Analysis and Recommendations

Analysis of the selected examples described above does not identify a single unified thread, except that they all represent issues which in one way or another came to the attention of Office of Federal Statistical Policy and Standards (OFSPS) or its predecessor, the Statistical Policy Division of OMB. Many other coordinated efforts, which do not come to the attention of OFSPS, go on constantly within and between departments. Unfortunately, it is equally true that many opportunities for coordination are missed for various reasons. Usually, the reason for the failure to coordinate involves a lack of knowledge on the part of the various actors, although there are sometimes more selfish, bureaucratic motivations involved.

Therefore, that the most important issue concerning the coordination of survey activities involves provisions for informing all possible interested parties that specific issues are to be explored or a particular survey is to be conducted.

Another element which was present in those cases where proper coordination did take place was the existence of a group with the authority to require coordination between the involved agencies. Problems within the same department can frequently be solved at the Secretarial level, if the department is sensitive to the issue of data coordination, and if the department has established a mechanism to inform the constituent parts about what is taking place within the department.

In some very large departments, notably DHEW, coordinating bodies have to be developed at lower levels such as the Education Data Acquisition Committee the Education Division. Portions of HEW contain almost as many constituent parts as some major departments. The existence of coordinating bodies at the subdepartmental level, however, does not eliminate the need for department level coordination.

Finally, there remains the need to have interdepartmental coordination. This role is principally the responsibility of the OFSPS.

Since much of the collection of Federal statistics is carried out through interagency agreements with a few major statistical agencies, notably the Bureau of the Census, the collecting agencies have a unique opportunity to help coordinate Federal surveys.

Two other mechanisms outside of the statistical system also have a major impact on the coordination or lack of coordination on Federal surveys. One is the budget process, and the other, the legislative process. The budget process at the departmental level, the

OMB level and the congessional level provides a mechanism for identifying overlap and forcing agencies to correct situations as needed. Through the budget mechanism, funds can be awarded to one agency and deleted from another agency's request. Furthermore, the "allowance letter" (which provides the agencies with their budget "mark" as well as instructions for using their appropriated funds) can be a helpful tool.

Complaints about lack of coordination are increasingly being made by segments of the public, especially industry groups, education and health groups and others. Congress has reacted by enacting legislation to control the situation. Some of this legislation directs specific Federal agencies to coordinate the data collection for their entire functional area.

In order for all agencies to be aware of proposed survey activities, a notice of intention should be submitted to OFSPS for any continuing survey including all longitudinal surveys and for any one time survey which proposes to interview 20,000 or more respondents. This notice of intention would permit OFSPS to review the concept. A short summary would also be published in the Statistical Reporter and thus be made available to all agencies and other interested parties to determine if there were other interests which need to be served.

A Strategy for the 1980's

During the past decade, the pressure for small area data has been building continuously. Many agencies recognize the needs of State and local areas for data for program planning, monitoring and evaluation. Recently, President Carter's new urban initiative. stressed the need for the development of an urban data base.

Several Federal agencies have attempted to meet the need for small area data either by expansion of national samples or by carrying out local area surveys in conjunction with national surveys.

The Current Population Survey has undergone constant sample expansion since its inception. Earlier increases were instituted primarily to increase the reliability of estimates derived from the survey and the need to examine problems of smaller segments of the population. Recent and proposed increases, on the other hand, have been for the purpose of providing data for states and other sub-national

areas.

In 1972, The Law Enforcement Assistance Administration established the National Crime

Survey. A part of this survey involved the collection of crime data from the largest cities. The rationale for this was that crime was a greater problem in the central cities than in other places. While this program was continued for a few years, it was finally discontinued in part because of the expense and in part because of the recommendation of a National Academy of Science panel that city collection should be discontinued in favor of modifying the sample to obtain SMSA data.

For the past several years, the National Center for Health Statistics has requested funds to expand the Health Interview Survey to provide data at the State level. These requests have not been approved, but the Center feels that such data would prove useful.

The Annual Housing Survey, as noted previously, collects data for 59 SMSA's on a rotating 3-year basis, as well as national data. These data are used for planning purposes by local governments and, as with all local data, provide an analytical tool for comparing different kinds of urban areas.

There is a substantial issue concerning the responsibility of the Federal Government for providing data for State and local planning. Granting the usefulness of the data, many argue that having the Federal Government collect it is a hidden form of revenue sharing and that it should be the responsibility of subnational governmental entities to pay for it.

A related issue is the increased concern for the imposition of Federal reporting burden on the public. This concern caused the OMB in 1976 to declare that Federal agencies could not collect data for subnational areas in surveys that are conducted annually or more frequently. This restriction did not affect the largest of the Federal surveys, the CPS, because the Comprehensive Employment and Training Act (CETA) requires that the survey provide data at the State level. However, the principle remains; the Federal Government should limit its intrusion into the lives of its citizens.

Yet the need for local area data remains, for the Federal Government, State and local governments and for the private sector. The problem then relates to how to develop data for small areas without imposing an excessive burden on the public or an excessive cost on the taxpayer.

The problem would be solved if there was a magic formula to provide small area estimates. Given the unavailability of true magic, statistical estimation may have to suffice. A great deal of work has already been done on the development of synthetic methods of developing small area estimates in some areas.

For some years, the BLS has used its "70-step" method for estimating local area unemployment. At the State level these estimates are benchmarked to the CPS. No one would say that this method generates a precise count of the unemployed, but it does provide a consistent estimate from place to place; which, while not perfect, is effective.

Other agencies should follow this practice. The development of regression, or other analytic methods, for generating local area estimates, based on survey or administrative data, is essential. The large national surveys should continue in order to provide benchmarks, but given the concerns of cost and burden, sample expansion in order to provide small area data should be at a bare minimum consistent with direct legal mandates.

The decennial and mid-decade censuses should provide the basic socioeconomic and demographic data for small areas which would be used to develop the small area estimators. Current planning calls for the mid-decade census to be organized in such a way as to maximize the amount of information which may become available while keeping burden at a minimum. The procedure being proposed is the use of "nested samples." (This process is described in the chapter on statistical methodology.) The strategic purpose for using the nested sample technique is to provide the specific data items needed to drive small area estimation models. However, at present no one knows what data are needed. A major effort will have to be carried out by the various Federal agencies to identify the correlates and develop questions to identify them. While we recognize that it may not be possible to develop estimators for all of the needed social and demographic measures, every effort still must be made.

A good faith effort must be made by the various collection centers to develop the methodology for making local area estimates in their area of concern. There should be no funding for sample expansion until such serious efforts have been made. Each agency which maintains a continuing survey should prepare by April 1, 1979 a research plan for the development of subnational estimates, if they feel that subnational data are needed in their area of concern. If such a plan is not developed, there should be no future consideration of funding to expand existing surveys to provide subnational data.

In order for this strategy to work, there will have to be strong interagency coordination of the content of both the decennial and mid-decade censuses. In order to accomplish this coordination, a permanent extension of the Federal Agency Council on the 1980 Census was agreed to by the Statistical Policy Coordination Committeee (SPCC). This will permit all

agencies to present their case for the collection of the specific data items needed to derive their estimation models.

In the past, the Federal Agency Councils, like most other interest groups, have presented their needs in terms of intuitive formulations of questions or merely in terms of a need. Beginning with the 1985 middecade census, Federal agencies will be expected to offer only tested inquires. The Census Bureau's testing responsibility should be limited to examining feasibility of the overall package, its cost and "order effects."

Until now the discussion has dealt with problems identified by various Federal agencies. These agencies following their own specific concerns not only could skew the data requirements but could omit im

portant issues which need to be addressed. To reduce this possibility, a series of statistical policy development seminars should be conducted. These seminars should be sponsored by the Statistical Policy Coordination Committee. State and local governments, the academic community, various interest groups as well as the general public would be involved in the seminars. The data requirements developed by the Census Bureau and other Federal agencies would be subjected to broad public scrutiny by these seminars.

This process should develop a process which will both reduce burden and provide the data needed for Federal, State and local planners as well as the private and academic sectors.

Introduction

Substantial concern has been voiced in the statistical community about maintaining and, where necessary, upgrading statistical expertise throughout the Federal Government. More than a decade ago, the American Statistical Association noted that the growing need for statistical data had not been accompanied by a corresponding increase in personnel trained in data collection, processing, and analysis. More recently, the International Statistical Institute appointed a Committee on the Integration of Statistics to make recommendations concerning the gap between theoreticians and practitioners in the statistical field. A seminar, "On the Transfer of Methodology Between Government and Academic Statisticians" has recently been conducted by the National Science Foundation and the American Statistical Association to examine a series of issues, including academic training needed for government statistical work, exchange arrangements for government and academic statisticians, in-service training, and cooperative arrangements between governmental and private (including quasigovernmental) statistical organizations.

This chapter of the Framework describes current and past programs to fully develop the potential of the professionals in the Federal statistical establishment so as to maximize efficiency and productivity in the entire Federal Statistical System. It further proposes methods for maintaining and upgrading statistical expertise.

In addition, this chapter contains information on the number and distribution of statistical personnel in the Federal Government.' Statistical personnel are dispersed throughout the Government. A great percentage, as can be expected, are located in the 38 "major" statistical agencies which are identified in the organization chapter of this Framework, and in a subset of these 38 which is composed of the five

'As used in this document, "statistical personnel" include those classified by the U.S. Civil Service Commission as professional statisticians, mathematical statisticians, economists, operations research employees, actuaries, and social science analysts, as well as the paraprofessional statistical assistants.

Identification of the Programs

Various kinds of professional staff training programs have been in operation in the major statistical centers during the past several years. The most extensive program started in 1963 in the National Center for Health Statistics (NCHS). This one-year work-study program, known as the Junior Professional Training Program, prepared entry level (GS-5 or 7) professionals to be professional statisticians. The trainees were employed immediately following college or graduate level education. They were rotated to three major substantive assignments in three different divisions of NCHS as well as to short assignments in the administrative division and the computer division. In addition, they participated in weekly seminars ranging from the organization and subject matter of NCHS and its divisions and programs to topics relating to statistical methodology. They also participated in a 1-week writing course as well as a 2week report writing and analysis course. There were approximately 10 trainees in each class. Financially supported after hours academic training was available, but optional. At the end of the year, the trainee was placed in a divisional assignment after consideration of the trainee's preference and the

'These agencies are the Bureau of the Census, the Bureau of Labor Statistics, the National Center for Education Statistics, the National Center for Health Statistics, and the Statistical Reporting Service.

needs of the NCHS divisions. The final decision was made with the concurrence of the affected division director and the senior staff of NCHS. The program was discontinued in 1974 because greater numbers of more experienced statisticians became available and, at the same time, NCHS lost its direct hiring authority for this program.

In addition to this structured program, NCHS has had a number of supplementary training programs. A number of agencies within the Department of Health, Education, and Welfare, including NCHS, have provided support for an annual 6-week summer session in statistics at one of the leading departments of statistics. The 1977 session, the 19th in the series which began in 1958, was held for a second year at Harvard University. Other universities at which the program has been given include Michigan, Minnesota, Stanford, North Carolina, Yale, Pittsburgh, Washington, California/Berkeley, Vanderbilt, and Texas/Houston. Although these sessions are designed for all those interested in statistics, primarily biostatistics, NCHS has taken advantage of the programs by sending 13 employees to them over the last 5 years. Other supplemental activities include special workshops and training sessions on the use of statistical software packages.

NCHS also encourages employees to take courses in statistics at local universities, usually at agency expense. In addition, it has had a program similar to the long-term training program aSRS, which is described below. Over the past 10 years, 21 employees have been supported by NCHS for 1 year of training at universities such as North Carolina, Georgetown, Oklahoma, North Carolina State, and Texas/Houston.

To provide training at the State and local level, NCHS established a series of short courses, usually lasting for a period of a week, under what is called the Applied Statistics Training Institute (ASTI). These were initially administered by NCHS staff located at its installation in the Research Triangle Park, North Carolina, and are now operated under the aegis of the Cooperative Health Statistics System. Experts both within and outside the Federal Government are recruited as teachers for the sessions. More recently, most of the administrative functions have been performed under contract.

The professional training program most similar to that of the NCHS was in the Bureau of the Census in the Department of Commerce. However, the Census Bureau program, known as the Statistical Intern Program, differed in some important ways. First, the Census Bureau program was for statistical personnel already in the employ of the Census Bureau. GS-7-11

employees (versus GS-5 and 7's in NCHS) with one or more years of Federal experience were eligible to apply. Applicants were required to be classified as statisticians, mathematical statisticians, economists, or social science analysts. The Census Bureau bulletin which described the program stated that it was "designed to identify junior level employees who display potential for assuming broader responsibilities and to make available to them developmental opportunities which they may not otherwise obtain in their present positions." Selection included a review of education and work experience, supervisory evaluation, a written qualifying examination, and an interview by the selection committee. A maximum of six professionals were selected each year. The program was in existence from 1957 to 1972, when the demand for it ceased. The Census Bureau was able to recruit and keep qualified junior professionals, and funds for the 1970 decennial census, to which this program's funding had been tied, ended in 1971.

The second major difference between the Census Bureau and NCHS programs was the content of the programs. The NCHS program was structured with weekly seminars. The Census Bureau program included three or four rotational assignments with an option to have an assignment outside of the Census Bureau, in the Commerce Department or elsewhere in the Federal Government. In addition, the Census Bureau more forcefully encouraged after hours academic training, and agreed to financially underwrite one course per semester (spring, summer, and fall). Although the Statistical Intern Program has been terminated, the Bureau continues to support after hours academic training, underwriting the cost of one course per semester for employees. At the present time, approximately 500 Bureau employees register for these courses each year.

Another major professional staff training program has developed in the Bureau of Labor Statistics since 1972 when a Division of Training was established separate from the Division of Personnel. A major thrust of the program is the development and delivery of program-related training courses for economists in the regional offices who collect data for a large number of surveys (wages and compensation, export and import prices, and construction labor and material costs) and work with State staffs engaged in Federal-State-local cooperative and grant programs (primarily in the manpower and occupational safety and health areas). These courses range from 2 or 3 days to 2 weeks, include initial and refresher training,

'U.S. Bureau of the Census. Census Bulletin Vol. XV, No. 23 (November 5, 1965).