Printer Friendly

Evaluation of the state personal income estimates.

THIS article presents the results of a BEA study of the reliability of the quarterly and annual estimates of State personal income. The study, which covered estimates for 1980-87, assessed reliability using several statistical measures to examine the size of the revisions made to the estimates. The findings of this study are intended to help users determine the suitability for their purposes of the estimates released at different stages of the estimating process (see the box on page 21 for a description of some of the uses of the State estimates). The following summarizes the principal findings of the study:

* The major sources of the revisions

to the quarterly percent changes

in the preliminary quarterly estimates

of State personal income

are farm proprietors' income and

wages and salaries.

* Largely reflecting wages and salaries,

the preliminary quarterly estimates

of total personal income

tend to be underestimated in fastgrowing

States and overestimated

in slow-growing States.

* Beginning in 1984, the reliability

of the second quarterly estimates

was improved by the incorporation

of quarterly data from employers'

payroll tax reports.

* The annual revisions of total personal

income are smaller than the

quarterly revisions.

The first section of this article provides an overview of the sources and methods for the State personal income estimates. The second section presents the revision measures and the findings of the study. The third section describes some changes in the estimating procedures that have been made since the period covered by the study. Overview of Sources ana

Methods The quarterly and annual estimates of State personal income are subject to refinements as BEA incorporates source data that are more complete, more detailed, or otherwise more appropriate than the information previously available. These source data are incorporated at specific stages in the estimating process, and successive estimates are released according to a set schedule. Because the quarterly State estimates are tied to the annual estimates that incorporate more detailed and more reliable source data, the quarterly estimates for all components are interpolations or extrapolations of the annual estimates. The State quarterly source data are used as indicators for the interpolation or extrapolation. In addition, the quarterly State estimates are controlled to the personal income total in the national income and product accounts (NIPA's).

A detailed statement of the sources and methods used for the annual and quarterly State personal income estimates is presented in State Personal Income: 1929-87. (Ordering information appears on the inside back cove of this issue.) A revised description of the source data and methods used for the State wage and salary estimate was presented in State Estimates of Wages and Salaries: A Methodologic Update," in the October 1989 SURVEY OF CURRENT BUSINESS.

Revision schedule for State estimates

The preliminary' quarterly estimates of State personal income are released 4 months after the close of the quarter. The "second" quarterly estimates are released 3 months later. In October and again the following April, the quarterly estimates for the preceding 3 years are revised to reflect revisions to the annual estimates; the "final' quarterly estimates used in the study were the most recent estimates available in early 1989, when the study began. The "preliminary" annual estimates of State personal income, which are based on the current quarterly estimates, are released in April, 4 months after the close of the year. Revised annual estimates, which are developed independently of and are prepared in greater component detail than the quarterly estimates, are released 4 months later, in August. For several succeeding years, the annual estimates are again revised in April and in August, as additional data become available; the final' annual estimates used in this study were the most recent estimates available in early 1989. Sources of revision

Revisions to the quarterly State estimates stem from three main sources: (1) Incorporation of new or revised quarterly State indicators, (2) revision of the national controls from the NIPA's and of the annual State estimates, and (3) revision of the seasonal factors. Any of these factors can affect a particular quarter's revision. Revisions to the annual estimates stem from the incorporation of additional source data or from revisions of the national controls from the NIPA'S. Table 1 provides a brief summary of the primary sources for (1) the preliminary quarterly estimates, (2) the second quarterly estimates, and (3) the detailed annual State estimates. The primary sources for the major components of State personal income are discussed in this section.

Wages and salaries.-Most of the preliminary quarterly estimates of wages and salaries are extrapolations based on Bureau of Labor Statistics (BLS) 790 State employment data at the Standard Industrial Classification (SIC) division level. For durable and nondurable manufacturing, the extrapolation indicator is the product of employment data and production workers' hours and earnings. The BLS-790 is a monthly survey of employment, average weekly hours, and average hourly earnings that is conducted in cooperation with the State employment security agencies. Data are collected for the pay period that includes the 12th of each month from a sample of over 300,000 nonagricultural establishments. For the States, employment data, but not hours and earnings data, are available at the SIC division level; employment data and, for production workers, hours and earnings data are available for durable and nondurable manufacturing. The BLS-790 State data are available to BEA 6 weeks after the close of the month. Except for coverage differences, such as the exclusion of wages and salaries of U.S. citizens stationed abroad, the State quarterly estimates of wages and salaries are controlled to-that is, they sum to-the NIPA estimates of wages and salaries released by BEA 3 months after the close of the quarter. (For a detailed description of the differences between State and national estimates of personal income, see table 1 in State Estimates of Wages and Salaries: A Methodological Update" in the October 1989 SURVEY.) The NIPA estimates are based on more complete nation BLS-790 information. Data at the two digit SIC level on employment and, for production and nonproduction workers, on average weekly hours and average hourly earnings are included in the national wage and salary estimates. For the nonmanufacturing industries, employment is a reliable indicator for State wages and salaries when average wage rates and hours worked are stable or when they change similarly in all States. However, average wage rates and hours worked can change differently among States. In addition, the employment estimates available for the first quarterly State personal income estimates are subject to revision because they are based on a sample survey, and that sample may not accurately reflect employment changes associated with births of firms. In manufacturing, the data available for estimating preliminary quarterly State wages and salaries in manufacturing include wage-rate information in the form of average weekly hours and average. hourly earnings for production workers. However, these source data do not cover wages and salaries for nonproduction workers; they also exclude profit-sharing and other lump-sum payments for all workers. These gaps in coverage have become more significant as the number of production workers relative to nonproduction workers has declined and as wage payments based on profit sharing programs have become more common in all industries, including manufacturing.

In the second quarterly estimates of wages and salaries, quarterly data from the ES-202 employers payroll tax reports replace the BLS-790 data as the extrapolator. The second quarterly estimates are controlled to the same NIPA wage and salary estimates as the preliminary quarterly estimates. The ES-202 reports, covering 96 percent of all wages and salaries, summarize returns that are required of all employers covered by State unemployment insurance law and by the unemployment compensation program for Federal employees; thus, these reports provide a virtual census of nonagricultural employment and wages. The data are at the four-digit SIC level of industry detail, by State and county, and they are made available to BEA 5 months after the close of the quarter. Quarterly wage and salary estimates are adjusted to remove seasonal patterns by using, in most cases, the Census X-11 ARIMA seasonal adjustment method. Although the seasonal patterns are usually stable, they sometimes change rapidly and lead to substantial revisions in the seasonal factors when they are updated to reflect the latest year's data. Lump-sum payments, which are included in wages and salaries, have caused large revisions in the seasonal factors. Payments in certain industries will be discussed later in more detail. The preliminary annual State estimates of wages and salaries are derived from the quarterly State estimates released each April. At that time, the estimates for the first three quarters of the calendar year are based on ES-202 data, and the estimate for the fourth quarter is based on BLS-790 data. These annual estimates are controlled to the BLS-790-based annual NIPA estimates released 3 months after the close of the year.

BEA's final State annual estimates for most of wages and salaries are based on data from the ES-202 reports and are prepared at the SIC two-digit level of industry detail. These annual estimates are controlled to the revised NIPA estimates released in July, which incorporate ES-202 data. The final quarterly estimates incorporate these final State and national annual estimates.

Farm proprietors' income.-For the preliminary quarterly, second quarterly, and final quarterly estimates, monthly data on farm subsidy payments by State from the U.S. Department of Agriculture (USDA) are used as the quarterly indicator for the government payments portion of net farm proprietors' income.

Because no monthly or quarterly State data exist for the detailed components of production expenses and inventory change, monthly data on cash receipts from farm marketings are used as the quarterly indicator for net farm proprietors' income excluding government payments for all three stages of the quarterly estimates.

The preliminary annual estimates of total net farm income (gross farm income less production expenses) are based on the USDA State estimates adjusted for definitional differences. The BEA State annual estimates of net farm proprietors' income are obtained by deducting BEA State estimates of net corporate farm income from BEA State estimates of total net farm income.

The final annual estimates of farm proprietors' income are based on revised annual State income and expense data from. USDA.

Because no quarterly State data for farm production expenses or inventory changes exist, the subsequent incorporation of annual State data for these items can cause substantial revisions in the quarterly estimates of State farm proprietors' income. Although farm proprietors' income amounts to only about 1 percent of personal income for the Nation, it is very important in a number of States. In fact, as the analysis in the next section shows, many of the largest revisions in percent changes in total personal income are in these States.

Components based on wages and salaries.-In the preliminary quarterly, second quarterly, and final quarterly estimates, the State wage and salary estimates are used as the indicators for the residence adjustment and for those components that are closely related to wages and salaries: Other labor income, the construction portion of proprietors'income, and personal contributions for social insurance. For personal contributions, total wages and salaries are used as the quarterly indicator; for construction proprietors' income, construction wages and salaries are used; for the residence adjustment and for other labor income, wages and salaries for each SIC division are used.

The preliminary annual estimates are derived from the quarterly estimates released each April. The final annual estimates are based on annual source data from a variety of agencies.

Because the quarterly estimates of these components are based on wages and salaries, revisions in these components reflect revisions to the quarterly wage and salary estimates as wen as the incorporation of annual source data.

Components based on annual trends. -No quarterly State indicators exist for the following components of personal income: Farm wages; pay of military reserves; nonfarm proprietors' income excluding construction proprietors' income; dividends, interest, and rent; and transfer payments excluding unemployment insurance benefits. For these components, which amount to about 37 percent of personal income for the Nation, quarterly State indicators are based on trend changes in State shares of the national total; the trend line is estimated using the latest 5 years of annual State and national estimates.

The preliminary annual estimates for these components are derived from the quarterly estimates released each April. The final annual estimates are based on annual source data from a variety of agencies. .

For the components that are based on annual trends, revisions result from the incorporation of the annual NIPA and State estimates. Most of these components are driven by national, rather than local, economic conditions. The quarterly State estimating procedure distributes these revisions smoothly across the quarters.

Dividends, interest, and rent amount to nearly 17 percent of personal income for the Nation; over four-fifths of this component is interest payments. Because the largest capital markets are national, fluctuations in rates of return generally are determined more by national than by local economic conditions. Interest rates, for example, tend to fluctuate throughout the country in response to changes in Federal monetary policy. Moreover, residents of a State need not invest their savings locally. Thus, individual State shares of national dividends, interest, and rent are unlikely to change sharply from quarter to quarter in response to local economic conditions.(1)

Transfer payments excluding unemployment insurance benefits amount to about 14 percent of personal income for the Nation. Over two-thirds of these transfers are social security benefits and other Federal retirement-related transfers for which State shares do not vary much from quarter to quarter. Public assistance payments (for example, supplemental security income, aid to families with dependent children, and food stamps) are more sensitive to local economic conditions, so quarterly extrapolations of these payments are subject to greater errors than are the extrapolations of retirement related transfer payments. However, public assistance payments amount to less than one-fifth of total transfer payments and less than 3 percent of personal income for the Nation.

The personal income components for which trend extrapolation is likely to be least satisfactory are farm wages and nonfarm proprietors' income. Farm wages amount to about 0.2 percent of personal income for the Nation, and nonfarm proprietors' income amounts to about 7 percent. Over one half of nonfarm proprietors' income consists of professional and other services, which are likely to have reasonably stable trends in the State shares of nonfarm proprietors' income for the Nation; however, proprietors are also important in a number of industries such as mining, forestry and fisheries, and real estate--that can be quite volatile and that vary substantially from State to State.

Revision Measures and

Findings of the Study

The reliability of the quarterly State personal income estimates is determined by the accuracy of the final estimates and the size of the revisions to the preliminary and second estimates. Errors in the final estimates arise mostly from causes that do not easily lend themselves to error quantification-for example, data gaps and nonsampling errors in source data. Consequently, this study considered reliability as reflected in measures of revision.

The study examined the size of the revisions in three State aggregates-personal income, nonfarm personal income, and wages and salaries between the preliminary, second, and final estimates of quarterly change and the preliminary and final estimates of annual change.(2)

Revision measures

Revisions can be measured in several ways. This study used the same measures that BEA has used to study revisions to the estimates of GNP and its components. 3 One measure analyzes the range of revisions, where revision is defined as the percent change in the final estimates minus the percent change in the preliminary estimates. Typically, ranges covering two-thirds and nine-tenths of the revisions are presented.

Other sets of measures-dispersion, relative dispersion, bias, and relative bias-provide summary measures that are defined as follows: Let P represent the percent change in the preliminary estimates, F the percent change in the final estimates, and n the number of observations.

Dispersion is the average of the absolute values of revision in percent changes between the preliminary and the final estimates: FORMULA OMITTED

Bias is the average of the revisions in percent changes between preliminary and final estimates: FORMULA OMITTED

Relative dispersion expresses th dispersion as a percent of the aver age of the absolute values of the final percent changes: FORMULA OMITTED

Relative bias expresses the bias a a percent of the average of the final percent changes: FORMULA OMITTED

Table 2 presents measures of the range, dispersion, bias, relative dispersion, and relative bias of the revisions between the preliminary and the final estimates of quarterly total personal income, nonfarm personal income, and wage and salary disbursements for 1980-87. It also presents the results of statistical tests of the significance of the biases for total personal income, nonfarm personal income, and wage and salary disbursements. Preliminary quarterly estimates

Range.-The range of the revisions between the preliminary and final estimates of total personal income is largest for the States of the Plains region. Of the nine States in which the differences between the lower and upper limits for nine-tenths of the revisions exceed 3.5 percentage points, five are in the Plains region chart 1). A comparison of chart 1 with chart 2, which shows the range of revisions for nonfarm personal income, reveals that farm income is the source of much of the variation in the range of revisions among States. After removing farm income, the range of revision for all of the Plains States is below 3.5 percentage points.

For nonfarm personal income, the range of revisions exceeds 3.5 percentage points for only four StatesDelaware, Montana, Wyoming, and Alaska. In terms of nonfarm personal income, these four States are small -Delaware, the largest' of the four, ranked 45th in the Nation in 1987.4 In addition, Alaska has a large special transfer payment, the Alaska Permanent Fund dividend payment, that began in 1982 after the discovery and development of oil on the North Slope. The introduction in 1985 of seasonal adjustment for this payment, which amounts to 4 percent of total personal income in Alaska, led to large revisions to the estimates of total personal income and nonfarm personal income for 1982-84. The revisions to the estimates of total personal income and nonfarm personal income in Alaska for 1985-87 are significantly smaller; however, the revisions are still large relative to other States.

Delaware, Montana, Wyoming, and Alaska are also among the seven States for which the range of revisions to the estimates of wages and salaries exceeds 3.5 percentage points (chart 3). The economies of these four States are strongly influenced either by a few large companies or by volatile industries, such as mining. In addition, the income estimates for Alaska, Montana, and Wyoming have strong seasonal patterns, the adjustments for which were revised substantially during the period covered by this study.

New Hampshire, Vermont, and Michigan are the other three States with large revisions to the estimates of wages and salaries. New Hampshire and Vermont are 'small'-ranking 37th and 50th in nonfarm personal income, respectively. Although Michigan ranks 9th in nonfarm personal income, its economy is heavily influenced by one industry-motor vehicle manufacturing.

Dispersion.-The measures of dispersion of total personal income also show the effect of the revisions to the estimates of farm income. The dispersion in the revisions for six States exceeds 2 percentage points; removing farm income reduces the number of States to one-Alaska. The relative dispersion of total personal income is more than 50 percent for 12 States (that is, for these States, the average size of the revisions exceeds 50 percent of the average quarterly change in total personal income) and is less than 30 percent for 14 States.

The relative dispersion in the revisions to the estimates of total personal income tends to be higher for small' States and lower for large" States. All of the 12 States with a relative dispersion of more than 50 percent ranked in the bottom half of the Nation in terms of total personal income. Of the 14 States with a relative dispersion of less than 30 percent, 10 ranked in the top half of the Nation in terms of total personal income.

The revisions to the estimates of wages and salaries show a greater relative dispersion than do the revisions to the estimates of nonfarm personal income. For nonfarm personal income, the relative dispersion is more than 50 percent for only 6 States and is less than 30 percent for 26 States. For wages and salaries, the relative dispersion is more than 50 percent for 18 States and is less than 30 percent for only 6 States.

Bias.-The study found statistically significant bias at the 5-percent confidence level in the preliminary quarterly estimates of total personal income for 10 States. For all 10 States, the bias was negative, which indicates that the preliminary estimates were understated. This result extends through the estimates of nonfarm personal income and of wages and salaries; except for Wyoming, each State that has statistically significant bias in its estimates of total or of nonfarm personal income also has statistically significant bias in its estimates of wages and salaries.

The variation in the bias in wages and salaries is highly associated with a variation in the growth of wages and salaries. Of the 13 States with significant bias in wages and salaries, 12 States are among the top 14 States in terms of growth during 1980-87. Further, of the 21 States with the most negative bias in wages and salaries, 17 States are among the top 21 States in terms of growth. Of the 22 States with zero or positive bias in wages and salaries, 17 States are among the bottom 22 States in terms of growth.

The direction and size of the bias in the estimates of wages and salaries vary among States and industries. The BEA coastal regions (the New England, Mideast, Southeast, and Far West regions), where fast-growing trade and service industries are concentrated, had the largest understatements in the preliminary estimates of the quarterly change in wages and salaries. Biases for all 4 coastal regions and for 18 of their 28 States were as negative as, or more negative than, -0.2 percentage point, which was the bias for the Nation. All four coastal regions had above-average wage and salary growth for 1980-87.

The BEA interior regions (the Great Lakes, Plains, Rocky Mountain, and Southwest regions), where slow-growing goods-producing industries are concentrated, had smaller understatements or had overstatements in the preliminary estimates of the quarterly change in wages and salaries. Biases for aH 4 interior regions and for 19 of their 21 States were positive or were less negative than the bias for the Nation. All four interior regions had below-average wage and salary growth for 1980-87.

Second quarterly estimates During the 1980's, the quality and timeliness of the quarterly ES-202 data, which is the primary data source for wages and salaries in the final quarterly estimates of State personal income, improved significantly. Therefore, beginning with the estimates for the third quarter of 1984, the ES-202 wage 202 and salary data were incorporated into the second estimates 7 months after the end of the quarter. Before then, the ES-202 data had been incorporated into the quarterly estimates twice a year with a lag of 1 0 months. Table 3 presents measures of the dispersion, bias, relative dispersion, and relative bias in the revisions to the preliminary-to-final and second-to-final estimates of quarterly total personal income, nonfarm personal income, and wages and salaries for 198487, the period in which the ES-202 data were incorporated into the second estimates. It also presents the results of statistical tests of the significance of the biases for total personal income, nonfarm personal income, and wages and salaries.

Dispersion.-Most States show a reduction in the size of the revisions to the estimates of wages and salaries that resulted from the incorporation of the quarterly ES-202 data into the second estimate. From the preliminary estimate to the final estimate, the dispersion in the revisions for 13 States is greater than or equal to 1 percentage point; from the second estimate to the final estimate, the dispersion for all States except Michigan is less than 1 percentage point (for Michigan, it is 1 percentage point)(chart 4).

Although the incorporation of wages and salaries is the most important difference between the preliminary and second estimates, farm income is also subject to some revision between these estimates. The effect of these revisions is apparent in table 3, where for some farm-oriented States the differences between the preliminary-to-final and second-to-final total personal income dispersions are greater than can be attributed to differences in wages and salaries alone.

Bias.-The bias in the second quarterly estimates of total personal income is similar to that found in the preliminary quarterly estimates. Statistically significant negative bias at the 5percent confidence level, present in the preliminary estimates for 23 States, is present in the second estimates for 19 States. There is some evidence that the bias in the second estimates is not as extreme as in the preliminary estimates. In the preliminary-to-final revisions, the bias in wages and salaries for 13 States is equal to or less than -0.6 percentage point, and the bias for 7 States is equal to or greater than 0.2 percentage point. In the second-to-final revisions, the biases for all States except Alaska fall between -0.6 and 0.2 percentage point.

An important factor limiting the improvement in the reliability of the estimates achieved by the introduction of the ES-202 data into the State wage and salary estimates is that the second estimate of wages and salaries is controlled to NIPA estimates that have not yet incorporated the ES-202 data.

The annual ES-202 data are not incorporated into the annual and quarterly NIPA estimates until July. Therefore, the bias in the quarterly NIPA estimates of wages and salaries continues to affect the second State estimates even after the ES-202 data are incorporated into the State estimates. For this reason, the biases in the second-to-final revisions are negative for most of the regions and States. In none o the 19 States in which the bias is statistically different from zero is it also statistically different from the bias for the Nation.

Preliminary annual estimates

Many applications of the State personal income estimates are based on the annual estimates. The initial annual estimates that can be used for such purposes-whether for a calendar year or a fiscal year-are derived as the sum of the quarterly estimates. Table 4 shows the measures of revisions between the preliminary annual estimates and the final annual estimates that were available when the study began.

For 1980-87, the relative revisions between the preliminary annual estimates and the final annual estimates are smaller than the relative revisions between the preliminary quarterly estimates and the final quarterly estimates shown in table 2. Only 3 States, all in the Plains region, have a relative dispersion in total personal income of 30 percent or more; 16 States have a relative dispersion of less than 10 percent. The revisions in the annual estimates are smaller for three main reasons: (1) State ES-202 wage and salary data for three quarters are incorporated into the preliminary annual estimates for most of the period, (2) State annual data on farm proprietors'income are incorporated into the preliminary annual estimates, and (3) annual estimates are not affected by seasonal adjustments.

Recent Developments

Beginning with the preliminary annual estimates for 1989 published in the April 1990 SURVEY, BEA stopped controlling the annual estimates for the most current year to the BLS based 790-U.S. totals for wages and salaries in the NIPA'S. The U.S. totals for wages and salaries now used for the preliminary annual State estimates are based on ES-202 data for three quarters. This methodological change is expected to reduce the revisions between the preliminary and final annual estimates of total personal income.

In addition, new procedures to deal with changes in seasonal payment patterns were used in the April 1990 estimates of wages and salaries for the service industry and in the October 1990 estimates for the finance, insurance, and real estate industries. In recent years, lump-sum payments in these industries have been unusual both in size and in timing; these payments, by significantly affecting the quarterly seasonal pattern of the wage and salary estimates, have made the existing seasonal factors less reliable. In the service industry, the size and the timing of lump-sum payments paid by personal service corporations to their owner-employees has changed in recent years, reflecting provisions of the Tax Reform Act of 1986 and the Revenue Act of 1987. In the finance, insurance, and real estate industries, the amount of lump-sum payments to security and commodity brokers has also changed, for a variety of reasons: The slowdown in brokerage business after the stock market crash of October 1987, the reduction in the number of mergers and acquisitions of companies, and the reduction in demand for new issues of high-yield bonds. As a result of these developments, the revisions to the 1988 and 1989 preliminary and second estimates were larger than normal. If the size and timing of lump-sum payments stabilize, the size of revisions for years after 1989 can be expected to be similar to those for 1984-87.
COPYRIGHT 1990 U.S. Government Printing Office
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1990 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Brown, Robert L.; Stehle, James P.
Publication:Survey of Current Business
Date:Dec 1, 1990
Words:4887
Previous Article:National income and product accounts: selected NIPA tables, NIPA charts and reconciliation and other special tables.
Next Article:U.S. international transactions, third quarter 1990.
Topics:


Related Articles
State personal income, first quarter, 1987.
State personal income, 1986-88: revised estimates.
Relationship between personal income and adjusted gross income, 1987-88.
State personal income, 1987-89: revised estimates.
Regional perspectives; state personal income: methodological improvements.
County and metropolitan area personal income, 1986-88.
Personal income by state and region, first quarter 1992.
State and regional tables.
Personal income and per capita personal income by state and region.
State Personal Income, Third Quarter 2000.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters