Longitudinal Analysis Using the BLS Business Registry. Brian MacDonald and Kenneth Le Vasseur. Coolangatta (AUSTRALIA) October 14-18, 1991

Transcription

1 Index Number: Titile: Author: Longitudinal Analysis Using the BLS Business Registry Brian MacDonald and Kenneth Le Vasseur Date: Country: Round Table: United States 6th Round Table Coolangatta (AUSTRALIA) October 14-18, 1991 NOTE: The following tables have not been reproduced: Longitudinal Analysis to Six Selected States Table 1: Births and Deaths by Industry Division and Single/Multi Status Table 2: Births and Deaths on a Percentage Basis by Industry Division Table 3: Employment Changes in Continuing Units by Industry Division

2 - 2 - Session No. 4 UNITED STATES Longitudinal Analysis Using the BLS Business Registry Brian MacDonald Kenneth Le Vasseur Bureau of Labor Statistics U.S. Department of Labor INTRODUCTION 1. Over the past decade, much debate has been generated about how to measure the incidence of firm births and deaths and their impact on aggregate economic change. Such debate has presented great challenges to economists and statisticians because of the complexities of defining what these terms really mean, capturing these events on a data base, and being able to separate change resulting from economic factors from change due to administrative factors. These complexities are further compounded by the fact that change is best tracked at the level of the individual unit or establishment. 2. While firm births and deaths are the topic which seems to command the greatest attention in this area, they are merely part of a larger issue: longitudinal analysis of economic flows. Births and deaths represent the most dramatic change in the structure of business. New companies arise and others disappear; new plants and stores of growing companies are built, while facilities of declining companies are closed. Although there is a large amount of change at the margins, i.e., births and deaths, there is also considerable change within both firms and establishments which were operating in the past and continue to operate today. Some stores expand, while others downsize; some plants add another shift, while others operate at far less than capacity. 3. Most data bases provide only snapshots of the economy at various points in time. By comparing aggregate data, one can readily establish that employment increased or decreased. However, that analysis is necessarily limited. The net change, which is a very commonly used approach, masks potentially significant--and possibly offsetting--movements. If there is little or no change in the aggregate employment level--much like we are seeing today--does that mean that we are in a static state? Of course not. It is just that most data bases were not designed to examine the gross flows, but rather only the net changes. 4. As indicated in past status reports and the 1989 paper entitled "Nonsampling Uses of the United States Bureau of Labor Statistics Registry," one of the planned outcomes of our registry redesign has been to develop capabilities for longitudinal analysis. Previously, the continuity of BLS data was impaired by the substantial impact of administrative procedures as defined by Unemployment Insurance (UI) tax laws.

3 - 3 - When businesses changed legal form or ownership, those changes were often captured on the data base as closures of the original entities and openings of new ones. The new registry eliminates much of the effect of such administrative factors. 5. With the implementation of the redesigned system and the improvement in the quality and quantity of data stored on the registry, BLS has begun to make progress in producing longitudinal data. The most obvious of these is the collection and storing of data for every quarter. Other important developments include the disaggregation of data to the establishment level and the implementation of a record linkage module to identify continuous records and reduce the number of false births and deaths. With these two components in place, BLS has been exploring the feasibility of producing summaries of business births, deaths, and continuous units by industry. These data will attempt to provide a new, dynamic view of the U.S. economy. 6. The purpose of this paper is to discuss some of the initial phases of this work and to present some preliminary results in analyzing employment data on a flow basis. More formalized, modified procedures will be implemented and more thorough analysis will be performed when data are more uniformly available at the establishment level. PROCESSING 7. The processing to extract longitudinal data consists of two major steps. The first consists of examining all employer records for the two time periods (quarters) of interest, determining whether or not they are continuous, and assigning an indicator to note that continuity. The second step consists of examining records, performing some basic calculations, and displaying aggregates based on those calculations and a particular classification scheme. These steps are explained in detail in this section. Record Linkage 8. The continuous unit matching process consists of several stages of record linkage to maintain continuous identification of records and reduce the number of false births and deaths that might otherwise be indicated by the UI system. 1 This module links records possessing essentially the same characteristics over time but which may have undergone changes in identification. The underlying consideration is that, despite a change in identification of a record, it should still be considered the same unit. 9. First, the input files (Quarterly Unemployment Insurance Name and Address Files) from the States for the current quarter are reviewed, edited, and corrected as needed. The updated files are matched to corresponding files from the previous quarter (actually four previous quarters of data from the registry, itself) by the UI account and reporting unit number.

4 - 4 - The remaining, unmatched records from the current quarter are then matched by predecessor identification to the remaining records on the current or previous quarter file to account for successions occurring within the quarter or between quarters. Records which match on either the basic UI account and reporting unit number or on predecessor identification have essentially been deemed continuous by the State. Consequently, they will also be considered continuous on the BLS registry. 10. Theoretically, predecessor identification should be reported by the States on all records which have undergone a transfer of ownership. However, in some instances, the State may not consider the new owner to meet the legal UI definition of successorship for the predecessor account and, therefore, not report a predecessor UI number. Other transfers may be missed by the State due to the timing of the change. To account for non-reporting of predecessor information, BLS processes records which do not have matching identification information through a record linkage procedure which attempts to establish a match based on the records in question having a combination of critical data elements in common. 11. Previously unmatched records are processed through several iterations of "weighted" matching. Each iteration uses a different group of elements for blocking. The data elements used to block records in this process are considered key ingredients for defining continuity. The data elements (e.g, name, address, telephone number, industry, county, etc.) on records which match during the blocking are then compared for each record pair. Match weight points are assigned for each data element that matches, with the weights varying by data element. If the number of match points exceeds a set threshold, the two records will be considered a match. If one record matches more than one other record, the match with the highest number of points over the threshold is considered the correct one. Once records are considered to be a match, they are removed from further processing, thereby reducing the number of unmatched (non-continuous) records. 12. Current quarter records which match on either UI and reporting unit number, predecessor identification, or a critical combination of data elements are assigned a data base identification number which corresponds to a previous quarter record. Non-continuous records are assigned a new data base identification number, and previous quarter records for which no match has been found on the current quarter will be considered inactive but will remain on the (four quarter) registry for three additional quarters. This is the point in routine processing at which data are loaded to the registry. Classification and Tabulation 13. The second major step begins with deleting master records (representing UI account totals) and government records from the data base "load" files described above. Current quarter data are matched to data from the earlier quarter's corresponding load file by data base identification number.

5 - 5 - This allows us to take advantage of the predecessor identification and mechanical matching process described above. Thus, we eliminate many of the false births and deaths which would result from direct use of the State input tapes (i.e, matching on UI and reporting unit number only). 14. Unmatched records from the previous quarter are classified as deaths and unmatched records from the current quarter are considered births. Records which appear in both quarters are classified as continuous. 15. Births and deaths are classified by single unit or multi-unit status, depending on whether they are part of a UI account with one, or more than one, establishment in the latter period. The employment for the third month of each quarter is summed by industry, and a unit count is generated. Detailed tables provide this information at all levels of industrial detail from the 4-digit industry to major industry division. Net changes, i.e., births minus deaths, in both units and employment are also calculated. Birth and death rates can also be calculated relative to the base period. 16. Continuous units are also classified by single or multi status and at the same level of industrial detail. They are further classified by (1) decreasing employment, (2) increasing employment, and (3) constant employment for the purpose of computing employment flows. "Declining" units are simply those which show an employment decline of any magnitude from the third month of the earlier quarter to the third month of the current quarter. "Growth" units are those which grow at a rate exceeding a chosen parameter. Units with identical employment or employment increases of less than the parameter are considered "static" or constant. The net employment change for the aggregate of these categories is also calculated. SCOPE OF STUDY 17. For this particular study, we used data for six States: Alabama, Colorado, Idaho, Kansas, Minnesota, and New Mexico. We wanted the scope to be small enough to be able to track individual units of significance if necessary, yet large enough so that the data might be somewhat representative of the entire nation. A four quarter span was chosen to provide an adequate number of births and deaths. The specific States were chosen to minimize any potential impact of changes in employer reporting. All six were reporting consistently over the period of the study, and most were reporting at the establishment level. (A noticeable increase in establishment-level reporting eliminated several States from consideration.) 18. The time period chosen was the most recent span (fourth quarter 1989 to fourth quarter 1990) for which data were available. The employment parameter used to define growth units was 4.2%, based on a factor of 1.5 applied to the annual growth rate (2.8%) for the period when our software was originally developed. As noted above, the scope was restricted to private sector units.

6 FINDINGS Tables 1, 2, and 3 present some aggregate results. Table 1 contains unit counts and employment levels for births and deaths over the four-quarter span by industry division and single or multi unit status. Table 2 contains birth and death data as a percentage of the fourth quarter 1989 base period by industry division. Table 3 shows the employment levels of continuous units for growing, declining, and "static" units, again by industry division. 20. Our analysis of these, and more detailed data not presented here, thus far has been limited due to both time constraints and the artificial nature of our six State aggregate. Nevertheless, some interesting observations can be made. One obvious result is the large number of both births and deaths, especially relative to the employment base of 5.8 million. Over a four-quarter span, 13% of the units have disappeared and been replaced by an even larger number of new units. The death rate appears to be about 3% per quarter, and the birth rate nearly 4%. The employment loss of 389,000 due to deaths was more than offset by a gain of 507,000 as a result of births. The net employment growth of 118,000, representing 85% of the total employment growth, far exceeds the growth within continuous units. 21. The net employment change due to births and deaths is positive for all industry divisions, and the net unit change is positive for all but one. Single unit employers, which make up about 85% of continuous units, account for 82% of births and 84% of deaths but only 54% of the birth employment gain and 67% of the death employment loss. Birth rates are relatively similar across industry divisions; death rates are a bit more varied. The data show a relatively high rate of both births and deaths in construction, and relatively low rates in manufacturing. Although the death rate is not substantially higher than average, the percentage of employment lost due to deaths in finance, primarily commercial and savings banks, and insurance is quite high. 22. Looking at the incidence of births and deaths by quarter, births occurred rather evenly. Of the four quarters, the most births (28%) occurred in first quarter and the least (23%) in fourth quarter. Deaths occurred more disproportionately (33%) in first quarter, with between 21% and 24% occurring in the other quarters. 23. Regrettably we did not tabulate the employment change within each of the growing, declining, and static groups of continuous units. The net of the three groups, however, pales in comparison with the net change due to births and deaths. As noted, less than 15% of the net employment growth of 138,000 was due to expansion of existing units. Although not shown in the tables presented here, the net employment growth of 20,000 resulted from a net decline in multi units of 18,000 and a net increase in single units of 38,000. The net employment growth in continuous units translates into an average change of just over 0.05 persons per unit.

7 FUTURE WORK As the quality of our data improves even further, and establishment-level reporting becomes more universal, more extensive economic analysis can be performed. The impact of administrative factors on the continuity of our data is continually diminishing. Following a major shift to the establishment level in first quarter 1991, tracking units will become even easier and the results more reliable. As reporting practices stabilize, the number of States studied will increase. At the national level, more detailed industry data can be studied. Various levels of geographic aggregation can be created. Data for the U.S., individual States, and regions, such as New England, can be studied. 25. As indicated in the BLS status report, 2 work is proceeding on our firm linkage system. Once that is completed, we will be able to look at growth within firms, not just units, and decompose that change into the net of births and deaths and change in existing units. We also anticipate conducting analysis by size of firm and size of unit. 26. We also expect to make some refinements to our software, most notably to add a calculation of the employment change within each of the growing, declining, and static groups. We expect to look at the monthly employment data, particularly for births and deaths, to gauge the impact of using data for the third month rather than the quarterly average, first nonzero month, or last nonzero month. 1 See past status reports, particularly "Progress Report on Improvements to the U.S. Bureau of Labor Statistics Business Survey Frame," presented at the Fifth International Roundtable on Business Survey Frames, Williamsburg, Virginia. 2 See "Progress Report: BLS Business Registry," prepared for the Sixth International Roundtable on Business Survey Frames, Coolangatta, Australia. For more detail on firm linkage, see "Exact Matching for Constructing BLS' Enterprise Register," prepared by Brian MacDonald and Catherine Armington for the Fifth International Roundtable on Business Survey Frames, Williamsburg, Virginia.