Estimating Non-Disclosures When Creating Employment Databases

There are three primary datasets containing non-disclosed elements that are used to estimate IMPLAN employment and labor income data: The BLS’ Quarterly Census of Employment & Wages (QCEW), the BEA’s Regional Economic Accounts (REA), and the Census Bureau’s County Business Patterns (CBP). Therefore, in order to complete the IMPLAN database, the non-disclosed values must be estimated and controlled for each regional dataset. This article describes these methods.

THREE STEPS TO ESTIMATING MISSING COUNTY BUSINESS PATTERN (CBP) ELEMENTS

While the CBP data lag one year behind the IMPLAN data year, they provide the number of establishments by employee-size classification for all NAICS codes in all places, even in those cases where the employment is not disclosed. For this reason, CBP estimates provide initial estimates of non-disclosed QCEW data in some cases.

Estimate the missing elements by taking the midpoints of each employment-size-class and multiplying these by the number of establishments in each size class. This provides an initial estimate of employment.
Add the 6-, 5-, 4- ,3- and 2-digit NAICS estimates from the bottom-up to make the first adjustment to the non-disclosed elements with the corresponding NAICS. This ensures that the 6-digit NAICS add to the 5-digit NAICS, the 5-digit NAICS add to the 4-digit NAICS, the 4-digit NAICS add to the 3-digit NAICS, and the 3-digit NAICS add to the 2-digit NAICS.
A top-down pass is made so that non-disclosed elements are adjusted again to ensure that all data add to the overall total. This procedure is performed on the national, state, and county data. This adjusting provides a complete set of CBP employment data that is internally consistent within a county, state, or nation.

ESTIMATING NON-DISCLOSURES IN THE QCEW DATA

QCEW data are current but also come with non-disclosures and, unlike the CBP data, provide only total establishment counts (i.e., they do not provide the number of establishments by employee-size classification) for non-disclosed items. To estimate the non-disclosed elements of the QCEW data, a number of methods are used, depending on data availability. The first option is to turn to recent past years’ disclosed raw QCEW values, applying a growth rate based on state or U.S. data. If no recent past raw data are disclosed for a particular NAICS code in a particular geography, the next option is to turn to the CBP data. If there are no CBP data available for this particular NAICS code and geography, we use ratios from the parent NAICS code or geography applied to the QCEW establishment counts, which are always disclosed.

Values for all lower-level NAICS codes are then controlled to higher-level parent NAICS codes, across all NAICS levels. Adjustments are made only to the elements that are non-disclosed; adjustments are not made to disclosed elements.

DERIVING REA STATE NON-DISCLOSURE ADJUSTMENTS

BEA Regional Economic Accounts (REA) data are integral to the IMPLAN data creation process because they cover all sectors and are one of the few sources of Proprietor Employment and Income data, as well as Employee Compensation. U.S. 3-digit REA employment and income data are reported without non-disclosures; however, the state and county level data do have non-disclosed elements. Estimation of non-disclosed state values are made while ensuring that the state values add up to the U.S. values and that the individual state sectors also sum to the more aggregated state sectors.

Disclosing Wage and Salary Employment (SA27 Tables)

State REA values are matched to corresponding CEW employment value. The CEW problem sectors (farm, railroad, and military) are not a problem with REA data as there are few non-disclosures in these sectors at the state level in REA.
After plugging in the initial estimates, state values are RASed using U.S. values as controls for the row values and the 1-digit State REA values as the column control.

Disclosing Wage and Salary Income

The first estimate is the corresponding state level CEW income/employment ratio times the state W&S employment derived above.
After plugging in the initial estimates, the state values are RASed using the U.S. as controls for the row values and the 2-digit State REA values as the column control.

Disclosing Total Employment (SA25 Tables)

The four component BEA Gross State Product data is a source of information at the state level which will tell us whether there is any self-employment income in the state.

If no proprietor employment is reported, then total employment is equal to wage and salary employment, the value of which is derived in the previous step ‘Disclosing Wage and Salary Employment’.
In some cases, a single 3-digit non-disclosure remains within a 2-digit group which can be derived through subtracting all disclosed 3-digit data from the 2-digit control value. Conversely, there may be proprietor employment and no corresponding wage and salary employment. For sectors for which both wage and salary employment and proprietor employment are non-disclosed, the first estimate is based on U.S. proprietor employment to wage and salary employment ratios for that sector.
Initial estimates are controlled to known totals at various stages in the process.

DERIVING REA COUNTY NON-DISCLOSURE ESTIMATES

Disclosing REA Employee Compensation and W&S Income

State and County 6-digit CEW income data are aggregated to the REA sectoring scheme (the estimation of CEW data is described above).
To get our first estimate of Employee Compensation (EC), we either project a historical disclosed REA value for that county and industry or apply the state’s ratio of REA EC to CEW W&S Income to the county’s CEW W&S Income.
To get our first estimate of W&S Income, we apply the state ratio of W&S Income to EC to the county’s EC value.
These estimates are then adjusted as necessary to ensure that more-detailed sectors sum to their more-aggregate parent sectors.

Disclosing REA Wage & Salary Employment

State and County 6-digit CEW employment data are aggregated to the REA sectoring scheme (the estimation of CEW data is described above).
To get our first estimate of W&S Employment we apply either the county’s ratio of CEW Employment to CEW W&S Income to the county’s W&S Income estimate or the state’s CEW W&S Employment to CEW W&S Income ratio to the county’s W&S Income estimate.
These estimates are then adjusted as necessary to ensure that more-detailed sectors sum to their more-aggregate parent sectors.

Disclosing REA Proprietor Employment

If Total Employment is disclosed, we simply subtract W&S Employment from Total Employment to get Proprietor Employment. Otherwise, if the parent sector’s Proprietor Employment is non-disclosed, we estimate the child sectors’ Proprietor Employment by applying the state’s ratio of Proprietor Employment to EC for that sector to the county’s EC value. If the parent sector’s Proprietor Employment is disclosed, we distribute its value to its children sectors based on state proportions of children to same parent.
These estimates are then adjusted as necessary to ensure that more-detailed sectors sum to their more-aggregate parent sectors.

Disclosing REA Proprietor Income

If Labor Income is disclosed, we simply subtract W&S Income from Labor Income to get Proprietor Income. Otherwise, if the parent sector’s Proprietor Income is non-disclosed, we estimate the child sectors’ Proprietor Income by applying the state’s ratio of Proprietor Income for that sector to Total EC for all sectors to the county’s Total EC value. If the parent sector’s Proprietor Income is disclosed and the child sector has the opposite sign as the parent sector, we apply the state childrens’ Proprietor Income-to-EC ratios to the county childrens’ EC values. If the parent sector’s Proprietor Income is disclosed and the child sector has the same sign as the parent sector, we subtract the disclosed childrens’ values from the parent’s value and then distribute the leftover to the non-disclosed child sectors based on state ratio of Proprietor Income for that child to state’s leftover parent to distribute (that is, the state parent less the same child sectors subtracted from the county parent).
These estimates are then adjusted as necessary to ensure that more-detailed sectors sum to their more-aggregate parent sectors.

DISTRIBUTING DISCLOSED 3-DIGIT REA EMPLOYMENT AND INCOME DATA TO IMPLAN SECTORING

With a complete disclosed set of 3-digit REA income (national income being adjusted to NIPA) and employment data, it is now possible to distribute data to the 536 IMPLAN sectors using the disclosed CEW data.

Distributing W&S Employment and Employee Compensation to IMPLAN Sectoring

The 3-digit adjusted REA W&S employment is distributed to IMPLAN sectors based on the CEW data that has been aggregated to the IMPLAN sectoring scheme.
State estimates are forced to sum to the national value.
County data are then forced to sum to the corresponding state values.
A proportion of some sectors’ activity (employment, output, income, etc.) gets reclassified into other sectors. This follows the BEA “redefinitions” practice and is designed to reassign products from producing industries in which they are secondary product to the industries where those products are primary.

Distributing Proprietor Employment and Income to IMPLAN Sectoring

Census data on employing proprietors per establishment and non-employing proprietors per establishment are used along with CEW establishment counts (aggregated to IMPLAN sectoring) to give a first estimate of proprietors.
These first estimates are used to distribute the REA control.
These final estimates are then multiplied by the REA control’s Proprietor Income per Proprietor to obtain Proprietor Income for each IMPLAN sector.

Special Considerations for Distribution

Agricultural and Construction sectors are not defined by 6-digit NAICS in IMPLAN. Agriculture is a commodity based sector and construction is based off of Census construction type categories. Therefore, other distributors are used (see Special Sectors) instead of CEW data.

CEW data also does not cover Railroad transportation, but there is a one-to-one correspondence between 3-digit REA data and IMPLAN sectoring, so no distribution is necessary.

There are industries, at the county level, for which 3-digit REA income and employment are available but there are no corresponding 6-digit CEW data (indicating self-employment in the industry for the county but no wage and salary workers). For these cases, the 3-digit REA data are distributed to industries based on the state distribution. If this distribution places less than 0.4 employees to a particular industry, then that piece of the distribution is added to the largest component of the distribution. Finally, a sector with less than 0.25 employees before redefinitions and balancing is zeroed out.