personal income on a pension disbursement basis for the mid-1990s. This involved removing from personal income several pension related items and adding back an estimate of pension disbursements. The items that were removed from personal income included employee contributions to pension plans (such as 401(k) contributions) that are now included in wages and salaries, employer contributions that are now included in the category “employer contributions for employee pension and insurance funds,” and investment earnings on pension accounts (dividends, interest, and rent) that are currently included in property income.

Using pension disbursements in place of pension contributions and earnings lowers the estimate of US personal income for 1997 by $154 billion or 2.2 percent. Adjusted personal income was lower for all states except Florida. The states that experienced the largest gain in personal income shares were Florida, Arizona, Delaware and Michigan. Relative losers were DC, Maryland, Virginia, Alaska and Hawaii. All of these losing states have a large federal government presence, with large federal government contributions to pensions. Of course, these estimates reflect the current relationship between the number of retirees and working people in the US and in the states, a relationship that is expected to change with the retirement of the baby-boom generation. The impact of replacing pension accruals with disbursements in any given year will also depend on the strength of the stock market, as required employer contributions into pension plans decline with the appreciation of pension plan assets.

There are several source data problems with estimating personal income on a pension disbursement basis. These problems are an issue at the national level and they are even more acute at the subnational level where less source data are available or where, owing to small sample issues, the source data might be less reliable. One problem is that the unemployment insurance data used to generate wage and salary estimates do not break out the portion that employees contribute to pension plans. BEA’s preliminary research used data from the Office of Personnel Management (OPM) and the Census Bureau for the government sector and IRS Form 5500 data for the private sector to break out employee contributions at the national level. Wages and salaries were used to distribute the national estimates to the state level. The drawback of this approach at the national level is the timeliness of the IRS data from Form 5500. At the state level, employee contributions in different industries have different geographic coverage that the use of private wages will not capture. In addition, employees can contribute a variable amount up to a certain limit, and the state wages will not reflect that option.

Another important issue with regard to the source data available for estimating pension disbursements concerns pension rollovers. Lump sum distributions frequently occur for cash balance and defined contribution plans when employees leave their firms. Whether these distributions are rolled over into a new retirement plan or whether they are retained for spending or paying off debt is a crucial distinction for measuring personal income on a pension disbursement basis. According to data from the Survey of Income and Program Participation, Moore and Muller (2002) determined that, while “historically most distributions have not been rolled over, the majority of the dollar value of all distributions has been rolled over. For example, 73 percent of distributed dollars were rolled over in 1993, and 79 percent were rolled over in 1996” (page 33). Similarly, Sabelhaus and Weiner (1999) estimate from IRS micro data that 70 to 77 percent of distributions were rolled over in 1995 (pages 600-601).

BEA reviewed state and national information from IRS’s Compliance Research Information System (CRIS), a sample of seven percent of all 1040 filers. BEA estimated that total rollovers in 1999 were approximately $227 billion. One of the problems with the CRIS database is that it has not been edited. The quality of the database must be reviewed over time. Large errors were noticed for some categories at the state level. In addition, the database excludes nonfilers of IRS Form 1040.

BEA has recently been working with the IRS Statistics of Income Division to provide BEA with state level sample data on information returns that will allow the Bureau to adjust for nonfilers and to determine type of distributions from the 1099R. Data on distributions from pension and IRA accounts, and their disposition (whether rolled-over or paid to the individual), may be gleaned from taxpayer information returns (Forms 1099R and 5498). Administrative records and Census Bureau survey data are available for lump sum payments, refunds, and transfers from government retirement plans, but these government payments constitute only a small portion of all lump sum retirement payments.

The foregoing discussion considers including pension disbursements in an alternative definition of personal income, where pensions are limited to defined benefit and defined contribution plans. However, there are other tax-preferred schemes, such as, IRAs and annuities that facilitate savings to provide income during retirement. BEA currently treats contributions to these other schemes as personal savings. One can envision an alternative definition of personal income that includes not only pension distributions, but also distributions from IRAs and other tax-preferred schemes, that is, that reflects total retirement payments.16 Constructing this alternative measure would involve removing contributions to these other tax-preferred schemes, removing the property income earned on the schemes, and adding total retirement payments. As with pension rollovers, care would need to be taken to net out distributions that are rolled over into other similar tax preferred accounts.

The rationales for including disbursements from other tax-preferred retirement schemes parallel those given for including pension disbursements—an income measure capturing these disbursements may better measure current spending capacity and the tax base.

It is important to note that the alternative measures of personal income proposed above do not strictly fit the HSH concept of income. Specifically, when retirement assets are disbursed and used for consumption, they may lead to a decline in retirement asset balances. Thus, consumption may be supported by a decline in net worth, in contrast to the HSH concept of maintaining net worth.

The foregoing has discussed the inclusion of distributions of pensions and other retirement schemes into alternative measures of personal income. All of these schemes have the attribute that they are tax-preferred and were designed to provide income in retirement. But, private savings more generally may be viewed as generating assets that provide income during retirement or more generally that provide income for consumption smoothing over the life cycle. A more expansive definition of income might include the money flows from realized capital gains or the increase in wealth associated with unrealized capital gains. These more expansive personal income definitions come closer to measuring changes in net wealth and hence closer to the HSH concept of income.

Data on changes in the net worth of households and nonprofit organizations are published quarterly by the Federal Reserve Board in the Flow of Funds (see Table R 100). An examination of this table makes clear some drawbacks of a very broad measure of income that includes unrealized capital gains. Specifically, such a measure would be extremely volatile and would not be predictive of spending patterns. For example, the net worth of households and nonprofit organizations increased by over $4.9 trillion, or 63 percent of US personal income, in 1999, largely through holding gains on assets. In contrast, net worth declined by about $1.4 trillion in the second quarter of 2002 and it declined by $1.6 trillion in the third quarter of 2002. Such increases and decreases were not accompanied by corresponding changes in consumption. Thus, a very broad measure of income defined as the change in net worth does not seem useful for measuring spending capacity. Further, since unrealized capital gains are not taxed, such a broad measure does not serve as a good proxy for the tax base. A more useful expanded definition of income might include only realized capital gains.

12 – The Census Bureau also produces annual estimates of median household income for states and counties, based on models using data from the ASEC, the decennial census, administrative records, and BEA’s personal income. The estimate are available with greater lag than the state household income tabulations from the CPS.

13 – See Mead, McCully, and Reinsdorf (2003).

14 – Household transfer payments to non-profits are treated as household outlays in the NIPAs.

15 – In fact, however, defined contribution plans frequently contain provisions that allow employees to borrow on these accounts, so that the liquidity constraint assumption is weak.

16 – It should be noted that while IRAs were created to provide retirement income, the funds may be withdrawn before retirement for a variety of purposes. Thus, it is not strictly true that an alternative measure of personal income that incorporates IRA disbursements is reflecting only retirement payments.


This paper has compared two of the more widely used measures of household income—BEA’s personal income and the Census Bureau’s CPS money income. It has also presented alternative estimates of money income developed by the Census Bureau to better measure economic well-being and it has discussed how alternative BEA measures of personal income might be developed that better measure the capacity to spend or the tax base. It is clear that there is not one single definition of household income that can serve all purposes. Instead the BEA and the Census Bureau have and will continue to provide an array of measures that address different user needs.


After adjusting BEA personal income to the same scope as CPS money income, the BEA estimate remains $806 billion higher. The paper surmises that at least half of this might be due to CPS underreporting. Is this a reasonable surmise or could the BEA estimate be too high? Does possible underreporting in the CPS have any implications for income distribution measures?

Should BEA pursue generating alternative income estimates along the lines discussed in the paper? That is, should BEA generate alternative income measures that include retirement income disbursements and possibly realized capital gains? How well would these alternative measures address user needs to better measure the capacity to spend or the tax base?

Are there other user needs not discussed in the paper that could be met by alternative BEA measures of income? If so, what alternative definition of income might be proposed to meet these needs?

The Census Bureau has been producing estimates of income and poverty based on alternative definitions of income for a long time and, within the latest income report release, has begun to highlight some of these measures much more than previously. Are the measures that the Census Bureau has begun to highlight appropriate? Are there others they should highlight as well?

For future Census Bureau income reports, are there other components of well-being that should be added to Census Bureau research into broadened income measures? For example, alternative measures now examine the effect of one type of mandatory expense (taxes) on income. Should future research include the effect of other “mandatory” expenses, such as work-related or health-related expenses?

Should the Census Bureau consider using model-based techniques based on the relationships between administrative and survey data to produce improved experimental estimates of household income?

Is the reconciliation of BEA’s personal income and the Census Bureau’s money income of sufficient value that it should be updated and published on a regular basis? Is there value in extending the reconciliation to the state level?

Farm Data Procedures


The BEA Benchmark I-O tables provide us with Output, Employee Compensation (EC), and Gross Operating Surplus (GOS) by IMPLAN Sector. They also provide us with production functions for each IMPLAN Sector. The data is available for all 14 farm Sectors. However, the data does not separate GOS into its component parts (Proprietor Income (PI) and Other Property Income (OPI)), nor do they provide Employment estimates. We use the latest BEA Benchmark data with other data sources from the same year to create ratios that are used in the annual data creation process:

  1. Output-per-EC by farm Sector: The BEA Benchmark provides EC and Output by farm Sector.
  2. EC-per-W&S Worker by farm Sector: The BEA Benchmark does not provide any employment estimates. Thus, to get ‘Benchmark’ W&S Employment by farm Sector, we distribute benchmark-year BEA REA “All Farm” W&S Employment based on benchmark-year BLS CEW W&S Employment by farm sector and then calculate ratios from these estimates. While CEW has more sector detail than REA, we use the REA total W&S Employment figure since the CEW data only cover roughly 90% of W&S employment for the farm sectors. We only use the CEW data for distribution of the REA W&S Employment value. We then combine these W&S Employment estimates with the BEA Benchmark EC data to calculate EC-per-W&S Worker ratios by farm Sector.
  3. PI-per-Output by farm Sector: The BEA Benchmark does not directly give us PI by farm Sector, but rather gives us GOS by farm Sector. We use an agriculture-wide average ratio between PI and GOS (using PI from the 2007 BEA REA data, which has just one “All Farm” Sector) to estimate Benchmark PI by farm Sector. These ratios are then used to distribute the latest BEA REA “All Farm” PI data amongst the 14 farm Sectors. We then combine these PI estimates with the BEA Benchmark Output data to calculate PI-per-Output ratios by farm Sector.
  4. Output-per-Proprietor by farm Sector: Again, the BEA Benchmark does not provide any employment estimates. We estimate Benchmark Proprietor Employment by distributing benchmark-year BEA REA farm proprietor employment based on the farm count per farm sector from the latest Census of Agriculture. We then combine these Benchmark Proprietor Employment estimates with the BEA Benchmark Output data to calculate output-per-proprietor ratios by farm Sector.


  1. Each year, we obtain estimates of agricultural Output by commodity by state from USDA’s Economic Research Service (ERS) and National Agricultural Statistics Service (NASS). The data generally are current and empirical (i.e. based on observation and survey, rather than trend extrapolations), so we begin with Output data.
  2. ERS and NASS do not cover some states with low production, and can omit certain crops with low production. ERS often will report some of those low-production crops in its “miscellaneous” category. We attempt to reallocate those low-production crops into the appropriate Sector. We also use the Census of Agriculture to estimate crop production for states whose production is not reported by ERS or NASS. For both of these reasons, our estimates for a particular crop Sector often are somewhat larger than those reported by ERS or NASS.
  3. As of IMPLAN’s 2015 data set, we no longer control agricultural estimates to BEA for several reasons. The BEA release of farm cash receipts was released after we produced agricultural estimates. Additional discussions with BEA revealed that they primarily use ERS data, which is one of IMPLAN’s primary sources, so controlling to BEA estimates adds relatively little value. Furthermore, BEA’s commodity-level estimates are for cash receipts, which excludes crops put into inventory and home consumption, both of which drive intermediate expenditures. However, one benefit of controlling to BEA that we wanted to keep is that it theoretically corrects for ERS’ tendency to overestimate the Output of the Miscellaneous Crops Sector; therefore, we implemented an adjustment factor based on the ratio of BEA Benchmark Output to benchmark-year ERS output. We do, however, compare our Output estimates to estimates from NIPA, BEA REA, ERS, and the BEA Industry Series to ensure that they are reasonable.
  4. These state values are distributed to the counties by using the ratio of county physical production to state physical production from the latest Census of Agriculture (which comes out every 5 years). For example, if County A has 10% of the state’s corn sales (or acres if sales is not available, and farms if acreages is not available), then it gets 10% of the state’s Annual Cash Receipts value for corn.
    • For non-disclosed counties in the Census of Agriculture, we multiply the average value of a particular commodity per unit of volume (or per farm in the absence of volume data) in the state that produces that commodity by the county-level data that produce that commodity (number of farms is always disclosed), giving us an estimate of total value of that commodity at the county level. We adjust estimates based on these ratios to control totals when possible.
    • Annual Cash Receipts data are not available for all crops at the state level. In these cases, the average ratio of state production to national production from the Census of Agriculture is used, and controlled to the state total for a higher-level aggregation, if that aggregation accounts for non-disclosed crop values.1

1 ERS and NASS occasionally change their reporting for aggregate categories between a) including the sum of non-disclosed values for subcategories in their aggregate category values and b) making the aggregate value simply the sum of disclosed children. We adjust our processes accordingly.


There is no data source for employment by agricultural commodity, even at the U.S. level.

  1. We make our first estimates of employment and income by IMPLAN Sector applying the Benchmark ratios to annual Output by commodity (both described above).
  2. We control those estimates of employment and income to annual BEA REA estimates of “All Farm” employment and income (both for W&S workers and proprietors).
  3. Each state’s output, employment, and income are forced to sum to the U.S. totals, after which the counties’ values are forced to sum to the state values.


The BEA releases state-level GDP data, at approximately the 3-digit NAICS level, that includes the break-out of GDP into EC, TOPI, and GOS. These data collapse NAICS 111, Crop Production, and 112, Animal Production, into a single “Farms” Sector. GOS consists of Proprietor Income (PI) and Other Property Income (OPI), so OPI for 3-digit NAICS is derived by subtracting our estimates of PI (described above) from GOS. These 3-digit control values are distributed to the IMPLAN Sectors based on the latest BEA Benchmark’s characteristics for GOS and TOPI and by using data from the ERS Agriculture Resource Management Survey (ARMS) data as described below.


Updated source data and method for forecasting lagged state GDP data for farm sectors (IMPLAN Sectors 1-14):

  • At the national level, we now use growth rates from NIPA table 7.3.5, which reports national components of value added in the current Implan reference year. We apply the growth rates to BEA REA data, to maintain consistency with REA definitions and concepts.
  • At the state level, we use growth in total farm output rates to project value added growth. We have empirical sources for current-year agricultural output by state. This has the result of better approximating future BEA estimates of farm value added. Previously, we used only EC, which was extrapolated from REA total farm EC and current-year output estimates.
  • These state-level projections are then controlled to the national projections.

In Data Year 2015, we also incorporated USDA ERS Agriculture Resource Management Survey (ARMS) data to estimate components of value added by commodity at the national and state levels:

  • ARMS data report sub-components of value added components for certain commodities and certain states.
    • Only major agricultural states are included in ARMS. For non-covered states, we form initial estimates of OPI and TOPI by farm Sector according to national ratios of OPI and TOPI to state-level farm data. The national ratios are based on ARMS data, as described below.
    • ARMS reports government payments, i.e., subsidies, real estate and other property taxes (a large component of gross TOPI for farm sectors), interest (a component of OPI), and depreciation (another component of OPI).
    • Certain ARMS commodities line up well with IMPLAN commodities, including soybeans (over 90% of IMPLAN’s Oilseeds sector), wheat and corn (collectively over 90% of IMPLAN’s Grains sector), dairy, cattle, poultry, and other livestock. We make minor adjustments to oilseeds and grains to scale the ARMS values to account for non-covered commodities, e.g., canola in the Oilseeds sector and rice in the Grains sector. We do not use ARMS data in cases where the commodity classifications do not align well with IMPLAN Sectors.
  • Since ARMS data covers only certain parts of value added components, we don’t use the ARMS data outright but rather to distribute REA “Farm Total” value added components to IMPLAN farm Sectors.
    • For example:
      • First, we estimate, according to ARMS, the share of all real estate taxes in a state that go to dairy. Let’s say it is 5%.
      • If the ARMS data is lagged with respect to the IMPLAN reference year, we then use state-level output by commodity data from the lagged year and the current year to estimate the change in dairy’s share of the state’s total farm output. If dairy increased from 10% to 20% of the state’s agricultural Output, we multiply the 5% share of real estate taxes by 2, for a 10% share. We check for and downwardly adjust abnormally high or low changes in shares.
      • We then take 10% of our projected REA Gross TOPI value as the Gross TOPI that belongs to dairy. We perform an analogous calculation for subsidies, and then calculate net TOPI from that.
  • After using ARMS data where available, we use our pre-existing method of making initial estimates of value added components based on 5-year BEA benchmark data for farm commodities where ARMS coverage does not align with IMPLAN Sectoring.
  • The final estimate for farm GDP for a state is controlled to our projection of REA farm value added, which is consistent with past practices.



Generally, we prefer our own estimates of employment to QCEW, BEA, and the Census of Agriculture for a variety of reasons. Among those reasons: QCEW does not cover proprietors, which compose a significant share of farm employment; QCEW misses some W&S employment; BEA REA has employment data only at the “farm” level of detail, though it includes proprietor employment; the Census of Agriculture releases employment data only every 5 years and measures employment differently than our other data sources: it measures the number of unique human beings who worked on a farm as opposed to the “jobs” those humans filled.  For example, if a farm had 6 humans who worked 2 months each, sequentially in a year, the Census of Agriculture would report that as 6 jobs, whereas in other data sources (and in IMPLAN), this is considered just one job – one job that happens to be filled by 6 different temporary workers.  Our data attempts to correct for these omissions and inconsistencies. Since the agriculture data are, to a large extent, derived, analysts who have local agriculture data that also correct for these omissions and inconsistencies (e.g., from a survey) are encouraged to use their data when building their IMPLAN models. However, in the absence of such data, we encourage people to use IMPLAN’s estimates.

Special Sectors for Employment Data

There are several sectors that are not covered by the CEW data that require different data points and additional techniques to capture.

These special Sectors include the following:

  • Farm sectors
  • Construction sectors
  • Government sectors




The primary data sources for state-level agricultural output by sector are the NASS Value of Production and ERS Annual Cash Receipts data sets, both from the USDA.  The county-level data from these sources are not consistent enough for our use; thus, we use data from the latest Census of Agriculture to estimate county-level farm sector output.  Census of Agriculture data is also used to estimate state-level values not disclosed by NASS or ERS.

More details can be found in this article.

Employment and Income

BLS CEW does have estimates of wage and salary employment for farms with about 90% coverage. However, the CEW data for farm sectors are particularly difficult to integrate because BLS CEW data establishments are not reclassiffied year to year, while farmer commodity production is. For example, a given farmer will plant either corn or soybeans (2 separate IMPLAN sectors) based on that year’s prices and/or how late in the year they are able to plant their crops. Also, the CEW data does not include propietors.

The BEA’s Regional Economic Accounts (REA) program estimates county-level employment and income data, including proprietors, but these are farm totals that are not broken down by type of agricultural commodity. Therefore, we have developed procedures to generate first estimates of employment and income by commodity and county based on annual output values (described above) and various ratios from the Census of Agriculture and state-level data. These first estimates are controlled to the total farm employment value given by the REA data, as well as to higher geographic levels. 

Some of the state and county farm sectors are subject to large adjustments when controlled to the national totals. This is a result of inconsistencies between sources. Our benchmark dataset is the published ERS data for output. Since the agriculture data is, to a large extent, derived, analysts with local agriculture data are encouraged to use it when building their IMPLAN models.

More details can be found in this article.


IMPLAN construction sectors are not NAICS-based, but rather are defined by the Census Bureau’s types of construction.  There are 10 new construction and 3 maintenance and repair construction sectors in IMPLAN’s current sectoring scheme.


Employment and Income

The BLS’ CEW data and the BEA’s REA data provide construction employment and income values at the national, state, and county level for a single aggregate construction sector.  These values are split amongst IMPLAN’s 13 construction sectors using various ratios from the latest BEA Benchmark I-O tables.  “In-house” construction activity performed by non-construction industries is “re-defined” into the appropriate construction sectors, following BEA Benchmark conventions. All estimates are then controlled to higher geographic levels.



National output values for the new construction sectors come from BEA NIPA tables 5.4.5-Private Fixed Investment in Structures by Type and 5.9.5. Gross Government Fixed Investment by Type. National output for the maintenance and repair construction sectors are based on the ratio of maintenance and repair to new construction from the BEA’s Annual Gross Domestic Product (GDP) by Industry series.  These values already contain “redefined” construction activity and thus do not require any redefinition procedures on our part. These national values are distributed to states and counties based on employment and income. 





IMPLAN data includes several types of government activity. “Administrative” or “general” government is considered an Institution in IMPLAN. There are payroll-only sectors that are purchased exclusively by administrative government spending patterns. In contrast, government enterprises are sectors that have profiles similar to those of industries. The Bureau of Economic Analysis (BEA) defines government enterprises as “Government agencies that cover a substantial portion of their operating costs by selling goods and services to the public and that maintain their own separate accounts.” The BEA Benchmark I-O tables treat government enterprises as industries and IMPLAN follows this convention. IMPLAN has 8 government enterprise sectors. All remaining government agencies are part of “general” or “administrative” government.


Most data sources other than the BEA Benchmark table do not distinguish government enterprises from administrative government establishments. The BLS’ CEW program, for example, reports employment and wages for government establishments by NAICS code and by ownership type (federal, state, or local). We assign government establishments to enterprise or administrative payroll sectors according to the BEA Benchmark table. If a government enterprise makes a particular commodity, then any government establishment that fits the NAICS code of that commodity is classified as a government enterprise. For example, in the latest (2007) BEA Benchmark table, “Electric power generation, transmission, and distribution,” which corresponds to NAICS 2211, is made by private industries in the electrical sectors but also by “State and local government electric utilities.” Accordingly, any government establishment owned by state or local government under NAICS 2211 is classified in IMPLAN sector 522 or 525, respectively. If the BEA Benchmark table does not show a government enterprise sector making a commodity, then any government establishment in that commodity’s corresponding NAICS code is classified as administrative government. Tables 1 through 3 show IMPLAN sector assignments for establishments by ownership type and NAICS code. Note that military employment is not covered by any of our NAICS-based sources and is estimated from military employment data reported by the BEA


Sector Description NAICS Code
518 Postal Service 491 Postal Service
519 Federal Electric Utilities 2211 Power Generation and Supply
520 Other Federal Government Enterprises

44-45 Retail Trade

481 Air Transportation

531 Real Estate

51213 Motion Picture and Video Exhibition

5241 Insurance Carriers

722 Food Services and Drinking Places

8121 Personal Care Services

535 *Employment and Payroll Federal Government Non-Military All other NAICS codes
536 *Employment and Payroll Federal Government Military Not reported on a NAICS basis


Sector Description NAICS Code
521 State Government Passenger Transport 485 Transit and Ground Passenger Transportation
522 State Government Electric Utilities 2211 Power Generation and Supply
523 Other State Government Enterprises

2212 Natural Gas Distribution

2213 Water, Sewage and Other Systems

7112 Spectator Sports

7132 Gambling Industries

72112 Casino Hotels

5241 Insurance Carriers

521 Monetary Authorities – Central Bank

5221 Depository Credit Intermediation

531 Real Estate

562 Waste Management and Remediation Services

4453 Beer, Wine and Liquor Stores

481 Air Transportation

482 Rail Transportation

483 Water Transportation

487 Scenic and Sightseeing Transporation

531 *Employment and Payroll State Government Non-Education All other NAICS codes
532 *Employment and Payroll State Government Education 61


Sector Description NAICS Code
524 Local Government Passenger Transport 485 Transit and Ground Passenger Transportation
525 Local Government Electric Utilities 2211 Power Generation and Supply
526 Other Local Government Enterprises

2212 Natural Gas Distribution

2213 Water, Sewage and Other Systems

4453 Beer, Wine and Liquor Stores

5241 Insurance Carriers

531 Real Estate

562 Waste Management and Remediation Services

7112 Spectator Sports

481 Air Transportation

482 Rail Transportation

483 Water Transportation

487 Scenic and Sightseeing Transportation

521 Monetary Authorities – Central Bank

5221 Depository Credit Intermediation

72112 Casino Hotels

7132 Gambling Industries

533 *Employment and Payroll Local Government Non-Education All other NAICS codes
534 *Employment and Payroll Local Government Education 61


Primary data sources for state-level agricultural output by sector are the NASS Value of Production and ERS Annual Cash Receipts data sets, both from the USDA. The county-level data are not consistent enough for our use. Data used to estimate county level farm sector output come from the Census of Agriculture. We use estimates from the census of Agriculture to estimate values not disclosed by NASS or ERS.