| Federal
Committee on Statistical
Methodology Office of Management and Budget |
FCSM
Home ^ Methodology Reports ^ |
Statistical Policy Working Paper 21 - Indirect Estimators in Federal Programs
Click HERE for graphic. Statistical Policy Working Paper 21 Indirect Estimators in Federal Programs Prepared by Subcommittee on Small Area Estimation Federal Committee on Statistical Methodology Statistical Policy Office Office of Information and Regulatory Affairs Office of Management and Budget July 1993 MEMBERS OF THE FEDERAL COMMITTEE ON STATISTICAL METHODOLOGY (July 1993) Maria E. Gonzalez, Chair office of Management and Budget Yvonne M. Bishop Daniel Melnick Energy Information Substance Abuse and Mental Administration Health Services Administration, Warren L. Buckler Robert P. Parker Social Security Administration Bureau of Economic Analysis Cynthia Z.F. Clark Charles P. Pautler, Jr. National Agricultural Bureau of the Census Statistics Service David A. Pierce Steven Cohen Federal Reserve Board Administration for Health Policy and Research Thomas J. Plewes Bureau of Labor Statistics Zahava D. Doering Smithsonian institution Wesley L. Schaible Bureau of Labor Statistics Roger A. Herriot National Center for Fritz J. Scheuren Education Statistics internal Revenue Service C. Terry Ireland Monroe G. Sirken National Computer Security National Center for Center Health Statistics Charles D. Jones Robert D. Tortora Bureau of the Census Bureau of the Census Daniel Kasprzyk Alan R. Tupek National Center for National Science Foundation Education Statistics Nancy Kirkendall Energy Information Administration PREFACE The Federal Committee on Statistical Methodology was organized by OMB in 1975 to investigate issues of data quality affecting Federal statistics. Members of the committee, selected by OMB on the basis of their individual expertise and interest in statistical methods, serve in a personal capacity rather than as agency representatives. The committee conducts its work through subcommittees that are organized to study particular issues. The subcommittees are open by invitation to Federal employees who wish to participate. Working papers are prepared by the subcommittee members and reflect only their individual and collective ideas. The Subcommittee on Small Area Estimation was formed in 1991 to document the uses of indirect estimators by Federal statistical agencies to prepare and publish estimates. An indirect estimator uses values of the variable of interest from a domain and/or time period other than the domain and time period of the estimate being produced. Users of indirect estimators should consider the errors to which these estimates are subject. Eight programs that publish indirect estimators are described in this report. These programs sometimes respond to legislative requirements or, alternatively, to State data needs. The programs and sponsor agencies are: infant and maternal health characteristics (National Center for Health Statistics (NCHS)); personal income, annua1 income, and gross product (Bureau of Economic Analysis); postcensal population estimates for counties (Bureau of the Census (BOC)); employment and unemployment for States (Bureau of Labor Statistics); cotton, rice, and soybean acreage (National Agricultural Statistics Service (NASS)); livestock inventories, crop production, and acreage (NASS); disabilities, hospital utilization, physician and dental visits (NCHS); and median income for 4-person families (BOC). The Subcommittee on Small Area Estimation was chaired by Wesley L. Schaible of the Bureau of Labor Statistics, Department of Labor. MEMBERS OF THE SUBCOMMITTEE ON SMALL AREA ESTIMATION Wesley L. Schaible (Chair) Bureau of Labor Statistics Department of Labor Robert E. Fay Bureau of the Census Department of Commerce Joe Fred Gonzalez National Center for Health Statistics Department of Health and Human Services Linnea Hazen Bureau of Economic Analysis Department of Commerce William C. Iwig National Agricultural Statistics Service Department of Agriculture John F. Long Bureau of the Census Department of Commerce Donald J. Malec National Center for Health Statistics Department of Health and Human Services Alan R. Tupek National Science Foundation ACKNOWLEDGMENTS This report is the result of the collaborative efforts of members of the Subcommittee on Small Area Estimation of the Federal Committee on Statistical Methodology. Subcommittee members volunteered a considerable amount of time over a two year period to complete individual chapters in the report. Chapter authors are identified at the beginning of each chapter. Although the introductory and concluding chapters were authored by the Subcommittee Chair, they resulted from discussions which included the entire Subcommittee as well as other interested parties. Throughout the preparation of the report, a number of reviewers read drafts and provided valuable comments. The Subcommittee thanks Maria Gonzalez, Chair of the Federal Committee on Statistical Methodology, for her support and contributions throughout the development and preparation of the report. The Subcommittee also expresses its appreciation to the members of the Federal Committee on Statistical Methodology for reviewing the report and providing many useful suggestions. The Subcommittee extends special thanks to the following Committee members: Yvonne Bishop, Cynthia Clark, Robert Parker, David Pierce, Thomas Plewes, Monroe Sirken, Robert Tortora, and, in particular, Fritz Scheuren. The Subcommittee also thanks Alan Dorfman, Steve Mlier, and especially Robert Casady, all of the Bureau of Labor Statistics, for helpful discussions and comments. In addition, the Subcommittee extends its thanks to Gordon Brackstone and the staff of Statistics Canada, to Wayne Fuller of Iowa State University, and especially to Graham Kalton of Westat, Inc. for valuable comments and the time so generously provided to review the report. TABLE OF CONTENTS Chapter 1. Introduction and Summary........................ 1-1 Chapter 2. Synthetic Estimation in Follwback Surveys at the National Center for Health Statistics.............. 2-1 Chapter 3. State, Metropolitan Area, and County Income Estimation......................................... 3-1 Chapter 4. Postcensal Population Estimates: States, Counties and Places....................................... 4-1 Chapter 5. Bureau of Labor Statistics' State and Local Area Estimates of Employment and Unemployment.................. 5-1 Chapter 6. County Estimation of Crop Acreage Using Satellite Data...................................... 6-1 Chapter 7. The National Agricultural Statistics Service County Estimates Program......................................... 7-1 Chapter 8. Model-Based State Estimates from the National Health Interview Survey................................... 8-1 Chapter 9. Estimation of Median Income for Four-Person Families by State......................................... 9-1 Chapter 10. Recommendations and Cautions.................. 10-1 iv CHAPTER 1 Introduction and Summary 1.1 Introduction Federal statistical agencies produce estimates of a variety of population quantities for both the nation as a whole and for subnational domains. Domains are commonly defined by demographic and socioeconomic variables. However, geographic location is perhaps the single variable used most frequently to define domains. Regions, states, counties, and metropolitan areas are common geographic domains for which estimates are required. Federal agencies use different data systems and estimation methods to produce domain estimates. Those systems designed for the purpose of producing published estimates use standard, direct estimation methods. Data systems are designed within time, cost and other constraints which restrict the number of estimates that can be produced by standard methods. However, the demand for additional information and the lack of resources to design the required dam systems have led federal statistical agencies to consider non-standard methods. Estimation methods of a particular type, referred to as small area or indirect estimators, have sometimes been used in these situations. The purpose of this report is to document, in a manner that will facilitate comparisons, the practices and estimation methods of the federal statistical programs that use indirect estimators. Only programs that use indirect estimators for the production of published estimates are included; whether a data system is based on a census (including administrative records) or a: sample survey has no bearing on the inclusion of a program. The focus of this report is on the method by which estimates are produced. The methods and practices of eight programs are documented here; three are located in the Department of Commerce, two in the Department of Agriculture, two in the Department of Health and Human Services, and one in the Department of Labor. Other applications of indirect estimators occur in federal statistical agencies but descriptions of these applications have not been included in this report. Most of these methods were not included because they were not used, to produce published estimates. This publication restriction, a somewhat arbitrary indicator of program importance, keeps the number of programs included to a manageable level but leads to the omission of other interesting methods (for example, Fay and Herriott 1979). This introductory chapter includes brief discussions of small area estimation terminology; definitions of direct and indirect estimators; some characteristics, of indirect estimators; and summary descriptions of the programs included in the report. Each program is documented, following a standard format, in the individual chapters of the report. The intent is to create program descriptions that will not only provide complete, self-contained documentation for each individual program but also facilitate comparisons among programs. Although the focus of the report is on estimation methods, the description of each program includes material on program history, policies, evaluation practices, estimation methods, and current, problems and activities. In addition to the standard chapter format, attempts have been made to employ common notation throughout the chapters to facilitate comparisons of estimation methods. The report, concludes with a number of recommendations and cautions. 1.2 Terminology Terms used to describe indirect estimators can be confusing. Increased interest in non- traditional estimators for domain statistics has occurred recently among survey statisticians and, even though the term "small area estimator" is commonly used, uniform terminology has not yet evolved. This term is frequently used because in most applications of these estimators the domains of interest have been geographic areas. However, the word "small" is misleading. It is the small number of sample observations and the resulting large variance of standard direct estimators that is of concern, rather than the size of the population in the area or the size of the area itself The word "area" is also misleading since these methods may be applied to any arbitrary domain, not just those defined by geographic boundaries. Other terms used to describe these estimators include "local area" (Ericksen 1974), "small domain" (Purcell and Kish 1979), subdomain" (Laake 1979), "small subgroups" (Holt, Smith, and Tomberlin 1979), subprovincial" (Brackstone 1987), "indirect" (Dalenius 1987), and "model-dependent" (Sarndal 1984). The term "synthetic estimator" has also been used to describe this class of estimators (NIDA 1979) and, in addition, to describe a specific indirect estimator (NCHS 1968). Survey practitioners sometimes refer to indirect estimators as "model-based" whereas this term is rarely, if ever, used to describe direct estimators. However, direct estimators can be motivated by and justified under models as readily as indirect estimators. There is also lack of agreement on what to call the class of direct estimators. In addition to "direct" (Royall 1973), authors have used "unbiased" (Gonzalez 1973), "standard" (Holt, Smith, and Tomberlin 1979), "valid" (Gonzalez 1979), and "sample-based" (Kalton 1987). In the remainder of this paper, the words "direct" and "indirect" will be used to describe traditional and small area estimators, respectively. 1.3 Direct and Indirect Estimators Perhaps the most common measure of error of an estimator is the mean square error, composed of the sum of the variance of the estimator and the squared bias of the estimator. Biases can rarely be estimated with any degree of confidence. If an estimator is unbiased or approximately unbiased, the variance of the estimator, which can be estimated from the available data, is a satisfactory measure of error of the estimator. This leads to the selection of estimators that are unbiased or approximately unbiased in most applications. Such estimators allow data systems to be designed so that estimates with a predictable level of error can be produced with high probability and, in addition, estimated measures of error can be provided to accompany estimates. Federal statistical programs are generally designed using direct estimators which are unbiased, or approximately unbiased, under finite population sampling theory. Samples are often used and, given adequate resources, the sample design specifies population and domain sample sizes large enough to produce direct estimates that meet reliability requirements for the survey. When a domain sample size is too small to make a reliable domain estimate using the direct estimator, a decision must be made whether to produce estimates using an alternative procedure. The alternative estimators considered are those that increase the effective sample size and decrease the variance by using data from other domains and/or time periods through models that assume similarities across domains and/or time periods. These estimators are generally biased, but if the mean square error of the alternative estimator can be demonstrated to be small compared to the variance of the direct estimator, the selection of the alternative estimator may be justified. In extreme situations, there may be no sample units in the domain of interest and, if an estimate is to be produced, an alternative estimator will be required. Indirect estimators have been characterized in the Bayesian and empirical Bayes literature as estimators that "borrow strength" by the use of values of the variable of interest from domains other than the domain of interest. This approach can be used to provide a working definition of direct and indirect estimators for a broad class of population quantities including means and totals. A direct estimator uses values of the variable of interest only from the time period of interest and only from units in the domain of interest. An indirect estimator uses values of the variable of interest from a domain and/or time period other than the domain and time period of interest. Three types of indirect estimators can be identified. A domain indirect estimator uses values of the variable of interest from another domain but not from another time period. A time indirect estimator uses values of the variable of interest from another time period but not from another domain. An estimator that is both domain and time indirect uses values of the variable of interest from another domain and another time period. Indirect estimators depend on values of the variable of interest from domains and/or time periods other than that of interest. These values are brought into the estimation process through a model that, except in the most trivial case, depends on one or more auxiliary variables that are known for the domain and time period of interest. To the extent that applicable models can be identified and the required auxiliary variables are available, indirect estimators can be created to produce estimates. Perhaps the simplest example of an indirect estimator is the use of the sample mean of the entire sample as the estimator for a specific domain. For example, the use of the mean from a national sample as an estimate for a particular state. To the extent that information related to the variable of interest is available for the state, an indirect estimator which is "better" than the national mean can be defined. The availability of auxiliary variables and an appropriate model relating the auxiliary variables to the variable of interest are crucial to the formation of indirect estimators. However, the definition of direct and indirect estimators does not depend on whether or not auxiliary variables from outside the domain or time period of interest are used. The clear distinction between direct and indirect estimators made in the discussion above reflects the situation during the design stage of a data system. However, when estimators reflect the realities associated with data system implementation, the distinction becomes a little less clear. For example, nonresponse is a common problem in dam collection efforts. To the extent that nonresponse occurs, even direct estimators must rely on model-based assumptions relating the known information for responders to the unknown information for nonresponders. 1.4 Organization of Program Chapters As discussed in the previous section, indirect estimators borrow strength and can be classified into three types: domain indirect, time indirect, and domain/time indirect. In addition to this classification, indirect estimators are commonly expressed in different forms, that is different algebraic expressions. Each of the eight programs described in this report uses one of the following three common indirect estimators: synthetic, regression, or composite. The order of chapters describing programs follows this classification of estimators. That is, the program that uses a synthetic estimator is presented first in Chapter 2, followed by the programs that use regression estimators in Chapters 3 through 6; those programs that use composite estimators are presented in Chapters 7 through 9. Some of the programs have used different estimators at different times; however, emphasis is placed on the estimator that was last used to publish estimates. As with all indirect estimators, synthetic estimators may be domain indirect, time indirect, or domain and time indirect. For example, a domain indirect synthetic estimator for a population total in domain d and time t may be written as Click HERE for graphic. Regression estimators may be direct or, like the synthetic estimator, domain indirect, time indirect, or domain and time indirect depending on how the parameters are estimated. For example, a domain indirect regression estimator for a population total may be written as Click HERE for graphic. Click HERE for graphic. It should be noted that not all indirect estimators are linear. For examples of nonlinear indirect estimators see MacGibbon and Tomberlin (1989) and Malec, Sedransk, and Tompkins (1993). This latter, nonlinear indirect estimator is being considered for use in conjunction with the National Health Interview Survey and is discussed in Chapter 8. 1.5 Characteristics of Indirect Estimators There are several fairly well-known characteristics of indirect estimators that are important for producers and users to keep in mind. o In general, indirect estimators have relatively small variances since they not only incorporate observations from the domain and time period of interest, but also, from other domains and/or time periods. The variance of a modified synthetic estimator is discussed by Holt, Smith, and Tomberlin (1979) and variances of several indirect estimators resulting from different prediction models are discussed by Royall (1979). Care must be taken since the variance alone may be a misleading measure of error. See, for example, Raback and Sarndal (1982) and Sarmdal and Hidiroglou (1989). o An indirect estimator will be biased if the model assumptions leading to the estimator are not satisfied. Even so, an indirect estimator may be a useful alternative to a direct estimator when the mean squared error of the indirect estimator is sufficiently small compared to the variance of the direct estimator. However, the magnitude of the bias is likely to vary with each application and estimation of biases is difficult. Gonzalez and Waksberg (1973) consider the problem of estimating the mean squared error of synthetic estimators, and Prasad and Rao (1990) discuss the estimation of the mean squared error of indirect estimators. Care must be taken when interpreting estimated mean squared errors of indirect estimators. Some approaches provide an average measure over all domains rather than a measure associated with a specific domain. Confidence intervals for biased estimators is a related issue that has, been addressed by Miller (1992). o For a given application and estimator, biases in different domains will differ since the model will likely be a better representation of reality in some domains than in others. In general, when an indirect estimator is used to produce estimates for a number of domains, the distribution of estimates will have a smaller variance than the corresponding distribution of domain population values. This is a result of the tendency for indirect estimators to have relatively small biases when domain population values are close to the average total population value and, when domain population values are not close to the overall population value, to have relatively large directional biases which make the estimates closer to the overall population value. There is considerable evidence illustrating this characteristic (Gonzalez and Hoza 1978; Schaible et al. 1977 and 1979; and Heeringa 1981). Not all indirect estimators display this characteristic to the same extent. Spjovoll and Thomsen (1987), Lahiri (1990), and Ghosh (1992) have addressed this problem and suggest constrained approaches. o From a model-based, prediction point of view, direct and indirect estimators are model unbiased under the model that generates the estimator. A direct estimator is robust against model failure in the sense that it is unbiased, not only under the domain/time specific model which generates the estimator, but under each of the models associated with the corresponding indirect estimators. Indirect estimators are not robust in the same sense. However, the domain indirect estimator and the time indirect estimator are both more robust against model failure, in a similar sense, than the estimator that is both domain and time indirect. The bias of indirect estimators, under the domain and time specific model, is a source of concern that results in a reluctance to fully accept indirect estimators in many applications. An example and additional discussion of this aspect of indirect estimator bias is given in Schaible (1993). 1.6 Program Summaries The programs described in this report were initiated in response to a variety of needs and directives. Several are a direct result of legislative requirements to allocate federal funds. Others were created in response to state needs for data and/or to standardize estimation methods across states. Others are viewed as research programs that periodically publish estimates when an improved methodology has been developed. Table 1 below allows a comparison of summary information on the programs described in Chapters 2 through 9 in this report. The eight programs that use indirect estimators to publish estimates are located in five large statistical agencies. In some instances, a program produces estimates for a single variable; in other instances, estimates are produced for numerous variables. States and counties are the only domains for which indirect estimates are presently published. Four of the programs publish estimates for states, three for counties, and one for both states and counties. There is considerable variability in the frequency with which estimates are published. Two programs publish estimates only periodically, every few years. The remainder publish indirect estimates on a fixed schedule: four annually, one annually with selected estimates on a quarterly schedule, and one monthly. As noted above, a variety of indirect estimators are used to produce estimates. Synthetic, regression, and composite estimators that borrow strength over domains, over time, or over both domain and time are found among these programs. The estimation procedures for six of the programs are based on data from sample surveys. There is no sampling involved in the procedures used in the two programs that produce estimates of personal income and postcensal populations. Given the differing demands on Federal statistical agencies, it is not surprising that considerable variation is seen in the programs described in this report. Further investigations and improvements in the quality of indirect estimates published by Federal agencies are needed. It is hoped that recognition of the differences, as well as the similarities, in these programs will help provide a foundation for this further effort. Click HERE for graphic. REFERENCES Brackstone, G. J.. (1987), "Small Area Data: Policy Issues and Technical Challenges," in Small Area Statistics, New York: John Wiley and Sons. Dalenius, T. (1987), "Panel Discussion" in Small Area Statistics, New York: John Wiley and Sons. Ericksen, E.P. (1974), "A Regression Method for Estimation Population Changes for Areas," Journal of the American Statistical Association, 69, 867-875. Fay, R.E. and Herriott, R.A. (1979), "Estimates of Income for Small Places: An Application of James-Stein Procedures to Census Data," Journal of the American Statistical Association, 74, 269-277. Ghosh, M. (1992), "Constrained Bayes Estimation With Applications," Journal of the American Statistical Association, 87, 533-540. Gonzalez, M.E. (1973), "Use and Evaluation of Synthetic Estimates," Proceedings of the Social Statistics Section, American Statistical Association, 33-36. Gonzalez, M.E. (1979), "Case Studies on the Use and Accuracy of Synthetic Estimates: Unemployment and Housing Applications" in Synthetic Estimates for Small Areas (National Institute on Drug Abuse, Research Monograph 24), Washington, D.C.: U.S. Government Printing Office. Gonzalez, M.E. and Hoza, C. (1978), "Small-Area Estimation with Application to Unemployment and Housing Estimates," Journal of the American Statistical Association, 73, 7- 15. Gonzalez, M.E. and Waksberg, J (1973), "Estimation of the Error of Synthetic Estimates," paper presented at the first meeting of the International Association of Survey Statisticians, Vienna, Austria, 18-25 August, 1973. Heeringa, S.G. (1981), "Small Area Estimation Prospects for the Survey of Income and Program Participation," Proceedings of the Section on Survey Research Methods, American Statistical Association, 133-138. Holt, D., Smith, T.M.F., and Tomberlin, T.J. (1979), "A Model-Based Approach to Estimation for Small Subgroups of a Population," Journal of the American Statistical Association, 74, 405- 410. Kalton, G. (1987), "Panel Discussion" in Small Area Statistics, New York: John Wiley and Sons. Laake, P. (1979), "A Prediction Approach to Subdomain Estimation in Finite Populations," Journal of the American Statistical Association, 74, 355-358. Lahiri, P. (1990), "Adjusted Bayes and Empirical Bayes Estimation in Finite Population Sampling," Sankhya B, 52, 50-66. MacGibbon, B. and Tomberlin, T.J. (1989), "Small Area Estimation of Proportions Via Empirical Bayes Techniques," Survey Methodology, 15-2, 237-252. Malec, D., Sedransk, J., and Tompkins, L. (1993), "Bayesian Predictive Inference for Small Areas for Binary Variables in the National Health Interview Survey." In Case Studies in Bayesian Statistics, eds., Gatsonis, Hodges, Kass and Singpurwalla. New York: Springer Verlag. Miller, S.M. (1992), "Confidence Interval Coverage for Biased Normal Estimators," Proceedings of the Section on Survey Research Methods, American Statistical Association. National Center for Health Statistics (1968), Synthetic State Estimates of Disability (PHS Publication No. 1759), Washington, D.C.: U.S. Government Printing Office. National Institute on Drug Abuse (1979), Synthetic Estimates for Small Areas (NIDA Research Monograph 24), Washington, D.C.: U.S. Government Printing Office. Prasad, N.G.N. and Rao, J.N.K. (1990), The Estimation of the Mean Squared Error of Small Area Estimators," Journal of the American Statistical Association, 85, 163-171. Purcell, N.J. and Kish, L. (1979), "Estimation for Small Domains," Biometrics, 35, 365-384. Raback, G. and Sarndal, C.E. (1982), "Variance Reduction and Unbiasedness for Small Area Estimators," Proceedings of the Social Statistics Section, American Statistical Association, 541- 544. Royall, R.A. (1973), "Discussion of papers by Gonzalez and Ericksen," Proceedings of the Social Statistics Section, American Statistical Association, 42-43. Royall, R.A. (1979), "Prediction Models in Small Area Estimation," in Synthetic Estimates for Small Areas (National Institute on Drug Abuse, Research Monograph 24), Washington, D.C.: U.S. Government Printing Office. Sarndal, C.E. (1984), "Design-Consistent versus Model-Dependent Estimation for Small Domains," Journal of the American Statistical Association, 79, 624-631. Sarndal, C.E. and Hidiroglou, M.A. (1989), "Small Domain Estimation: A Conditional Analysis," Journal of the American Statistical Association, 84, 266-275. Schaible, W.L. (1993), "Use of Small Area Estimators in U.S. Federal Programs," in Small Area Statistics and Survey Designs, Vol. 1, Central Statistical Office, Warsaw, Poland. Schaible, W.L., Brock, D.B., and Schnack, G.A. (1977), "An Empirical Comparison of the Simple Inflation, Synthetic and Composite Estimators for Small Area Statistics," Proceedings of the Social Statistics Section, American Statistical Association, 1017-1021. Schaible, W.L., Brock, D.B., Casady, R.J., and Schnack, G.A. (1979), Small Area Estimation: An Empirical Comparison of Conventional and Synthetic Estimators for States, (PHS Publication No. 80-1356), Washington, D.C.: U.S. Government Printing Office. Spjovoll, E. and Thomsen, I. (1987), "Application of Some Empirical Bayes Methods to Small Area Statistics," Proceedings of the International Statistical Institute, Vol. 2, 435-449. CHAPTER 2 Synthetic Estimation in Followback Surveys at The National Center for Health Statistics Joe Fred Gonzalez, Jr., Paul J. Placek, and Chester Scott National Center for Health Statistics 2.1 Introduction and Program History The National Center for Health Statistics (NCHS) through its vital registration system collects and publishes data on vital events (births and deaths) for the United States (NCHS 1989). NCHS produces national, State, county, and smaller area vital statistics for sociodemographic and health characteristics which are available from birth and death certificates. The Division of Vital Statistics of NCHS produces annual summary tables for the United States showing trends in period and cohort fertility measures and characteristics of live births. Also, NCHS produces detailed tabulations by place of residence and occurrence for each State, county, and city with a population of 10,000 or more by race and place of delivery and place of residence for population-size groups in metropolitan and nonmetropolitan counties within each State by race, attendant and place of delivery, and birth weight. These statistics are based on a complete count of vital records. In addition to the limited vital statistics tabulations which are produced annually, there has always been a continuing need for more detailed national and State level estimates of health status, health services, and health care utilization related to vital events. Because vital records (birth and death certificates) serve both legal and statistical purposes, they provide limited social, demographic, health, and medical information. Each vital record is a one page document with extremely limited information. The data from these vital records can be augmented, however, through periodic "followback" surveys. These surveys are referred to as "followback" because they obtain additional information from sources named on the vital record. A followback survey is a cost effective means of obtaining supplementary information for a sample of vital events. From the sample it is possible to make national estimates of vital events according to characteristics not otherwise available. Examples of supplementary information which may be needed by health researchers, health program planners, and health policy makers are: mother's smoking habits before and during pregnancy; complications of pregnancy; drug or surgical procedure to induce or maintain labor; amniocentesis during pregnancy; electronic fetal monitoring; respiratory distress syndrome; infant jaundiced; medical x-ray use; birth injuries; and, congenital anomalies. Periodic followback surveys respond to the changing data needs of the public health community without requiring changes in the vital record forms. The specific NCHS followback surveys that will be discussed in this chapter are the 1980 National Natality Survey (NNS) and the 1980 National Fetal Mortality Survey (NFMS) (NCHS 1986). In order to produce State estimates for certain health characteristics not available on the vital records, synthetic estimation (NCHS 1984a, 1984b) was applied to national data from the 1980 NNS and 1980 NFMS. In addition to the usual appeal of using synthetic estimation over direct estimation, especially when small sample sizes are concerned, synthetic State estimates were compared to direct State estimates as well as the "true" values for a limited number of variables from State vital statistics via fetal death records and birth and death certificates. 2.2 Program Description, Policies and Practices The 1980 NNS is based on a probability sample of 9,941 from a universe of 3,612,258 live births that occurred in the United States during 1980. The NNS sample included a four-fold oversampling of low birth weight infants. The live birth certificate represents the basic source of information. Based on information from the sample birth certificates, eight page Mother's questionnaires were mailed to mothers who were married. These mothers were asked to provide information on prenatal health practices, prenatal care, previous pregnancies, and social and demographic characteristics of themselves and their husbands. Each mother was also asked to sign a consent statement authorizing NCHS to obtain supplemental information from her medical records. If the mother did not respond after two questionnaires were sent by mail, a telephone interviewer attempted to complete an abbreviated questionnaire and to obtain a consent statement. To ensure their privacy, unmarried mothers were not contacted. As a result of sending the Mother's questionnaire only to married mothers, the 1980 NNS population of inference for data collected through the Mother's questionnaire was 2,944,580 live births. Regardless of the mother's marital status, questionnaires were mailed to the hospital's and to the attendants at delivery (for example, physicians or nurse-midwives) named on the birth certificates. A questionnaire was sent to the hospital for each sample birth that occurred either in a hospital or en route to a hospital. If the mother signed a consent statement authorizing NCHS to obtain supplemental medical information, a copy was included with the questionnaire. The focus of the hospital questionnaire was on characteristics of labor and delivery, health characteristics of the mother and infant, information on prenatal care visits, and information on radiation examinations and treatments received by the mother during the 12 months before delivery of the sample birth. For the hospital component of the 1980 NNS, the population of inference was 3,580,700 live births. The 1980 NNS is composed of information from birth certificates and information from questionnaires sent to married mothers, hospitals, attendants at delivery, and providers of radiation examinations and treatments. The survey represents an extensive source of information concerning specific maternal and child health conditions and obstetric practices for live births in the United States. The 1980 NNS response rates were 79.5 percent for mothers, 76.1 percent for hospitals, and 61.6 percent for physicians. The 1980 NFMS is based on a probability sample of 6,386 fetal deaths (out of a universe of 19,202 fetal deaths) with gestation of 28 weeks or more, or delivery weight of 1,000 grams or more, that occurred in the United States during 1980. The report of fetal death represent the basic source of information in this survey. Married mothers, hospitals, attendants at delivery, and providers of radiation examinations and treatments were surveyed under the same conditions as those described for the 1980 NNS. The 1980 NFMS populations of inference for all fetal deaths, fetal deaths in hospitals, and fetal deaths to married mothers were 19,202, 18,930, and 14,790, respectively. The same questionnaires were used for both surveys. Although some questions pertained only to live births and other pertained only to fetal deaths, instructions to skip inappropriate questions were included in the questionnaires. The sampling design for the NFMS was developed so that the NFMS would be large enough to permit comparisons between live births in the NNS and fetal deaths in the NFMS. The 1980 NFMS response rates were 74.5 percent for mothers, 74.0 percent for hospitals, and 55.0 percent for physicians. Table 1 presents the 1980 NNS and NFMS distribution of sample cases of live births and fetal deaths by State of occurrence. As shown in Table 1, it may be possible to produce direct State level estimates of certain health characteristics for some of the larger States. However, the sample sizes for most States are generally too small to produce reliable direct State estimates. This was the main justification for exploring synthetic State estimation as an alternative for producing State level estimates. 2.3 Estimator Documentation The underlying rationale for synthetic estimation is that the distribution of a health characteristic is highly related to the demographic composition of the population (NCHS 1984a). It is assumed that differences in the prevalence of the characteristics between two areas are due primarily to differences in demographic composition (e.g. age, race, sex, etc.). That is, it is assumed that a particular measure would be the same in two States that had the same population composition with respect to certain demographic variables. This rationale was used to select the demographic variables that were deemed to be the most appropriate and relevant to the 1980 NNS and NFMS in order to produce Synthetic State estimates. The following is the basic estimator that was used to produce Synthetic State estimates of proportions for certain health variables from the 1980 National Natality Survey (NNS) and the 1980 National Fetal Mortality Survey (NFMS). Click HERE for graphic. Table 2 gives an illustration of the computation of the synthetic State estimate of the percent jaundiced infants in Pennsylvania in 1980. The stub of Table 2 shows the 25 demographic cells (race, age of mother, and live-birth order groups) that were used to produce the Synthetic State estimates. Column (1) shows the national (based on the 1980 NNS) estimates of percent of live births that were jaundiced in each of the respective 25 demographic cells. Column (2) shows the number of hospital births (derived from State Vital Registration System) within the 25 demographic cells in Pennsylvania. Column (3), the estimated number of jaundiced live births in Pennsylvania, is computed by taking the product of entries in columns (1) and (2) within each of the 25 respective cells. Finally, the Synthetic State estimate is found by taking the ratio of the sum of column (3) to the sum of column (2). Since there were three different populations of inference (all vital events, vital events to married mothers, and vital events in hospitals) for each of the 1980 NNS and NFMS, appropriate State aggregates of vital events were incorporated into the calculation of corresponding synthetic State estimates (NCHS 1984a, 1984b). 2.4 Evaluation Practices The following is a description of some of the tabulations that were produced. Table 3 gives Synthetic State estimates of 11 health characteristics of mothers and infants for five selected States. A complete listing of all 57 NNS/NFMS health variables for which Synthetic State estimates were produced can be found in Tables 2-8 in (NCHS 1984a, 1984b). Click HERE for graphic. The synthetic State estimates are subject to sampling error because they are based on corresponding national estimates derived from the 1980 NNS and NFMS by race, maternal age, and live-birth order group. Therefore, the standard errors of the synthetic State estimates are relatively small because they are based on the standard errors of the national estimates. The standard errors for the NNS and NMFS were estimated by a balanced-repeated-replicated procedure using 20 replicate half samples. This procedure estimates the standard errors for survey estimates through the observation of the variability of estimates based on replicate half samples of the total sample; This variance estimation procedure was developed and described by McCarthy (NCHS 1966, 1969). Although the synthetic State estimate has a relatively small standard error, it is subject to bias. The bias is a measure of the extent to which the national maternal age, race, live-birth order specific estimates differ from the true values for a given State. The closer the demographic variables used in the synthetic estimate come to accounting for all the interstate variation in a particular health characteristic, the smaller the bias will be. Unfortunately, the bias cannot be computed without knowing the true values. However, through the U.S Vital Registration System, true State values for vital events (collected through birth and death certificates) are known for a limited number of available sociodemographic and health characteristics. Therefore, we can compare certain synthetic estimates with their corresponding true values. This yields a degree of confidence for the synthetic estimates of similar characteristics which cannot be checked against the true values from State vital statistics. Thus, the evaluation of this study only provides an indicator of the quality of the synthetic State estimates. The last two columns of Table 4 show the mean square error (MSE) of the NNS synthetic. estimates as compared with the MSE of the NNS direct estimates. The MSE of an estimate x is the variance of x plus the square of the bias of x, i.e. Click HERE for graphic. 2.5 Current Problems And Activities Work is currently underway at NCHS to produce synthetic State estimates from the 1988 National Maternal and Infant Health Survey (NMIHS) which is very similar to its predecessor the 1980 NNS and NFMS. In the NMIHS 9,953 out of a universe of 3,898,922 live-birth certificates are linked with mothers' responses on 35-page questionnaires about the mothers' prenatal health behavior, maternal health, the birth experience, and infant health. The 1988 NMIHS live birth estimates will be used to produce synthetic State estimates by infant's race. birth weight, and maternal age and marital status. Click HERE for graphic. Click HERE for graphic. Click HERE for graphic. Click HERE for graphic. REFERENCES National Center for Health Statistics: Vital Statistics of the United States, 1987 Vol. 1, Natality, DHHS Pub. No. (PHS) 89-1100. Public Health Service, Washington. U.S. Government Printing Office, 1989. National Center for Health Statistics, K.G. Keppel, R.L. Heuser, P.J. Placek, et al.: Methods and Response Characteristics, 1980 National Natality and Petal Mortality Surveys. Vital and Health Statistics, Series 2, No. 100. DHHS Pub No. (PHS) 86-1374. Public Health Service, Washington. U.s. Government Printing Office, Sept. 1986. National Center for Health Statistics: State Uses of Followback Survey Data, R.L. Heuser, K.G. Keppel, C.A. Witt, and P.J.Placek, Presented at the Annual Meeting of the Association for Vital Records and Health Statistics, July 9-12, 1984, Niagara Falls, NY. National Center for Health Statistics: R.L. Heuser, K.G. Keppel, C.A. Witt, and P.J. Placek, Synthetic Estimation Applications form the 1980 National Natality Survey (NNS) and the 1980 National Fetal Mortality Survey (NFMS), Presented at the NCHS Data Use Conference on Small Area Statistics, August 29-31, 1984, Snowbird, Utah. National Center for Health Statistics, P.J. McCarthy: Replication: An Approach to the Analysis of Data From Complex Surveys. Vital and Health Statistics, Series 2, No. 14, PHS Pub No. 1000. Public Health Service. Washington, U.S. Government Printing Office, April 1966. National Center for Health Statistics, P.J. McCarthy: Pseudoreplication: Further Evaluation and Appliication of the Balanced Half-Sample Technique. Vital and Health Statistics. Series 2, No. 31. DHEW Pub No. (HSM) 73-120. Health Services and Metal Health Administration. Washington. U.S. Government Printing Office, Jan. 1969. * Chapter 8 (authored by Donald Malec) of this report contains several references on small area estimation as applied to the National Health Interview Survey of the National Center for Health Statistics. CHAPTER 3 State, Metropolitan Area, and County Income Estimation Wallace Bailey, Linnea Hazen, and Daniel Zabronsky Bureau of Economic Analysis 3.1 Introduction and Program History 3.1.1 Program Description The Bureau of Economic Analysis (BEA) maintains a program of State and local area (county and metropolitan area) economic measurement that centers on the personal income measure. This program originated in 1939 when estimates of income payments to individuals by State were first published. At the national level, personal income is the principal income measure in the personal income and outlay account, one of the five accounts that compose the national income and product accounts. The State and local area personal income estimates are derived by disaggregating the detailed components of the national personal income estimates to States and counties. Estimates for all other geographic areas are made by aggregating either the State or county estimates in the appropriate combinations. This building block approach permits estimates for areas whose boundaries change over time, such as metropolitan areas, to be presented on a consistent geographic definition for all years. 3.1.2 Uses of the State and Local Area Income Estimates BEA's State and local area income estimates are widely used in the public and private sector to measure and track levels and types of incomes received by persons living or working in an area. They provide a framework for the analysis of each area's economy and serve as a basis for decision making in both the public and private sectors. Personal income is among the measures used in evaluating the socioeconomic impact of public- and private-sector initiatives; for example, it is widely used in preparing the environmental impact statements required by the National Environmental Policy Act of 1969. One of the first uses made of State personal income estimates (or a derivative) was as a variable in formulas for allocating Federal funds to States. The most often used derivative is per capita personal income, which is computed using the Census Bureau's estimates of total population; these population estimates are described by Long in Chapter 4 of this report. At present, BEA's State personal income estimates are used by the Federal Government to allocate over $92 billion annually for various Federal domestic programs, including the medical assistance (Medicaid) program, and the aid to families with dependent children program. Table 3.1 highlights the major Federal Government programs which use BEA personal income estimates in allocation formulas for Federal domestic assistance funds. Federal agencies also use the components of personal income in econometric models, such as those used to project energy and water use. The U.S. Forest Service is using these estimates to identify resource dependent rural areas and to allocate funds for their economic diversification as required by the National Forest-Dependent Rural Communities Economic Diversification Act of 1990. The U.S. Census Bureau uses the BEA estimates of State per capita personal income as the key predictor variable in its estimates of mean annual income for 4-person families by State. These estimates are described by Fay, Nelson, and Litow in Chapter 9 of this report. During the past decade, State governments have substantially increased their use of the State personal income estimates. The estimates are used in the measurement of economic bases and in models developed for planning for such things as public utilities and services. They are also used to project tax revenues. In recent years, legislation that limits a State's expenditures or tax authority by the level of, or changes in, State personal income or to one of its components has been enacted in 16 States. These 16 States account for nearly one-half of the U.S. population. Some of these States used BEA's annual State personal income estimates; the others use fiscal year estimates derived from BEA's quarterly State personal income estimates (ACIR, 1990). State governments also use the local area estimates to measure the economic base of State planning areas. University schools of business and economics, often worldng under contract for State and local governments, use the BEA local area estimates for theoretical and applied economic research. Businesses use the estimates to evaluate markets for new or established products and to determine areas for the location, expansion, and contraction of their activities. Trade associations and labor organizations use them for product and labor market analyses. 3.1.2 A History of BEA's Regional Income Estimates In the mid-1930's, BEA's predecessor began work on the estimation of regional income as part of the effort to explain the processes and structure of the Nation's economy. As a result of its work, it produced a report that showed State estimates of total "income payments to individuals' in May 1939 (Nathan and Martin, 1939). These income payments were defined as the sum of (1) wages and salaries, (2) other labor income and relief, (3) entrepreneurial withdrawals, and (4) dividends, interest, and net rents and royalties. In 1942, the State estimates of wages and salaries and entrepreneurial income were expanded to include a further breakdown by broad industry group--agriculture, other commodity-producing industries, distribution, services, and government. The industry breakdown was for 1939, when the availability of census information on payrolls and the employed labor force by industry and by State made possible more reliable estimates than for prior years (Creamer and Merwin, 1942). The estimates for most nonagricultural industries and for the military services were based on reports in which establishments, not employees, were classified by State and in which the State of residence of the employees was not indicated; therefore, the estimates for these industries were on a "place-of-work" (where-earned) basis. No systematic adjustment was made in the total income payments series to convert the estimates to a "place-of-residence" (where-received) basis. However, using the limited information that was available, residence adjustments were made for a few States for the per capita series. During the later 1940's and early 1950's, BEA continued to work on improving these estimates by seeking additional source data and by improving the estimating methods that were used. The industrial detail of the wage and salary estimates was expanded to include each Standard Industrial Classification (SIC) division and additional detail for some SIC divisions. As one result of the major reworking and expansion of the national income and product accounts, BEA developed State personal income--a measure of income that is more comprehensive than State income payments. During the 1960's and 1970's, BEA continued its work to provide more information about regional economies. Annual State estimates of disposable personal income were published in the April 1965 Survey of Current Business (Survey), and the first set of quarterly estimates of State personal income was published in the December 1966 Survey. Estimates of personal income for metropolitan areas were published in the May 1967 Survey, for nonmetropolitan counties in the May 1974 Survey, and for metropolitan counties in the April 1975 Survey. In the late 1970's, BEA introduced annual estimates of employment for States, metropolitan areas, and counties. Refinement of the residence adjustment procedures and a fuller presentation of industrial detail for earnings--the term introduced to cover wages and salaries plus other labor income plus proprietors' income--emerged in the estimates published in 1974. The residence adjustment procedures had been extended to all States in 1966, but the residence adjustment estimates (i.e., the net flows of interstate commuters' earnings), along with earnings by industry on a place-of-work basis, were not published explicitly until 1974. 3.2 The Regional Economic Measurement Program 3.2.1 Estimating Schedule for State and Local Area Personal Income Series The annual estimates of State personal income for a given year are subject to successive refinement. Preliminary estimates, based on the current quarterly series, are published each April, 4 months after the close of the reference year, in the Survey. The following August, more reliable annual estimates are published. These estimates are developed independently of the quarterly series and are prepared in greater component detail, primarily from Federal and State government administrative records. The annual estimates published in August are subsequently refined to incorporate newly available information used to prepare the local area estimates for the same year. These revised State estimates, together with the local area estimates, are published the following April. The annual estimates emerging from this three-step process are subject to further revision for several succeeding years (the State estimates in April and August and the local area estimates in April), as additional data become available. For example, the 1992 State estimates that were first released in April 1993 will be revised in August 1993 and in April and August of subsequent years; the 1991 local area estimates that were first released in April 1993 will be revised in April of 1994 and of subsequent years. The routine revisions of the State estimates for a given year are normally completed with the fourth April publication, and the local area estimates, with the third April publication. After that, the estimates will be changed only to incorporate a comprehensive revision of the National Income and Product Accounts--which takes place approximately every 5 years--or to make important improvements to the estimates through the use of additional or more current State and local area data. Quarterly estimates of State personal income, which are available approximately 4 months after the close of the reference quarter, are published regularly in the January, April, July, and October issues of the Survey. In,October and again the following April, the quarterly series for the 3 previous years is revised for consistency with the revised annual estimates. In January and July, at least the quarter immediately preceding the current quarter is revised. 3.2.2 Availability of State and Local Area Estimates The State and local area personal income and employment estimates are available through the Regional Economic Information System (REIS), which operates an information retrieval service that provides a variety of standard and specialized analytic tabulations for States, counties and specified combinations of counties. Standard tabulations include personal income by type and earnings by industry, employment by industry, transfer payments by program, and major categories of farm gross income and expenses. These tabulations are available from REIS in magnetic tape, computer printout, microcomputer diskette and CD-ROM; some of the tabulations are also available electronically on the Department of Commerce's Economic Bulletin Board, available through the National Technical Information Service. In addition, summary tabulations of the State and local estimates are published regularly in BEA's major publication, the Survey. An extensive set of State-level historical estimates is available (BEA, 1989). BEA also makes its regional estimates available through the BEA User Group, members of which include State agencies, universities, and Census Bureau Primary State Data Centers. BEA provides its estimates of income and employment for States, metropolitan areas, and counties to these organizations with the understanding that they will make the estimates readily available. Distribution in this way encourages State universities and State agencies to use data that are comparable for all States and counties and that are consistent with national totals; using comparable and consistent data enhances the uniformity of analytic approaches taken in economic development programs and improves the recipients' ability to assess local area economic developments and to service their local clientele. 3.3 BEA Annual State and County Personal Income Estimation Methodology 3.3.1 Overview The following discussion will focus on the annual estimates of State and county personal income. BEA's quarterly State personal income, annual State disposable personal income, annual State and county full- and part-time employment, and gross State product (GSP) estimates are produced in a manner similar to those described below. (The methodologies for quarterly State personal income and for annual State disposable personal income are described in BEA (1989, pp. M-32-37); the methodology for GSP is described in BEA (1985) and in Trott, Dunbar, and Friedenberg (1991, pp. 43-45). The personal income of an area is defined as the income received by, or on behalf of, all the residents of the area. It consists of the income received by persons from an sources, that is, from participation in production, from both government and business transfer payments, and from government interest. Personal income is measured as the sum of wage and salary disbursements, other labor income, proprietors' income, rental income of persons, personal dividend income, personal interest income, and transfer payments less personal contributions for social insurance. Per capita personal income is measured as the personal income of the residents of an area divided by the resident population of the area. At the national level, personal income is part of the personal income and outlay account, which is one of five accounts in a set that constitutes the national income and product accounts. Such accounts do not now exist below the national level; however, personal income has long been available for States and local areas. In addition, GSP, which corresponds to the national measure gross domestic product, and some elements of personal outlays (personal tax and nontax payments) are available for States but not for local areas. GSP is estimated separately from State personal income, but the two measures share most of the elements of wages and salaries, other labor income, and proprietors' income by State of work. For a tabular representation of the relationships among gross domestic product, State earnings, and GSP, see Table 2 in Trott et. al. (1991, p. 44). 3.3.2 Differences Between the National and Subnational Estimates The definitions underlying the State and local area estimates of personal income are essentially the same as those underlying the national estimates of personal income. However, the national estimates of personal income include the labor earnings (wages and salaries and other labor income) of residents of the United States temporarily working abroad, whereas the subnational estimates include the labor earnings of persons residing only in the 50 States and the District of Columbia. Specifically, the national estimates include the labor earnings of Federal civilian and military personnel stationed abroad and of residents who are employed by U.S. firms and are on temporary foreign assignment. An "overseas" adjustment is made to exclude the labor earnings of these workers from the national totals before the totals are used as controls for the State estimates. An important classification difference between national and subnational estimates relates to border workers--that is, residents of the United States who work in adjacent countries (such as Canada) and foreigners who work in the United States but who reside elsewhere. At the national level, the net flow of the labor earnings of border workers and the labor earnings of U.S. residents employed by international organizations and by foreign embassies and consulates in the United States are included in the measurement of the "rest-of-the-world" sector. At the State and local area levels, however, only the labor earnings of U.S. residents employed by international organizations and by foreign embassies and consulates in the United States are treated as a component of personal income. Border workers are treated as commuters and their earnings flows are reflected in personal income through the residence adjustment procedures. Statistical differences between the national and subnational series may reflect the different estimating schedules for the two series. The State and local area estimates usually incorporate source data that are not available when the national estimates are prepared. The national estimates are usually revised the following year to reflect the more current State and local area data. 3.3.3 Sources of Data BEA uses information collected by others to prepare its estimates of State and local area personal income. Generally, two kinds of information are used to measure the income of persons: Information generated at the point of disbursement of the income and information elicited from the recipient of the income. The first kind is data drawn from the records generated by the administration of various Federal and State government programs; the second kind is survey and census data. The following are among the more important sources of the administrative record data: The State unemployment insurance programs of the Employment and Training Administration, Department of Labor; the social insurance programs of the Social Security Administration and the Health Care Financing Administration, Department of Health and Human Services; the Federal income tax program of the Internal Revenue Service, Department of the Treasury; the veterans benefit programs of the Department of Veterans Affairs; and the military payroll systems of the Department of Defense. The two most important sources of census data are the censuses of agriculture and of population. (BEA uses little survey data to prepare the State and local area estimates; however, the Department of Agriculture makes extensive use of surveys to prepare the State farm income estimates and the county cash receipts and crop production estimates that BEA uses in the derivation of the farm income components of personal income.) The data obtained from administrative records and censuses are used to estimate about 90 percent of personal income. Data of lesser scope and relevance are used for the remaining 10 percent. When data are not available in time to be incorporated into the current estimating cycle, interim estimates are prepared using the previous year's State or county distribution. The interim estimates are revised during the next estimating cycle to incorporate the newly available data. Using data that are not primarily designed for income measurement has several advantages and disadvantages. Using administrative record data and census data, BEA can prepare the estimates of State and local area personal income on an annual basis, in considerable detail, at relatively low cost, and without increasing the reporting burden of businesses and households. However, because these data are not designed primarily for income measurement, they often do not precisely "match" the series being estimated and must be adjusted to compensate for differences in content (definition and coverage) and geographic detail. 3.3.4 Controls and the Allocation Procedure The national estimates for most components of wages and salaries and transfer payments, which together account for about 75 percent of personal income, are based largely on the sum of subnational source data, and the procedure used to prepare the State and county estimates causes only minor changes to the source data. For other components of personal income, either detailed geographic coding is not available for all source data, or more comprehensive and more reliable information is available for the Nation than for States and counties. For these reasons, the estimates of personal income are first constructed at the national level. The subnational estimates are constructed as elements of the national totals, using the subnational data. Thus, the national estimates, with some adjustment for definition, serve as the "control" for the State estimates, and the State estimates, in turn, serve as controls for the county estimates. The State estimates are made by allocating the national total for each component of personal income to the States in proportion to each State's share of a related economic series. Similarly, the county estimates are made, in somewhat less component detail, by allocating the State total. In some cases, the related series used for the allocation may be a composite of several items (e.g., wages, tips, and pay-in-kind) or the product of two items (e.g., average wages times the number of employees). In every case, the final estimating step for each income estimate is its adjustment to the appropriate higher level total. This procedure is called the allocation procedure. The allocation procedure, as used to estimate a component of State personal income, is given by Click HERE for graphic. The source data that underlie the national estimates are frequently more timely, detailed, and complete than the available State and county data. The use of the allocation procedure imparts some of these aspects of the national estimates to the subnational estimates and allows the use of subnational data that are related but that do not always precisely match the series being estimated. The use of this procedure also yields an additive system wherein the county estimates sum to the State totals and the State estimates sum to the national total. 3.3.5. Place of Measurement Personal income, by definition, is a measure of income received; therefore, estimates of State and local area personal income should reflect the residence of the income recipients. However, the data available for regional economic measurement are frequently recorded by the recipients' place of work. The data underlying the estimates can be viewed as falling in four groups according to the place of measurement. (1) For the estimates of wages and salaries, other labor income, and personal contributions for social insurance by employees, most of the source data are reported by industry in the State and county in which the employing establishment is located; therefore, these data are recorded by place of work. The estimates based on these data are, subsequently adjusted to a place-of-residence basis for inclusion in the personal income measure. (2) For nonfarm proprietors' income and personal contributions for social insurance by the self-employed, the source data are reported by tax-filng address. These data are largely recorded by place of residence. (3) For farm proprietors' income, the source data are reported and recorded at the principal place of production, which is usually the county in which the farm has most of its land. (4) For military reserve pay, rental income of persons personal dividend income, personal interest income, transfer payments, and personal contributions for supplementary medical insurance and for veterans life insurance, the source data are reported and recorded by the place of residence of the income recipients. 3.3.6 Sources and Methods for Annual State and County Income Estimates 3.3.6.1 Framework Personal income is estimated as the sum of its detailed components; the major types of payments that comprise those components are shown in Table 3.2, together with the related percents of personal income and the principal sources of data used to estimate the components. The following methodology presentation consists of a section for each of the six types of payment and a section for the residence adjustment. The methodologies for some types of payment and for many of the individual income components are omitted from this presentation, but a complete presentation is available (BEA 1991, pp. M-7-27). 3.3.6.2 Wage and salary disbursements Wage and salary disbursements, which accounted for about 58 percent of total personal income at the national level in 1990, are defined as the monetary remuneration of employees, including the compensation of corporate officers; commissions, tips, and bonuses; and receipts in kind that represent income to the recipient. They are measured before deductions, such as social security contributions and union dues. The estimates reflect the amount of wages and salaries disbursed during the current period, regardless of when they were earned. The following description of the procedures used in making the estimates of wage and salary disbursements is divided into three sections: Wages and salaries that are covered under the unemployment insurance (UI) program, wages and salaries that are not covered under the UI program, and wages and salaries that are paid in kind. Wages and salaries covered by the UI program The estimates of about 95 percent of wages and salaries are derived from tabulations by the State employment security agencies (ESA's) from their State employment security reports (form ES-202). These tabulations summarize the data from the quarterly UI contribution reports filed with a State ESA by the employers subject to that State's UI laws. Employers usually submit reports for each "county reporting unit"--i.e., for the sum of all the employer's establishments in a county for each industry. However, in some cases, an employer may group very small establishments in a single "statewide" report without a county designation. Each quarter, the various State ESA's submit the ES-202 tabulations on magnetic tape to the Bureau of Labor Statistics (BLS), which provides a duplicate tape to BEA. The tabulations present monthly employment and quarterly wages for each county in Standard Industrial Classification four-digit detail. (The ES-202 tabulations through 1987 reflect the 1972 SIC, and those for 1988-90, the 1987 SIC.) Under the reporting requirements of most State UI laws, wages include bonuses, tips, gratuities, and the cash value of meals and lodging supplied by the employer. The BEA estimates of wage and salary disbursements are made, with a few exceptions, at the SIC two-digit level. However, the availability of the ES-202 data in SIC four-digit detail facilitates the detection of errors and anomalies; this detail also makes it possible to isolate those SIC three-digit industries for which UI coverage is too incomplete to form a reliable basis for the estimates. In this case, the SIC two-digit estimate is prepared as the sum of two pieces: The fully covered, portion, which is based on the ES-202 data, and the incompletely covered portion, which is estimated as described in the section on wages and salaries not covered by the UI program. The ES-202 wage and data do not precisely meet the statistical and conceptual requirements for BEA's personal income estimates. Consequently, the data must be adjusted. to meet the requirements more closely. The adjustments affect both the industrial and geographic patterns of the State and county UI-based wage estimates. Adjustment for statewide reporting.--Wages and salaries reported for statewide units are allocated to counties in proportion to the distribution of the wages and salaries reported by county; the allocations for each State are made for each private-sector industry (generally at the SIC two-digit level) and for five government components. Adjustment for industry nonclassification.--The industry detail of the ES-202 tabulations regularly shows minor amounts of payroll that have not been assigned to any industry. For each State and county, the amount of ES-202 payrolls in this category is distributed among the industries in direct proportion to the industry-classified payrolls. Misreporting adjustment.--This adjustment--the addition of estimates of wages and salaries subject to UI reporting that employers do not report--is made to the ES-202 data for all covered private-sector industries. At the national level, the estimate for each industry is made in two parts--one for the underreporting of payrolls on UI reports filed by employers and one for the payrolls of employers that fail to file Ul reports (Parker, 1984). The source data necessary to replicate this methodology below the national level are not available. Instead, the national adjustment for each industry is allocated to States and counties in proportion to ES-202 payrolls. Adjustments to government components.--Alternative source data are substituted for the ES-202 data when the latter series reflects excessively large proportions of Federal civilian payrolls that are not reported by county or of State government payrolls that are apparently reported in the wrong counties. For Federal civilian wages and salaries, the alternative source data are tabulations of employment by agency and county prepared by the Office of Personnel Management. For State government wages and salaries, the alternative source data are place-of-work wage data derived from an unpublished tabulation of journey-to-work (JTW data from the 1980 Census of Population. (All income estimates using 1980 Census of Population data will be updated to incorporate 1990 Census of Population data in a regional comprehensive revision to be released in the spring of 1994.) Adjustments for noncovered elements of UI-covered industries.--BEA presently makes adjustments for the following noncovered elements: 0 Tips; 0 Commissions received by insurance solicitors and real estate agents; 0 Payrolls of electric railroads, railroad carrier affiliates, and railway labor organizations; 0 Salaries of corporate officers in Washington State; 0 Payrolls of nonprofit organizations exempt from UI coverage because they have fewer than four employees;. 0 Wages and salaries of students employed by the institutions of higher education in which they are enrolled; 0 Allowances paid to Federal civilian employees in selected occupations for uniforms; and 0 Salaries of State and local government elected officials and members of the judiciary. Except for tips, these elements are exempted from State UI coverage. Tips are covered by the various UI laws. BEA assumes that this form of income payment is considerably underreported, and it therefore makes additional estimates of tips in industries where tipping is most customary. National and State estimates of each of the noncovered elements are made (based on either direct data or indirect indicators). These estimates are added to the ES-202 payroll amount for the industry of the noncovered element to produce the final estimates for that industry. Because of the lack of relevant data, county estimates are made by allocating the final State total by the distribution of ES-202 payrolls for the appropriate industry. Wages and salaries not covered by the UI program Eight industries are treated as noncovered in making the State and county estimates of wage and salary disbursements: (1) Farms, (2) farm labor contractors, (3) railroads, (4) private elementary and secondary schools, (5) religious membership organizations, (6), private households, (7) military, and (8) "other." The estimates for these industries are based on a variety of sources. For example, the estimates for railroads ar based mainly on employment data provided by the Association of American Railroads, and the estimates for the military services are based mainly on payroll data provided by the Department of Defense. See BEA (1991) for the methodology for the noncovered industries. Wages and salaries paid in kind The value of food, lodging, clothing, and miscellaneous goods and services furnished to employees by their employers as payment, in part or in full, for services perfomed is included in the wage and salary component of personal income and is referred to as "pay-in-kind." The estimates for UI-covered industries are prepared as an integral part of total wages and salaries for those industries, based on the ES-202 data. The estimates for most on the noncovered industries are based on pertinent employment data. See BEA (1991) for the methodology for pay-in-kind. 3.3-6.3 Other labor income Other labor income (OLI), which accounted for about 5.5 percent of total personal income at the national level in 1990, consists primarily of employer contributions to private pension and welfare funds; these employer contributions account for approximately 98 percent of OLI. The "all other" component of OLI consists of directors' fees, judicial fees, and compensation of prisoners. Employer contributions for social insurance, which are paid into govemment-administered funds, are not included in OLI; under national income and product account conventions, it is the benefits paid from social insurance funds--which are classified as transfer payments--that are measured as part of personal income, not the employer contributions to the funds. Employer contributions to private pension and welfare funds Private pension and profit-sharing funds, group health and life insurance, and supp1emental unemployment insurance.--The larger part of the national estimates of employer contributions to private pension and welfare funds is developed from Internal Revenue Service tabulations of data from proprietorship and corporate income tax returns published in Statistics of Income. However, these data are not suitable for making the subnational estimates because most multiestablishment corporations file tax returns on a companywide basis instead of for each establishment and because the State in which a corporation's principal office is located is often different from the State of its other establishments. As a result, the geographic distribution of the data tabulated from the tax returns does not necessarily reflect the place of work of the employees on whose behalf the contributions are made. For private-sector employees, the State and county estimates of employer contributions to private pension and profit-sharing funds, group health and life insurance, and supplemental unemployment insurance are made, for all types of employer contributions combined, at the SIC two-digit level, the same level of industrial detail as the wage and salary estimates. The national total of employer contributions for each industry is allocated to the States and counties in proportion to the estimates of wage and salary disbursements for the corresponding industry. The use of subnational wage estimates to allocate the national estimates of employer contributions to private pension and welfare funds is based on the assumption that the relationship of contributions to payrolls for each industry is the same at the national, State, and county levels. The procedure reflects the wide variation in contribution rates--relative to payrolls--among industries (and therefore reflects appropriately the various mixes of industries among States and counties). It does not reflect the variation in contribution rates among States and counties for a given industry. The Federal Government makes contributions to a private pension fund, called the Thrift Savings Plan, on behalf of its civilian employees who participate in the Federal Employees Retirement System (mainly employees hired after 1983). In the absence of direct data below the national level, the national estimate is allocated to States and counties in proportion to the estimates of Federal civilian wages and salaries. State government contributions to private pension plans consist of annuity payments made by State governments on behalf of selected employee groups--primarily teachers. The State estimates are based on direct data from the Teachers Insurance and Annuity Association/College Retirement Equities Fund. The county estimates are prepared by allocating the State estimates in proportion to the estimates of State and local government education wages and salaries. In the absence of direct data below the national level, the national estimates of Federal, State, and local government contributions to private welfare funds on behalf of their employees are allocated to States and counties in proportion to ES-202 employment data for each level of government. Privately administered workers' compensation.--The State estimates for this subcomponent are based mainly on direct data provided by the National Council on Compensation insurance and by the Social Security Administration; the county estimates for each SIC two-digit industry reflect the geographic distribution of wages and salaries. The methodology for this income component is given in BEA (1991). "All other" OLI The methodology for "all other" OLI--primarily directors' fees and jury and witness fees--is given in BEA (1991). The State and county estimates for directors' fees--the largest of these subcomponents--reflect the geographic distribution of wages and salaries in each industry. 3.3.6.4 Proprietors' Income Proprietors' income, which accounted for about 8.5 percent of total personal income at the national level in 1990, is the income, including income-in-kind, of sole proprietorships and partnerships and of tax-exempt cooperatives. The imputed net rental income of owner-occupants of farm dwellings is included. Dividends and monetary interest received by proprietors of nonfinancial business, monetary rental income received by persons who are not primarily engaged in the real estate business, and the imputed net rental income of owner-occupants of nonfarm dwellings are excluded; these incomes are included in dividends, net interest, and rental income of persons. Proprietors' income, which is treated in its entirety as received by individuals, is estimated in two parts--nonfarm and farm. Nonfarm prorrietgrt' income Nonfarm proprietors' income is the income received by nonfarm sole proprietorships and partnerships and by tax-exempt cooperatives. The State and county estimates of the income of sole proprietors and partnerships for all but three of the SIC two-digit industries are based on 1981-83 tabulations from Internal Revenue Service (IRS) form 1040, Schedule C (for sole proprietors), and form 1065 (for partnerships). Tabulations either of gross receipts or of profit less loss from the two forms combined are used either to attribute a national total to the States or as direct data. Two national totals are used for each industry: One for income reported on the income tax returns as adjusted to conform with national income and product accounting conventions--and one for an estimate of the income not reported on tax returns. For the adjustments for unreported income, no direct data are available below the national level. The national total for each industry is attributed to States in proportion to the IRS State distribution of gross receipts for the industry. For the reported portion of nonfarm proprietors' income, the State estimates for each of 45 industries are based on the IRS distribution of profit less loss for the industry, and the estimates for each of another 20 industries (together accounting for 3 percent of total nonfarm proprietors' income) are based on the IRS distribution of gross receipts for the industry. For the latter group, the ERS distribution of profit less loss, although preferable in concept, is not used as a basis for State estimates because the extreme year-to-year volatility of the State data suggests that they are unreliable. The 1983 State estimates prepared by the foregoing methodology are extended to later years based mainly on the number of small establishments in each industry as determined from the Census Bureau's County Business Patterns; see BEA (1989) for a full description of the methodology. For the three remaining industries, limited partners' income presents a special estimating problem. In these industries--crude petroleum and natural gas extraction, real estate, and holding and investment companies--limited partnerships are often used as tax shelters. Limited partners' participation in partnerships is often purely financial; their participation more closely resembles that of investors than that of working partners. Accordingly, the usual assumption that the State from which the partnership files its tax return is the same as the residence of the individual partners is unsatisfactory. No direct data on the income of partners by their place of residence are available. The national estimates of proprietors' income for these industries are attributed to States in the same proportion as dividends received by individuals (based on all-industry dividends reported on IRS form 1040). The State estimates of the income of tax-exempt cooperatives are based on data provided by the Rural Electrification Administration (for electric and telephone cooperatives) and the Agricultural Cooperative Service (for farm supply and marketing cooperatives); see BEA (1989) for the methodology. The methodology for the county estimates of nonfarm proprietors' income is similar to the State methodology, but less direct data are used for many industries because problems with data volatility are greater at the county level. See BEA (1991) for a full description of the county methodology. Farm proprietors' income The estimation of farm proprietors' income starts with the computation of the realized net income of all farms, which is derived as farm gross receipts less production expenses. This measure is then modified to reflect current production through a change-in-inventory adjustment and to exclude the income of corporate farms and salaries paid to corporate officers. Tables showing the derivation of State and county farm proprietors' income in detail are available from the Regional Economic Information System. The concepts underlying the national and State BEA estimates of farm income are generally the same as those underlying the national and State farm income estimates of the U.S. Department of Agriculture (USDA). The major definitional difference between the two sets of estimates relates to corporate farms. The USDA totals include net income of corporate farms, whereas the BEA personal income series, which measures farm proprietors' net income, by definition excludes corporate farms. Additionally, BEA classifies the salaries of officers of corporate farms as part of farm wages and salaries; USDA treats the corporate officers' salaries as returns to corporate ownership and as part of the total return to farm operators. The State control totals for the BEA county estimates of farm proprietors' income are taken from the component detail of the USDA State estimates, which are modified to reflect BEA definitions and to include interfarm intrastate sales. The methods used to estimate farm proprietors' income at the county level rely heavily on data obtained from the 1974, 1978, and 1982 censuses of agriculture and on selected annual county data prepared by the State offices affiliated with the National Agricultural Statistics Service (NASS), USDA. (Data from the 1987 Census of Agriculture will be incorporated into the estimates with the next cycle of comprehensive revisions.) The NASS data, which are described by Iwig in Chapter 7 of this report, are used, wherever possible, to interpolate and extrapolate to noncensus years. In addition, data from other sources within USDA, such as the Agricultural Stabilization and Conservation Service, are used to prepare a fairly detailed income and expense statement covering all farms in the State and county. For census years, BEA prepares county estimates of 46 components of gross income and 13 categories of production expenses. For intercensal and postcensus years, the component detail of the estimates for each State is set to take advantage of the best annual county data available for the State. Farm gross income includes estimates for the following items: (1) The cash receipts from farm marketing of crops and livestock (in component detail); (2) the income from other farm-related activities, including recreational services, forest products, and custom-feeding services performed by farm operators; (3) the payments to farmers under several government payment programs; (4) the value of farm products produced and consumed on farms; (5) the gross rental value of farm dwellings; and (6) the value of the net change in the physical volume of farm inventories of crops and livestock. Cash receipts from marketing is the most important component of farm gross income. The USDA generally has annual production, marketing, and price data available for preparing the State estimates for about 150 different commodities. However, annual county estimates of cash receipts--usually for total crops and for total livestock--are currently available for only 19 States (BEA 1991, fn. 15, p. M-14). For the other States, the USDA State estimates of cash receipts from the marketing of individual commodities are summed into the 13 crop and 5 livestock groups for which value-of-sales data are reported by county in the censuses of agriculture. The aggregates for the census years are then allocated by the related census county distributions. Estimates for intercensal years are based on supplemental county estimates of annual production of selected field crops and on State season average prices available from the State NASS offices, or they are calculated by straight-line interpolation between the census years and adjusted to State USDA levels. The county estimates of the remaining components of gross income, of production expenses, of the adjustment for interfarm intrastate transactions, and of the adjustment to exclude the income of corporate farms are based mainly on data from the censuses of agriculture and data provided by NASS and by the Agricultural Stabilization and Conservation Service. See BEA (1991) for a full description of the methodology. 3.3.6.5 Personal Dividend Income, Personal Interest Income, and Rental Income of Persons These components accounted for more than 17 percent of total personal income in 1990. Dividends are payments in cash or other assets, excluding stock, by corporations organized for profit to noncorporate stockholders who are U.S. residents. Interest is the monetary and d imputed interest income of persons from all sources. Imputed interest represents the excess of income received by financial intermediaries from funds entrusted to them by persons over income disbursed by these intermediaries to persons. Part of imputed interest reflects the value of financial services rendered without charge to persons by depository institutions. The remainder is the property income held by life insurance companies and private noinsured pension funds on the account of persons; one example is the additions to Policyholder reserves held by life insurance companies. Rental income of persons consists of the monetary income of persons (except those primarily engaged in the real estate business) from the rental of real property (including mobile homes); the royalties received by persons from patents, copyrights, and rights to natural resources; and the imputed net rental income of owner-occupants of nonfarm dwellings. The State and county estimates of dividends, interest, and rent are based mainly on data tabulated from Federal individual income tax returns by the Internal Revenue Service. The methodology for dividends, interest, and rent is given in BEA (1991). 3.3.6.6 Transfer payments Transfer payments are payments to persons, generally in monetary form, for which they do not render current services. As a component of personal income, they are payments by government and business to individuals and nonprofit institutions. Nationally, transfer payments accounted for almost 15 percent of total personal income in 1990. At the county level, approximately 75 percent of total transfer, payments are estimated on the basis of directly reported data. The remaining 25 percent are estimated on the basis of indirect, but generally reliable, data. For the State and county estimates, approximately 50 subcomponents of transfer payments are independently estimated using the best data available for each subcomponent. The methodology for all of these subcomponents is given in BEA (1991); the following items are presented here as examples. Old-age, survivors, and disability insurance (OASDI) payments.--These payments, popularly known as social security, consist of the total cash benefits paid during the year, including monthly benefits paid to retired workers, dependents, and survivors and special payments to persons 72 years of age and over; lump-sum payments to survivors; and disability payments to workers and their dependents. The State estimates of each OASDI segment are based on Social Security Administration (SSA) tabulations of calendar year payments. The county estimates of total OASDI benefits are based on SSA tabulations of the amount of monthly benefits paid to those in current-payment status on December 31, by county of residence of the beneficiaries. Medical vendor payments.--These are mainly payments made through intermediaries for care provided to individuals under the federally assisted State-administered medicaid program. Payments made under the general assistance medical programs of State and local governments are also included. The county estimates are based on available payments data from the various State departments of social services. For States where no county data are available, the county estimates are based on the distribution of payments made under the aid to families with dependent children program. Aid to families with depenndent children (AFDC).--This State-administered program receives Federal matching funds to provide payments to needy families. The State estimates are based on unpublished quarterly payments data provided by the SSA. The county estimates are prepared from payments data provided by the various State departments of social services. County data are no longer being received from some State for these States, the most recent available data are used for the county estimates for each subsequent year. State unemployment compensation.--These are the cash benefits, including special benefits authorized by Federal legislation for periods of high unemployment, from State-administered unemployment insurance (UI) programs. Most States report benefits directly by county, but a few report by local district office. In the latter case, local district office data are distributed among the counties within the jurisdiction of the local district office in proportion to the annual average number of unemployed persons estimated by the Bureau of Labor Statistics (BLS). When the State is unable to supply the county data in time to meet the publication deadline, a preliminary set of estimates is made and is revised the following year to incorporate the delayed county data. The preliminary county estimates are prepared by extrapolating the preceding year's estimates forward by the change in the BLS estimate of the annual average number of unemployed persons. Veterans life insurance benefit payments.--These are the claims paid to beneficiaries and the dividends paid to policyholders from the five veterans life insurance programs administered by the Department of Veterans Affairs. The county allocations of the combined payments of death benefits and dividends are based on the distribution of the veteran population. Interest payments on guaranteed student loans.--These are the payments to commercial lending institutions on behalf of individuals who receive low-interest deferred-payment loans from these institutions to pay the expenses of higher education. The State estimates are based on Department of Education data on the number of persons enrolled in institutions of higher education. The county allocations are based on the distribution of the civilian population. 3.3.6.6 Personal Contributions For Social Insurance Personal contributions for social insurance are the contributions made by individuals under the various social insurance programs. These contributions are excluded from personal income by treating them as explicit deductions. Payments by employees and the self-employed for social security, medicare, and government employees' retirement are included in this component. Also included are the contributions that are made by persons participating in the veterans life insurance program and in the supplementary medical insurance portion of the medicare program. The State and county estimates of personal contributions for social insurance are generally based either on direct data from the administering agency or on the geographic distribution of the appropriate earnings component; see BEA (1989 and 1991) for the full methodologies. 3.3.6.7 Residence Adjustment Personal income is a "place-of-residence" measure of income, but the source data for the components that compose more than 60 percent of personal income are recorded by place of work. The adjustment of the estimates of these components to a place-of-residence basis is the subject of this section. At the national level, place of residence is an issue only for border workers (mainly those living in the United States and working in Canada or Mexico and vice versa). At the State and county levels, the issue of place of residence is more significant. Individuals commuting to work between States are a major factor where metropolitan areas extend across State boundaries--for example, the Washington, DC-MD-VA MSA. Individuals commuting between counties are a major factor in every multicounty metropolitan area and in many nonmetropolitan areas. BEA's concept of residence as it relates to personal income refers to where the income to be measured is received rather than to "usual," "permanent," or "legal" residence. It differs from the Census Bureau's concept mainly in the treatment of migrant workers. The decennial census counts many of these workers at their usual place of residence rather than where they are on April 1 when the census is taken. Except for out-of-State workers in Alaska (where migrant workers are unusually important) and for certain groups of border workers, BEA assigns the wages of migrant workers to the area in which they reside while performing the work. Similarly, BEA assigns the income of military personnel to the county in which they reside while on military assignment, not to the county in which they consider themselves to be permanent or legal residents. Thus, in the State and local area personal income series, the income of military personnel on foreign assignment is excluded because their residence is outside of the territorial limits of the United States. Three of the six major components of personal income are recorded, or are treated as if recorded, on a place-of-residence (where-received) basis. They are transfer payments; personal dividend income, personal interest income, and rental income of persons; and proprietors, income. Nonfarm proprietors' income is treated as income recorded on a place-of-residence basis because the source data for almost all of this part of proprietors' income are reported to the IRS by tax-filing address, which is usually the filer's place of residence. The source data for farm proprietors' income are recorded by place of production, which is usually in the same county as the proprietor's place of residence. The remaining three major components--wages and salaries, other labor income (OLI), and personal contributions for social insurance--are estimated, with minor exceptions, from data that are recorded by place of work (point of disbursement). The sum of these components (wages plus OLI minus contributions) is referred to as "income subject to adjustment" (ISA). Residence adjustment procedure (excluding border workers The county residence adjustment estimates for 1981 and later years are based on those for 1980 because intercounty commuting data are available only from the decennial censuses of population. (Data from the 1990 Census of Population will be introduced into the residence adjustment estimates as part of the comprehensive revisions to the State and local area personal income estimates that are now underway.) The estimation of these adjustments can be understood using the example of a two-county area comprising counties f and g. The two-county example is easily generalized to more complex situations. Click HERE for graphic. data from the 1980 Census of Population on the number of wage and salary workers (W) and on their average earnings (E) by county of work for each county of residence: Click HERE for graphic. The initial 1980 BEA estimates were modified in three situations. First, for clusters of counties identified as being closely related by commuting (mostly multicounty metropolitan areas), modifications were made to incorporate the 1979 wage and salary distribution from the 1980 Census of Population. The 1979 wage and salary distribution from the 1980 Census of Population reflects the residential distribution of the income recipients as of April 1, 1980, regardless of where they were living when they received the wages and salaries.) These modifications are needed because in numerous cases the 1980-census JTW data and the source data for the BEA wage estimates are inconsistently coded by place of work. (For example, the source data may attribute too much of the wages of a multiestablishment firm to the county of the firm's main office, or the geographic coding of the Defense Department payroll data and of the JTW data may attribute a military base extending across county boundaries to different counties.) Initial county estimates of place-of-residence wages and salaries were derived as place-of-work wages and salaries plus net residence adjustment for wages and salaries. (For the calculation of this net residence adjustment, only the gross flows for wages and salaries were used.) Then, the initial 1980 BEA place-of-residence wage and salary estimates were summed to a total for each cluster. Finally, the BEA total for each cluster was redistributed among the counties of the cluster in the same proportion as the 1979 wage and salary distribution from the 1980 census. To facilitate the extension of the 1980 residence adjustment estimates to later years, the cluster-based modifications--derived as net additions to or subtractions from the initial residence adjustment estimates for each of the 1,287 counties--were expressed as gross flows between pairs of counties within the same cluster. In the simplest case--a two-county cluster--the additional gross flow was assumed to be from the county with the negative modification to the county with the (exactly offsetting) positive modification. Second, modifications were made for selected noncluster adjacent counties if large, offsetting differences occurred between the initial 1980 BEA estimates and the census wage data for these counties. These adjacent-county modifications were expressed as gross flows in the same way and for the same reason as the cluster-based modifications. Third, modifications were made for eight Alaska county equivalents (boroughs and census areas) to reflect the large amounts of labor earnings received by seasonal workers from out of State. The 1980-census JTW data reflect the "commuting" of many of these workers, and the initial 1980 residence adjustment estimates for a majority of the county equivalents did not require modifications. However, for eight county equivalents, the initial 1979 estimates yielded BEA place-of-residence wage and salary totals that were so much higher than the comparable census data that the could not be an accurate reflection of the wages of only the permanent residents. The 1979 residence adjustment estimates, although based mainly on the 1980-census JTW data, also reflect--at the appropriate one-tenth weight--1970-census JTW data.) Based on the assumption that the excess amounts were attributable to out-of-State migrant workers, these amounts were removed by judgmentauy increasing the JTW-based gross flows to the large metropolitan counties of Washington, Oregon, and California. Click HERE for graphic. Click HERE for graphic. As a last step, the total place-of-residence ISA OSA plus net residence adjustment) for each cluster is derived and then distributed to the counties of the cluster based on 1980 place-of-residence ISA extrapolated to later years by the percentage change in the IRS-based wage series. The net residence adjustment, estimate for each cluster county is calculated as place-of-residence ISA minus place-of-work ISA. 3.4 Evaluation Practices In the past few years, two major studies were undertaken by BEA to evaluate the State and local area income estimates: (1) a reliability study of the State quarterly personal income series and (2) a study of the accuracy of the county residence adjustment estimates. In addition, In March of this year, the U.S. General Accounting Office (GAO) completed a study of BEA's national and State estimates. 3.4.1 Evaluation of the State Quarterly Personal Income Series This study provided a detailed measurement and analysis of the reliability of quarterly and annual estimates of State personal income (Brown and Stehle 1990). The study, which covered the State estimates from 1980-87, assessed the reliability of State quarterly personal income using several statistical measures to examine the size of the revisions made to the estimates. One measure used 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. Other sets of measures used were dispersion, relative dispersion, bias, and relative bias. The findings of the study were intended to: (1) help BEA isolate particular problem areas in the production of these estimates; and (2) help users of these data determine the suitability for their purposes of the estimates released at different stages of the estimating process. The four principle findings of the study were: (1) 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; (2) largely reflecting wages and salaries, the preliminary quarterly estimates of total personal income tend to be underestimated in fast-growing States and overestimated in slow-growing States; (3) beginning in 1984, the reliability of the second quarterly estimates (that is, the estimates yielded by the first routine revision) was improved by the incorporation of quarterly data from employers' payroll tax reports (the ES-202 data), and (4) the annual revisions of total personal income are smaller than the quarterly revisions. 3.4.2 Residence Adjustment Reliability Study In October 1988, a study was completed which measured the reliability of the Census commuting data used to prepare the net residence adjustments for county personal income (Zabronsky 1988). While the impact of the residence adjustments are generally small at the region and State level, the residence adjustments constitute a large portion of total personal income for most counties in the U.S. In 1989 for instance, the absolute value of the net residence adjustments accounted for about 12.5 percent of total personal income for all counties, on average, while accounting for about 25 percent of total personal income in metropolitan area counties, on average. In this residence adjustment reliability study, a comparison of a Census file of county commuting data constructed from the journey-to-work question on the 1980 decennial census was made with a file of aggregate wages and salaries independently tabulated by Census. In the course of the study, comparisons between thejourney-to-work and aggregate wage series were explored across a variety of geographic demographic, and industrial detail to develop a comprehensive reliability profile for the census commuting data. The major conclusion of this study was that taking the 1980 Census aggregate income series as a benchmark measure of county wages and salary income, the 1980 census journey-to-work data proved to be a highly reliable source for measuring commuter's income in the development of BEA's county residence adjustment estimates. Although careful analysis of the Census journey-to-work wage data did reveal a bias in that series that was correlated with county size, wage amputations undertaken by BEA largely corrected the problem while commuting patterns between counties indicated that for the relevant comparisons, the Census journey-to-work data were consistent with the Census aggregate income-based wage series. 3.4.3 GAO Study of BEA's State and National Estimates The GAO study (GAO, 1993) was conducted in response to a request by the Honorable Ernest F. Hollings, Chairman, Committee on Commerce, Science, and Transportation, U.S. Senate. Senator Hollings expressed concern about press reports that alleged that BEA "did not incorporate, for political purposes, a downward revision of original employment levels into its October 1991 estimate of first quarter 1991 State personal income growth and its December 1991 estimate of first quarter 1991 gross domestic product (GDP) growth. The report concluded that "We found no evidence that BEA manipulated first quarter 1991 personal income or GDP estimates for political purposes. BEA generally followed its standard procedures for using employment data in these estimates and deviated from these procedures only when required by what we believe were reasonable technical judgments" (GAO, 1993, p. 1) 3.5 Current Problems and Activities BEA's regular publication schedules are carefully developed to take into account the needs of users, balanced against the responsibility to produce data of high quality. In general, the four-month lag of the State quarterly and preliminary annual State personal income data and the eight-month lag in the release of more detailed annual State personal income estimates are timely enough for most purposes and cause few hardships for the users of these series. For county and metropolitan area data, the fifteen-month lag required to produce these estimates is considered too long for many purposes and has limited the usefulness of these data. In an effort to address the issue of the timeliness of its local area estimates, BEA has recently been testing the feasibility of developing preliminary annual estimates of personal income for metropolitan areas and non-metropolitan portions of States. These estimates would be available with a seven-month lag. 3.6 Conclusions Rapid advances in computer technologies continue to provide improvements in the range of regional data available for local area estimation as well as in the timing of their availability. For example, the more timely availability of ES-202 wage and salary data coupled with BEA's improved computing capabilities and estimating procedures may allow for the much more timely release of preliminary income estimates for metropolitan areas. These rapid advances in computer technologies also continue to expand the ease of data transfer, storage, and manipulation. For example, BEA recently introduced a CD-ROM containing the local area personal income estimates; many data users can now acquire the entire set of estimates rather than placing an order each time they need some of the data. As in the past, it is anticipated that these advances in electronic capabilities will continue to expand the uses and users of BEA's regional estimates. Table 3. 1: Programs Using BEA Personal Income Estimates in Allocation Formulas for Federal Domestic Assistance Funds, Fiscal Year 1992 Program Program FY 1992 Obligations Number Name (Millions of $) --------------------------------------------------------------------- 17.235 Senior Community Service 395.2 Employment Program 84.126 Rehabilitation Services 1,783.5 84.154 Public Library Construction 29.8 and Technology Enhancement 93.020 Family Support Payments to 13,814.9 States (AFDC) 93.138 Protection and Advocacy for 19.1 Mentally Ill Individuals 93.630 Developmental Disabilities 90.2 Basic Support and Advocacy 93.645 Child-Welfare Services--State 273.9 Grants 93.658 Foster Care--Title IV-E 2,342.1 93.659 Adoption Assistance 201.9 93.778 Medical Assistance Program 72,502.7 (Medicaid; Title XIX) 93.779 Health Care Financing Research 78.4 93.992 Alcohol & Drug Abuse & Mental 292.0 Health Services TOTAL 92,823.7 ---------------------------------------------------------------------- Source: Office of Management and Budget and U.S. General Services Administration (1992), 1992 Catalog of Federal Domestic Assistance, Washington, DC: U.S. Government Printing office. For information about the grant formulas, see U.S. General Services Administration (1992), 1992 Formula Report to the Congress, Washington, DC: U.S. Government Printing Office. ---------------------------------------------------------------------------- Click HERE for graphic. REFERENCES Advisory Commission on Intergovernmental Relations (ACIR) (1990), Significant Features of Fiscal Federalism, Volume 1: Budget Processes and Tax Systems, M-169, pp. 10-13, Washington, DC: U.S. Government Printing Office. Bureau of the Census, U.S. Department of Commerce (1992), Statistical Abstract of the United States: 1992, Appendix II, Washington, DC: U.S. Government Printing Office. Bureau of Economic Analysis (BEA), U.S. Department of Commerce (1985), Expermental Estimates of Gross State Product by Industry, BEA Staff Paper 42, Washington, DC: National Technical Information Service. ______(1989), State Personal Income. 1929-87, Estimates and a Statement of Sources and Methods, Washington, DC: U.S. Government Printing Office. ______(1991), Local Area Personal Income, 1984-89, Volume 1: Summary, Washington, DC: U.S. Government Printing Office. Brown, R.L. and Stehle, J.E. (1990), "Evaluation of the State Personal Income Estimates," pp. 20-29, Survey of Current Business 70 (December, 1990). Creamer, D. and Merwin, C. (1942), "State distribution of Income Payments, 1929-4l," Sur