The longitudinal research database: status and research possibilities.

McGuckin, Robert H. ; Pascoe, George A., Jr.

The Longitudinal Research Database: Status and Research Possibilities

Introduction

THE Longitudinal Research Database (LRD) is a large micro database(1) of establishment-level data constructed by pooling information from the Census of Manufactures (CM) and the Annual Survey of Manufactures (ASM). It is housed within the Census Bureau at the Center for Economic Studies (the Center), which was established in 1982 to oversee the development of this database, to use the data to improve future Census Bureau data collection and reports, and to make the data available to outside users.

The construction of the database was itself a major achievement. It contains linked data from 5 censuses and 11 annual surveys. There are 2,311,794 individual establishment year records currently in the file, and it is updated as new data become available. Thus, the LRD is one of the most ambitious and comprehensive data sets available for the study of manufacturing, and it promises to provide an exciting and stimulating research environment for many years. At the same time, the sheer magnitude of the database, coupled with its complexity, means that researchers must take the time to fully understand the structure of the database before embarking on research. This paper outlines the development of the database, its structure and current status, and the possibilities for its use in economic research.

The discussion is organized into four sections. We begin with some general observations on the characteristics that researchers desire in a database. In particular, we focus on the need for micro-level detail to adequately examine many economic issues. These observations provide the framework for the more specific remarks in the remainder of the paper. These remarks include a brief section outlining the origins of the LRD. The main portion of the paper details the major components of the LRD, the kinds of information included in the database, and the related data sets available at the Center. Throughout, we try to describe the research conducted at the Center as a way of providing concrete examples of the kinds of activity the LRD will support. We then briefly discuss access to the database and conclude with some observations intended to provide an overall assessment of the usefulness and flexibility of the LRD.

The Need For Detail in a Database

Economic analysis has a profound influence on data development. Researchers often approach particular problems with a well-defined theory, sophisticated econometric or statistical techniques, and data that are inadequate or inappropriate for testing the theory. This situation provides the incentive for developing new data. The theory provides guidance and direction to the data development strategy. Unfortunately, the need for better data often occurs when an answer to a question is required in a timeframe too short to develop a new data set. Even if there is time, the costs of developing new data are often prohibitive. In these instances, the available data influence the theory and the econometric procedures used. Thus, data development also influences economic analysis.

In most research on production functions and total factor productivity, data availability dictates the estimation procedures. The absence of detailed data for specific producing units often causes researchers to use aggregate data in econometric specifications. Several recent papers using the LRD suggest the existence of substantial aggregation bias in estimates of productivity relationships.(2) Moreover, there are many productivity-related questions that simply cannot be examined with aggregate data. John Solow (1987) argues convincingly that it is impossible to determine whether energy is a complement or substitute for other inputs using aggregate data (for example, two-digit manufacturing industries).

As an example of the need for detailed data, consider the problem of the measurement of trade flows and the technological leadership of U.S. industry. Examinations of this problem have focused on the high-tech trade balance defined in terms of trade flows measured at the three-digit industry level. This level of aggregation was chosen because high-tech industries are distinguished from low-tech industries solely on the basis of research and development (R&D) to sales ratios. Use of this procedure means that low-tech products are often included in the high-tech industry category. For instance, the office and computing equipment industry (Standard Industrial Classification 357) includes high-tech products, such as electronic computers and peripheral computing equipment. It also includes low-tech products, such as adding machines and coin counters. Conclusions based on such aggregate numbers may be misleading.(3)

These examples show that the need for more detailed data is a central feature of economic research. This need cuts across all applied fields of economics. The LRD is a longitudinal micro database that consists of individual establishment (plant) data and that provides a substantial source of detailed data.

Other elements of data structure

Elements of data structure other than the level of aggregation are also important for determining the usefulness of a data set to researchers. Such elements are the aspects of the data used to classify individual records. Although it is unlikely that any list of categories of economic data would satisfy all researchers, it is possible to list typical categories that are required for most economic research. As might be anticipated from the title of this paper, we view time as one of the most important structural characteristics. Various cross-sectional aspects of data are also regularly desired in economic research. Although for some problems the plant may be the appropriate unit for analysis, the firm or enterprise affiliation of the plant is more important for other issues. The location, industry classification, and size of the plant are other important aspects of the data structure that are of particular interest to economic researchers. Each of these variables has been made a part of the basic key structure of the LRD. As the discussion proceeds, we will highlight these structural characteristics of the LRD, but we will also emphasize that the LRD has the flexibility to accommodate research requiring new key variables.

Origins of the LRD

In the late 1970's, the Census Bureau agreed to develop a longitudinal database of individual establishments based on data collected in the CM and the ASM. The project was carried out under the direction of Richard and Nancy Ruggles of Yale University. Initial funding was provided by the National Science Foundation (NSF), the Small Business Administration, and the Census Bureau. The product of this effort was the Longitudinal Establishment Database (LED), which contains data for establishments for 1972 to 1981.

The Center was created to facilitate access to the LED file. Much of the Center's early efforts at database development were focused on a balanced panel of the LED file called the Time Series File. However, it soon became obvious that a balanced panel strategy was inappropriate. Exits due to plant closings continually reduced the number of plants in the file. Adding to the decline in the number of plants operating continuously were changes in the sample design used to collect data in noncensus years. Furthermore, analysis of the births of new plants and firms had extensive direct policy and research interest. In particular, many of the questions of interest to researchers required a focus on the firm, not simply on plants.

These factors led the Center to rethink its strategy in early 1987. All CM data for 1963, 1967, 1972, 1977, and 1982 and ASM data for 1973 to 1985 were grouped into a distributed database, which was termed the Longitudinal Research Database. The change of the database name from LED to LRD was made to emphasize the new database structure used for updating and extracting microdata; to focus attention on the primary use of the data--research and analysis; and to eliminate any confusion that may have existed, because the Time Series File and LED file had become synonymous in the inds of some people. The main consequence of this substantial undertaking is that it is now possible to generate extracts of the data using a variety of selection keys, such as geographic location, industry, size, firm, etc. Panels can be selected that meet the needs of the researcher and that are not constrained to certain years. Consequently, this paper focuses on the LRD--an unbalanced panel from which various balanced and unbalanced time series may be obtained.

Contents of the LRD

To determine if the LRD is a useful data source requires a clear understanding of what the LRD contains. The two principal components of the LRD--the CM and the ASM--are fundamentally different. We will discuss the CM first, and then we will contrast it with the ASM.

We want to alert the reader that our discussion concentrates on methodological issues that the researcher must be careful about when conducting research. Such a discussion has a tendency to emphasize problems with the data. As already noted, the LRD has been successfully employed in a wide range of studies. The results of these studies show that the LRD is a rich data source with great potential as a research tool.

The Census of Manufactures component

The CM is an enumeration of all establishments whose primary activity is manufacturing, as classified by the Census Bureau according to the Standard Industrial Classification System (SIC). An establishment is defined as an economic unit, at a single location, where business is conducted or where services or industrial operations are performed. The basic unit of data collection is the establishment, and accordingly, one of the primary data keys in the LRD is the establishment.

Since 1954, the Census Bureau has obtained the mailing lists used for data collection from the Internal Revenue Service (IRS) and the Social Security Administration (SSA). For single-establishment companies, these lists are usually sufficient for data collection purposes. However, for multiestablishment companies, the Census Bureau must request additional information, in particular, the name and address of each of the company's establishments. (An interesting byproduct of this survey is a detailed description of the firm's legal form of ownership, which we will discuss later in this article.) The information from the Census Bureau survey of multiestablishment companies is combined with the information from the IRS and the SSA to form the Standard Statistical Establishment List, which forms the basis for both the CM and the ASM.

Although the CM is a complete enumeration of all manufacturing establishments, not all establishments actually report data to the Census Bureau. Some data items for some establishments are obtained from other Government agencies, and other data items for these establishments are estimated. After the 1963 CM, it was decided to reduce the reporting burden, particularly for small companies, by making greater use of the data in the records obtained from the IRS and the SSA. Beginning in 1967, some small companies were exempted from reporting their data to the Census Bureau. Instead, census-type statistics for these establishments were developed from IRS and SSA records. The information obtained from these records includes the firm's name and address, payroll, and gross business receipts. Other statistics for these small firms are estimated using industry averages in conjunction with this administrative information.

In 1972, approximately 120,000 small single-establishment manufacturing firms identified as having less than 10 employees were designated administrative record cases and were excused from filing reports. In 1977 and 1982, approximately 145,000 and 130,000 firms, respectively, were designated administrative record cases. (See Appendix A.) The impact of administrative record data on industry aggregates is slight; for manufacturing as a whole, administrative record cases accounted for only 1.2 percent of the value added in 1972, 1.7 percent in 1977, and 1.3 percent in 1982. However, these data may be important in particular industries and for certain research topics.

The information on sales and payrolls obtained from the IRS and the SSA appears to be of high quality. Moreover, the estimation techniques for the unobserved variables work well for aggregate data. However, the methods used to estimate values for the unobserved variables in these administrative record cases may produce less useful data for microeconomic projects. Researchers must determine if the Census Bureau estimation method or some alternative is more appropriate for their projects.(4)

The treatment of the data collected from the approximately 220,000 remaining establishments reflects the demands of primary Census Bureau users and the budget constraints. The Census Bureau's primary objective for both the CM and ASM is to publish useful and accurate current year aggregates. Consequently, the data are evaluated and edited with the accuracy of the aggregate statistics in mind. Little consideration is given to the time series or microaspects of the data. In designing sampling plans and other collection procedures, the time and expense required to edit the data for an individual establishment is weighed against the probable effect that data for that particular establishment will have on the aggregates. The result is that, during editing, data for larger establishments receive more careful evaluating and editing than the data for smaller establishments.

The Annual Survey of Manufactures component

There are two major differences between the CM and the ASM: In the ASM, the number of establishments is smaller, and fewer data items are collected.

The ASM is a sample of establishments drawn from the universe of establishments in the CM. The sample is selected during the year following each census and is used for data collection for 5 years. After 5 years, a new sample is drawn from the most recent CM.

The LRD contains data from the annual surveys for 1973 to 1985. These data were collected from four separate ASM panels--the survey samples drawn originally in 1969, 1974, 1979, and 1984. Although there is substantial overlap in the establishments present in each ASM sample, the correspondence is not perfect. Details of the sampling plan are therefore important in evaluating the possibilities of using a continuous panel of establishments. Moreover, since the sampling methodology for the ASM has changed over time and since these changes have a significant effect on the time series that can be derived from the LRD, we describe them in some detail.

For the panels selected for 1969 and 1974, an establishment's size, industry, and company affiliation determined the probability of selection. If an establishment of a multiestablishment company was included in the sample, all of the company's establishments were also required to report their data, regardless of size. Thus, all firms in the ASM sample for these years were complete in the sense that all their manufacturing establishments were included.

The probability of selection for a company is related to the size of its establishments.(5) All companies with a manufacturing establishment with 250 employees or more were selected. These large companies account for more than two-thirds of total manufacturing employment in each of the censuses conducted from 1963 forward. Companies with smaller establishments were assigned probabilities proportional to their size.

In 1979, under severe budget pressure, the Census Bureau adopted a new procedure for sample selection. The main change was that the probability of selection for any establishment was now solely a function of the size of the establishment itself. Company affiliation played no part in the sample design. All establishments with 250 employees or more in the 1977 Census of Manufactures were included in the 1979 sample panel. Smaller establishments were still sampled with probabilities proportional to their size, but the plants of multiestablishment companies were not included in the sample automatically if one of the company's other plants was chosen.

The 1979 panel captures about 91 percent of the total manufacturing activity (measured by total value of shipments) captured by the previous panel, but the number of sampled individual establishments was reduced significantly--from about 75,000 to about 55,000. The major effect of the change was that many small establishments of multiestablishment companies were excluded from the ASM sample. In turn, the number of companies for which complete data were collected was also substantially reduced. Approximately 5,000 companies, roughly half of the total number of companies in the ASM for which complete data would have been available under the old sampling design, reported for only a portion of their establishments under the 1979 sampling methodology. Consequently, any time series research that requires complete information on the activities of a company will have substantially fewer observations after 1979.

To compensate for the loss of information that resulted from the 1979 change, the 1984 ASM panel now includes all establishments of companies with value of shipments of $500 million or more in 1982. As before, establishments with 250 employees or more are always included in the sample, regardless of company size, and smaller establishments are selected with probabilities that are proportional to their size.

It is important to note that the sampling design has implications for analysis conducted on the basis of categorizations of the data other than at the national level. Consider, for example, the establishment location information in the LRD. The location of each establishment is coded by state, standard metropolitan statistical area, county, and place. A sample based on these codes permits analysis below the national level. However, the selection probabilities for the ASM sample make such analysis subject to potential error. Each ASM sample provides sufficient sample points to develop estimates for national totals. But since location is not a criterion used in determining the selection probability for a particular establishment, totals derived from aggregating the microdata may not be appropriate for subnational levels of aggregation. For example, developing county or State totals in ASM years requires reweighting the data. Similarly, irrespective of the aggregations involved, the use of data from survey years requires careful consideration of the sample selection process before estimating microeconomic models. As part of the Center's software development, we plan to provide data users with methods to account for such selection biases.

Summary of CM and ASM coverage

The LRD contains data for all large establishments for every year from 1972 to 1985. These data are likely to be of high quality due to the attention they receive during collection and editing. The data for smaller establishments are less reliable, because they receive less attention during editing. However, the sales and payroll data for the administrative record establishments are not subject to substantial response error.

The ASM samples are less likely to contain small establishments because of policies to reduce reporting burdens and costs. Moreover, the composition of the sample of smaller establishments changes every 5 years. Establishments with 250 employees or more remain in the ASM panels over time. Even though the available time series of firms is less after 1979 than before, there are still over 6,000 complete multiunit companies available for annual analysis, and there are substantially more available than that for census years. Taken together, these sampling procedures imply that time series over many years will contain primarily large establishments. Finally, although the sampling procedures limit the size of continuous panels available for research, several current projects are utilizing continuous panels of over 20,000 establishments.

Data items in the CM and ASM

From every manufacturing establishment with one employee or more, the CM collects data on the establishment's inputs of labor, materials, and capital; its output of products and services; its location; and the legal form of organization of the owning firm. Associated with each establishment record is a permanent identification number and location. Both of these items stay with the establishment from its birth until it shuts down. In addition, each plant is linked to a parent firm, and detailed status codes allow one to trace ownership changes over time.

These establishment-firm codes were used to identify mergers among the largest firms in each four-digit industry for the study of conglomerate mergers by McGuckin and Andrews (1987). The same codes were used for the Lichtenberg and Siegel (1987) study of ownership changes in continuously operated plants. Lichtenberg and Siegel examined the relationship between total factor productivity growth and ownership changes using the time series panel. The McGuckin-Andrews work examined the performance of acquired lines of business in the period following their acquisition by a firm not previously operating in the same industry. This study used census year data and includes analysis of closed and opened plants. The Lichtenberg and Siegel work used yearly observations on continuously operated plants derived from the CM and the ASM.

The ASM collects the same basic measures of economic activity as the CM, and, in addition, the ASM collects detailed information on assets, capital expenditures, rental payments, supplemental labor costs, retirements and depreciation (after 1976), and in selected years, the cost of purchased services. In survey years, however, less detailed information on materials consumption and the plant's product outputs is collected. Data on individual materials consumption are not requested in survey years. Additionally, in survey years, the value of products shipped is recorded only in terms of approximately 1,500 product classes, instead of the roughly 11,000 individual products used in census years.

A detailed description of the individual data items can be found in the LED Technical Documentation (1987). A brief list of the data items gives one a good idea of the breadth of coverage. On the input side, the LRD contains the following: Total employment, number of production workers, production worker hours, salaries and wages, supplemental labor costs, cost of materials, inventory stocks for finished products, work-in-process and materials, capital expenditures, rental payments, capital stocks of buildings and equipment, depreciation, retirements, and rents and repairs. Appendix B provides the complete list.

The output data include the value of shipments reported for each seven-digit product in CM years and for each five-digit product class in ASM years. Related information--such as value added, miscellaneous receipts, value of resales, and receipts for contract work--are also available for each establishment.

There are two important points to keep in mind when designing research projects with the LRD. First, the reporting unit for data collection is the establishment. The various inputs used by the establishment are not allocated to the specific products produced by the establishment. In most applications and for most Census Bureau published tabulations, a plant is classified by the industry that accounts for the plant's largest output. As noted, detailed information on the value of shipments and physical output of products, at the seven-digit level in census years and at the five-digit level in survey years, is available for each plant. The other variables are reported at the level of the entire establishment.

Second, price data, in the form of unit values, are only collected in census years.(6) The units (quantity) are not always well defined. For example, the seven-digit level of detail does not distinguish between a $200, 10-speed bicycle and a $1,000 racing bicycle. The absence of even this information outside of census years means that price series needed, for example, for deflation in production function estimation must be obtained from non-Census Bureau sources for annual time series analysis.

This problem was recognized early on by researchers studying total factor productivity. Fortunately, the Bureau of Industrial Economics (BIE) at the U.S. Department of Commerce published an SIC-based price series based on Bureau of Labor Statistics (BLS) data. This series has been used by several researchers working with the continuous panel.(7)

We want to make one final point with regard to the price data available in census years: These unit value figures are obtained by dividing total product (or establishment) value of shipments by the quantity produced. They represent an average value for all the outputs of the establishment or product class. They may represent the combined outputs of the plant better than the BLS prices, which are based on probability samples of products. There has been little research on the relative usefulness of these alternative measures. We explicitly raise this point, because there appears to be a tendency to deemphasize unit value collection as a way to meet budget reductions, which may be very shortsighted, since it is not clear that BLS price indexes are appropriate in all cases.(8)

Although there have been a number of specific research projects using the LRD, an NSF-sponsored Resources for the Future study is developing a complete data set for research into productivity issues. Phase I of the study established the feasibility of producing a balanced panel containing detailed output, price, and input data. Preliminary analysis of the information developed for selected industries was reported at the American Economic Association annual meeting in 1987. The goal of phase II of this work is to develop a full-scale data set incorporating the methodological lessons learned in phase I. Unfortunately, budget cuts will probably prevent the completion of phase II.

Related data files

The tendency for data availability to influence the development and testing of economic models is evident in many of the research projects undertaken at the Center and described previously. To most users, the data development efforts associated with the Center's research agenda are perhaps more interesting. In this section, we highlight several projects involving extensions of the LRD that have been driven by the requirements of particular research projects. Each of these extensions involved linking the LRD to another database. Some of these efforts, like the use of BIE price index data discussed previously, involved outside databases. Other examples involved specialized Census Bureau surveys.

In an extension of their 1987 paper, McGuckin and Andrews (1988) are linking stock market premium data and other financial statistics for a small sample of companies to LRD-based performance measures for acquired lines of business (market share, profits, and productivity). This effort is an attempt to reconcile the disparate findings regarding the gains to takeovers found in the literature. Financial market studies show substantial gains that are not observed in accounting studies.(9)

One future project, which could have big payoffs, would be the development of an association between Census Bureau identification numbers and numbers used to identify companies in public financial databases. Such a step would improve research possibilities at the Center. Currently, the linking of company-level data to LRD companies in the McGuckin-Andrews study is being made by name matches. A similar procedure has been used to match companies reporting R&D data in the NSF-sponsored R&D survey to companies in the LRD. This latter procedure has resulted in several published papers about large firms.(10) Currently, with supplemental NSF support, the R&D and LRD linking is being extended to small firms. Completion of this work will mean that the entire R&D survey data will be linked to the LRD.

Supplementing the LRD by including the operations of firms outside manufacturing would be useful in research.(11) Restricting analysis of a firm to its manufacturing activities is unnecessarily limiting.

There are several areas in which the Center is working to expand the LRD's compatibility with existing Census Bureau data. One major area is foreign trade; the increasingly global nature of the economy has made it necessary to merge foreign trade data with domestic statistics. Because the foreign trade data are collected on a product basis, it is sometimes difficult to reconcile these data with LRD data collected under the SIC system. The Center is currently heading up a task force at the Census Bureau that is examining the feasibility of producing trade-adjusted concentration and market penetration statistics for detailed product classes (five- and seven-digit). The project includes CM, ASM, and Current Industrial Reports data. If the product codes and firm identifiers can be successfully linked, then these data can also be linked to the LRD. One of the first studies will examine the impact of foreign imports on domestic markets. In turn, research involving the linked data should help refine edit procedures and provide for adjustments in collection procedures when necessary.

Finally, a major long-term interest of the Center is the exploitation of individual data collected through the population censuses and surveys. The Center has at least one project that will make use of both LRD and demographic information.(12) The Center also has recently become the repository for the relatively new Survey of Characteristics of Business Owners (CBO). This survey was first conducted in 1982, and there is hope that a new panel can be developed for 1987. It is the only Census Bureau survey that directly links the characteristics of business owners with the characteristics of the business they operate. This data will greatly expand our ability to examine the nature and characteristics of entrepreneurs.

Accessing the Data

Establishment data are collected by the Census Bureau under the authority of Title 13 of the United States Code. To protect confidentiality, Title 13 and the disclosure rules and regulations of the Census Bureau prohibit the release of information that could be used to identify or closely approximate the data for an individual establishment or enterprise. In practice, the Census Bureau considers disclosure protection a binding constraint, but it provides as much public information as possible within this constraint. Although the Census Bureau has well-defined procedures for evaluating and releasing aggregate data and tabulations, it does not have similar procedures for evaluating and releasing microdata files. As a result, only a limited number of outside researchers working at the Census Bureau as special sworn employees (such as NSF and Census Bureau research fellows and associates) have access to the LRD.(13)

The practical considerations that make it impossible to accommodate all demands for microdata by allowing outside researchers to work at the Census Bureau have led to considerable interest in the development of public use data files. The major structural characteristics of a public use data file would be similar to those of the original data file so that the important economic relationships among variables in the file would be maintained. Ideally, the public use data file would preserve the economic relationships with sufficient precision so that elasticities and other parameters of interest could be directly obtained without any need for processing by the Center.(14)

In line with the public use data concept, the provision of researchers with a mock file that they could use to debug programs written in Service Annual Survey or other standard packages for execution by the Center would be a way to increase the access to the LRD. For projects involving the new and relatively clean CBO database, we hope to be able to provide complete processing without the researcher having to obtain special employee status. For LRD projects, until we have developed better software for editing the data and have had more experience with it, most researchers will still need to visit the Center to examine the data.(15) Nonetheless, with the use of programs debugged outside the Center, the necessary time required at the Center would be reduced. This means that research costs would be reduced and the Center could accommodate more LRD users.

Concluding Comment

We began our discussion by emphasizing the need for detailed microdata in resolving important issues in economic research and policy. In closing, we note that the limit on detail in the LRD is imposed by the establishment collection unit. However, within this limit, available computer technology makes it possible to classify and aggregate the data in a variety of dimensions. No longer does data collection and dissemination need to be tied to only one system. In contrast to the past, when tabulations of the data have been restricted to SIC classifications and to particular localities, the use of the data can be the determining factor in classification.

This principle has been described recently in work conducted at the Center involving the SIC system.(16) After recounting numerous complaints and shortcomings that have been voiced about the SIC system, Abbott and Andrews (1988) examined how well it classifies the data under alternative conceptual frameworks that have been proposed as a basis for the SIC system (markets, production compatibility, etc.). They find that the current system is a compromise that satisfies no particular objective. Extensions of the research to show (through the use of cluster algorithms) how the LRD data would look under various classification criteria are currently under way. But the real message that we draw from their work is that the data are sufficiently detailed and rich to support many classifications developed from objectively determined criteria. One such criterion is the grouping of producers based on the closeness of their production technologies, as judged by input proportions.(17) There are other possibilities. Regardless of the desired categorizations of the data, the Center is attempting to build into the LRD software the flexibility to organize the raw observations according to research needs. (1.) A micro database is one composed of the individual observations

collected in a survey (the establishment-level observations in the Annual Survey of Manufactures, for example). The term distinguishes such data from aggregations of the survey observations, such as employment or value of shipments for an industry. (2.) Abbott (1988) shows that the use of aggregate industry price deflators leads to biased estimates of productivity growth and production functions estimated in first differences. Lichtenberg and Siegel (1987) found that failure to account for the diversified structure of a firm's production when applying price deflators has a substantial effect on estimates of the role of technical change in total factor productivity. Similar findings are also reported by Kokkelenberg and Nguyen (1987). Finally, in a recent theoretical paper, using examples from the Census Bureau's Survey of Plant Capacity and from earlier work performed under Center sponsorship, McGuckin and Zadrozny (1988) describe several econometric problems with existing work on capacity utilization, most of which employs aggregate data. (3.) A comparison of trade balances derived from allocating aggregate industries to high-tech and low-tech categories with those derived by aggregating information on individual products separated into high-tech and low-tech categories showed substantial level and trend differences. See McGuckin and Monahan (1987) and Abbott, McGuckin, Herrick, and Norfolk. (4.) To this end, the Center is developing software that will enable a researcher to select alternative estimation strategies. (5.) In this section, we focus on the size of the reporting unit in determining its probability of selection. In practice, the sampling design is more complex, including factors such as the existence of the unit in the previous panel and industry affiliation. In the past, location may also have been included in the sample design. It is not currently a criterion variable. (6.) Current Industrial Reports data are not linked to the LRD. These reports contain yearly and sometimes monthly unit value data for many detailed SIC classifications. The Center hopes eventually to link these data to the LRD. (7.) See Lichtenberg and Siegel (1988) and Hazilla and Kopp (1986). (8.) A recent paper by Lichtenberg and Griliches (1986) discusses these differences. (9.) See, for example, the paper by Ravenscraft and Scherer (1987), which uses accounting data, and the ones by McGuckin, Warren-Boulton, and Waldstein (1988) and Guerin-Calvert, McGuckin, and Warren-Boulton (1987), both of which report premiums based on financial market data. (10.) Lichtenberg (1987) and Guerard, Bean, and Andrews (1987). (11.) This could be accomplished in part by linking LRD companies to publicly available financial data. A better procedure, which the Center hopes to undertake, would be the development of longitudinal panels for census programs conducted outside manufacturing. Such a program is already under way for the agriculture census. (12.) See Davis and Haltiwanger (1987). (13.) The Center has begun to create public use microdata files. However, precise criteria for evaluating disclosure risk in economic microdata like those found in the LRD are not yet available. Masked microdata files of demographic data have been released by the Census Bureau. These files contain samples of 100,000 individuals or more. The skewed size distribution and the relatively small number of establishments in the LRD make the development of useful, disclosures free, public use files difficult. (14.) See McGuckin and Nguyen (1988) for an extended discussion and several proposals. (15.) In some cases, for projects involving data tabulations, arrangements can be made for the Center staff to undertake the data work directly. (16.) See Abbott and Andrews (1988). (17.) This type of procedure was used by Gollop and Monahan (1986) in constructing an index of diversification. They measured the closeness of products by the technologies of pure producers.

References Abbott, Thomas A., III (1988), "Price Dispersion in U.S. Manufacturing," Center for Economic Studies Working Paper, U.S. Bureau of the Census. Abbott, Thomas A., III, and Stephen H. Andrews (1988), "An Examination of the Standard Industrial Classification of Manufacturing Activity Using the Longitudinal Research Data Base," Center for Economic Studies Working Paper, U.S. Bureau of the Census. Abbott, Thomas A., III, Robert H. McGuckin, Paul Herrick, and Leroy Norfolk, "Advanced Technology Products and the U.S. Trade Balance," Center for Economic Studies Discussion Paper. Forthcoming. Davis, Steve J., and John Haltiwanger (1987), "Establishment-Specific Labor Demand Disturbances and Unemployment in U.S. Manufacturing Industries," Research proposal to the Center for Economic Studies, U.S. Bureau of the Census. Dunne, Timothy (1988), "Firm Entry and Industry Evolution in the U.S. Manufacturing Sector: Measurement and Analysis," Center for Economic Studies Working Paper, U.S. Bureau of the Census. Dunne, Timothy, and Mark J. Roberts (1986), "Measuring Firm Entry, Growth, and Exit with Census of Manufactures Data," Mimeo, Pennsylvania State University. Dunne, Timothy, Mark J. Roberts, and Larry Samuelson (1987), "The Impact of Plant Failure on Employment Growth in the U.S. Manufacturing Sector," Mimeo, Pennsylvania State University. Gollop, Frank M., and James L. Monahan (1986), "From Homogeneity to Heterogeneity: An Index of Diversification," Center for Economic Studies Working Paper, U.S. Bureau of the Census. Griliches, Zvi (1984), "Data Problems in Econometrics," National Bureau of Economic Research Technical Paper No. 39, July 1984. Guerard, John B., Alden Bean, and Stephen H. Andrews (1987), "R&D Management and Corporate Financial Policy," Management Science 33, No. 11 (November 1987). Guerin-Calvert, Margaret E., Robert H. McGuckin, and Frederick R. Warren- Boulton (1987), "State and Federal Regulation in the Market for Corporate Control," Antitrust Bulletin 32, Spring 1987. Hazilla, Michael, and Raymond Kopp (1986), "Plant Level Productivity 1972-81: Measurement Using a Large Panel of Manufacturing Establishments," Working Paper. Kokkelenberg, Edward C., and Sang V. Nguyen (1987), "Forecasting Comparison of Three Flexible Functional Cost Forms," 1987 Proceedings of the Business and Economic Statistics Section, American Statistical Association. The LED Technical Documentation (1987), Center for Economic Studies, U.S. Bureau of the Census. Lichtenberg, Frank R. (1987), "The Effects of R&D and Fixed Investment on Productivity," Paper presented at the Allied Social Science Association meeting, December 1987. Lichtenberg, Frank R., and Zvi Griliches (1986), "Errors of Measurement in Output Deflators," National Bureau of Economic Research Working Paper Series 2000, August 1986. Lichtenberg, Frank R., and Donald Siegel (1988), "Productivity and Changes in Ownership of Manufacturing Plants," Center for Economic Studies Working Paper, U.S. Bureau of the Census. McGuckin, Robert H., and Stephen H. Andrews (1987), "The Performance of Lines of Business Purchased in Conglomerate Acquisitions," Paper presented at the American Economic Association meeting in Chicago, December 27-30, 1987. McGuckin, Robert H., and Stephen H. Andrews (1988), "Post Acquisition Performance of Acquired Lines of Business: Do Stock Market and Accounting Data Tell Different Stories." Forthcoming. McGuckin, Robert H., and James L. Monahan (1987), "High Technology Goods and the U.S. Trade Deficit," Internal Report, U.S. Bureau of the Census. McGuckin, Robert H., and Sang V. Nguyen (1988), "Use of 'Surrogate' Files to Conduct Economic Studies with Longitudinal Microdata," Paper presented at the U.S. Bureau of the Census Fourth Annual Research Conference, March 1988. McGuckin, Robert H., Frederick R. Warren-Boulton, and Peter Waldstein (1988), "Analysis of Mergers Using Stock Market Returns," Economic Analysis Group Discussion Paper No. EAG 88-1, U.S. Department of Justice, Antitrust Division. McGuckin, Robert H., and Peter Zadrozny (1987), "Long Run Expectations and Capacity," Center for Economic Studies Working Paper, U.S. Bureau of the Census. Nguyen, Sang V., and Edward C. Kokkelenberg, (1987), "The Stock of Research and Development Knowledge and Multi-Factor Productivity Growth." Paper presented at the American Economic Association meeting in Chicago, December 27-30, 1987. Ravenscraft, David J., and F.M. Scherer (1987), Mergers, Sell-Offs, and Economic Efficiency (Washington, DC: Brookings Institution). Solow, John L. (1987), "The Capital-Energy Complementarity Debate Revisited," American Economic Review 77:605-614. Nguyen, Sang V., and Edward C. Kokkelenberg, (1987), "The Stock of Research and Development Knowledge and Multi-Factor Productivity Growth." Paper presented at the American Economic Association meeting in Chicago, December 27-30, 1987. Ravenscraft, David J., and F.M. Scherer (1987), Mergers, Sell-Offs, and Economic Efficiency (Washington, DC: Brookings Institution). Solow, John L. (1987), "The Capital-Energy Complementarity Debate Revisited," American Economic Review 77:605-614.