The Longitudinal Business Database (Jarmin and Miranda, Center for Economic Studies, 2002)

Summary

In this July 16, 2002 paper, Ron S. Jarmin and Javier Miranda of the Center for Economic Studies at the U.S. Census Bureau describe the Longitudinal Business Database (LBD). They explain that the Census Bureau, the largest federal statistical agency, collects data on businesses through its economic censuses and surveys, and that the LBD links these records over time. Unlike the earlier LRD, which covers only manufacturing, the LBD covers nearly all of the non-farm private economy and some public sector activities. The authors say they supplemented the Census Bureau's longitudinal identifiers with name and address matching to repair broken linkages that had produced spurious establishment births and deaths. The paper carries a disclaimer that its views are the authors' own.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Issuer: Source documents
Document type: Covid Corporate Financing Facility Ccff Information For Those Seeking To Participate In Th
Case: Covid Corporate Financing Facility Ccff Information For Those Seeking To Participate In The Scheme

Cite this document

“The Longitudinal Business Database (Jarmin and Miranda, Center for Economic Studies, 2002).” Original at www2.census.gov. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/covid-corporate-financing-facility-ccff-information-for-those-seeking-to-participate-in-the-scheme/, SHA-256 f5aa9958f7dc3beb624e72f0c68c2a0490f7a79d1ba6e3092198af386de8721b.

Full text

The Longitudinal
Business Database

July 16, 2002

Ron S. Jarmin1
Center for Economic Studies, U.S. Census Bureau
rjarmin@ces.census.gov

And

Javier Miranda
American University and
Center for Economic Studies, U.S. Census Bureau
jmiranda@ces.census.gov

1
Disclaimer: This paper reports the results of research and analysis undertaken by Census
Bureau staff. It has undergone a more limited review by the Census Bureau than its official
publications. This report is released to inform interested parties and to encourage discussion.
Any findings, conclusions or opinions are those of the authors. They do not necessarily reflect
those of the Center for Economic Studies or the U.S. Census Bureau.

We would like to thank Cathy Buffington, Lucia Foster, Shawn Klimek, C.J. Krizan, James
Monahan and Al Nucci, each of whom provided crucial assistance and guidance on this project.
We also wish to thank Randy Becker, John Haltiwanger, Paul Hanczaryk, Brad Jensen and
Richard Moore and seminar participants at Census, CAED 2001 and BLS for helpful comments.
I. Introduction

As the largest federal statistical agency and primary collector of data on
businesses, households and individuals, the Census Bureau each year conducts
numerous surveys intended to provide statistics on a wide range of topics about
the population and economy of the United States. The Census Bureau’s
decennial population and quinquennial economic censuses are unique, providing
information on all U.S. households and business establishments, respectively.

The censuses and most surveys are static snapshots of the populations they are
intended to describe. Researchers and statisticians, however, have long known
the value of longitudinal data that follow the same survey units over time (see
McGuckin and Pascoe, 1988). There are two ways to construct such datasets.
First, statistical agencies can explicitly design longitudinal surveys (examples
include the SIPP, NLSY and PSID). Alternatively, records from successive
survey years can be linked together. For the vast majority of the statistical
programs at the Census Bureau, the latter is the only alternative.

The Center for Economic Studies (CES) at the Census Bureau has a mandate to
construct, maintain and conduct research with longitudinal datasets. CES
maintains these and other micro datasets for use by economists, and other social
scientists. These data sets primarily contain information received from
respondents to Census Bureau censuses and surveys, and may contain data on
individuals, households or businesses. Traditionally, however, CES has focused
on business data, mostly from the manufacturing sector.

The first longitudinal dataset created at CES was the Longitudinal Research
Database (LRD)2. The LRD contains longitudinally linked plant level data from
the Censuses and Annual Surveys of Manufactures. Over the years, a large and
successful research program has been carried out using the LRD (for reviews
see Bartelsman and Doms 2000; Caves 1998).

This paper describes recent efforts at CES to create a new longitudinal research
dataset: the Longitudinal Business Database (LBD). The LBD is a major
improvement over existing longitudinal establishment datasets. Unlike the LRD,
which covers only manufacturing, the LBD covers nearly all the non-farm private
economy, as well as some public sector activities. Also, research using the LRD
found problems with broken longitudinal linkages that lead to spurious
establishment births and deaths (Dunne 1992). We supplemented the
longitudinal numeric identifiers assigned by the Census Bureau with name and
address matching to repair broken linkages. Other economy wide longitudinal
files, such as the Bureau of Labor Statistics’ Longitudinal Database (LDB)
(Spletzer 1997; Pivetz, Searson and Spletzer 2001) and the Small Business

2
The LRD actually evolved from the Longitudinal Establishment Database (LED) that was
constructed in the early 1980’s. Attempts to longitudinally link plant level data from the Annual
Survey of Manufactures were initiated as early as the late 1950’s (Monahan 1992).

2
Administration sponsored Business Information Tracking Series (BITS) at the
Census Bureau (Robb 1999) only extend back into the early 1990’s. The LBD,
by contrast, contains data back to 1975.

There has been talk of creating the LBD for several years (Nucci 1993), but due
to various constraints, serious work was delayed. The LBD provides
longitudinally linked data for all employer (i.e., those with paid employees)
establishments contained in the Census Bureau’s business register, the
Standard Statistical Establishment List (SSEL). Currently, the core linkage files
of the LBD have been constructed and work continues to add additional
components that will contain basic data items, such as payroll, employment,
location, industrial activity and firm affiliation.

The LBD will be invaluable to researchers examining entry and exit, gross job
flows and changes in the structure of the U.S. economy. The LBD will also be
useful to aid the Census Bureau in examining how its census and survey
programs describe the U.S. economy. While the LBD is useful as a stand-alone
research dataset, we intend it to be used in conjunction with other Census
Bureau establishment and firm level micro data. The LBD describes how
establishment units in various Census Bureau censuses and surveys are linked
over time. Identifiers on the LBD facilitate linking to other datasets.

In this paper, we will focus on describing the longitudinal linkages that are the
core of the LBD. We first discuss the source data we used. Next we describe
how we linked establishment records over time. We then discuss some of the
features of the LBD and how we think it can be useful to the Census Bureau and
the research community.

II. Source Data for the LBD

Various components of the LBD will contain information from a variety of sources
including the Business Register (or SSEL), Economic Censuses and surveys. To
construct the longitudinal linkages, however, we relied on only one source: the
business register.

Since 1972, the Census Bureau has maintained a general-purpose business
register for use by the Federal Statistical System. The business register is a
database of U.S. business establishments and companies. In 1968 the Office of
Management and Budget, which oversees all Federal Executive Branch
statistical activities, directed the Census Bureau to develop and maintain the
Standard Statistical Establishment List (SSEL) on behalf of all federal statistical
agencies. The SSEL is authorized under Title 13, U.S.C. (for more see U.S.
Bureau of the Census, 1979).

The SSEL is used as the frame for Census Bureau firm and establishment
surveys. It is also the source data for employment and payroll data summarized

3
by industry and geographic area in the County Business Patterns (CBP)
Program. Longitudinal business demographics and summary statistics derived
from the SSEL have been used by multiple government agencies and private
organizations.3 The SSEL is continuously updated with administrative data from
other federal agencies, as well as data collected by the Census Bureau. It has
undergone significant changes and enhancements over time. For example, the
SSEL has seen two major revisions in industry coding: the 1987 SIC revision,
and the adoption of NAICS in 1997. The Census Bureau is currently undertaking
a major redesign of the SSEL for the 2002 Economic Census. While this
redesign will have important implications for the LBD going forward, we do not
discuss it here. Instead, we focus on the characteristics of the SSEL over the
period of the current LBD.

Coverage of the SSEL4:

The SSEL covers legally operating entities operating in the U.S. and its
territories. Entities engaging in illegal or “underground” activities are not covered.
The SSEL covers only entities with paid employees. Before 1994, nonemployers
that were subject to Federal income tax were included in the SSEL in Economic
Census years. Since 1994, these are kept in a separate nonemployer file. Note
that the SSEL files stored at CES contain employer entities only.

Industry coverage in the SSEL depends on whether entities are privately or
government owned or controlled. For private entities, the SSEL covers all
industries except private households. However, entities in industries outside the
scope of the Economic Census are not broken into establishment units. The
industrial scope of the Economic Census has changed over time5. Currently, out
of scope industries include: Agriculture, Forestry and Fishing (SIC Division A),
railroads (SIC 40), U.S. Postal Service (SIC 43), Certificated Passenger Air
Carriers (part of SIC 4512), Elementary and Secondary Schools (SIC 821),
Colleges and Universities (SIC 822), Labor Organizations (SIC 863), Political
Organizations (SIC 865), Religious Organizations (SIC 866) and Public
Administration (SIC Division J). Most government owned or operated entities
(SIC Division J) are outside the scope of the Economic Census and their
establishments are not broken out on the SSEL (these entities are, however, in
scope for the Census of Governments and are included on the Government
Integrated Directory). The only exceptions are Wholesale Distributors of Beer,
Wine and Distilled Alcoholic Beverages (SIC 518), Liquor Stores (SIC 5912),
Central Reserve Depository Institutions (SIC 601), Federal and Federally-
sponsored Nondepository Institutions (SIC 611) and Hospitals SIC (806).

Statistical Units on the SSEL

3
A brief overview can be found in http://www.census.gov/econ/overview/mu0600.html
4
The section draws heavily on Walker (1997).
5
Finance, Insurance and Real Estate (FIRE) and Transportation, Communications and Utilities
(TCU) were out of scope prior to 1992.

4
The SSEL is constructed from a variety of data sources including administrative
records and Census surveys. As a result the SSEL covers different statistical
units. The primary unit on the SSEL is the business establishment. An
establishment is a single physical location where business is conducted. The
establishment is the economic unit used in the Quinquennial Economic
Censuses and in many of the Census Bureau’s other business surveys.

However, there are other statistical units in the SSEL that may or may not be
equivalent to the business establishment. These are important when constructing
the LBD. First, the EIN unit is an administrative construct. The IRS assigns
Employer Identification Numbers for tax reporting purposes. An EIN unit is
comprised of one or more establishments. Second, the enterprise is an
economic unit comprising one or more establishments owned by the same legal
entity. That is, the enterprise is the highest-level parent company that controls
more than 50% interest in its affiliated establishments.

The relationship between establishment, EIN and enterprise units

By far the most common entity on the SSEL is the single unit establishment.
This occurs when the parent enterprise does business at only a single physical
location. In this case the establishment, EIN and enterprise units will refer to the
same single economic entity. The organization of the single unit enterprise
facilitates the use of administrative data for statistical purposes. Namely, the
enterprise files with the IRS under a single EIN, which in turn represents a single
physical location or establishment. In this case, the enterprise (or firm) is
identical to the tax reporting entity, which is in turn identical to the establishment,
which is the statistical unit for the Economic Census.

The other primary entity on the SSEL is the multi unit establishment. These units
represent establishments owned by enterprises conducting business in multiple
locations. The organization of multi unit enterprises complicates the use of
administrative data for statistical purposes.

The Census Bureau wants to provide data that describes the economy in as
much industrial and geographic detail as possible. The most logical way to
achieve the Census Bureau’s goals of providing both industrial and geographical
detail is to break the activities of multi unit enterprises up by establishments.
This also has the important advantage that it matches the reporting unit for single
unit enterprises.

Sources of administrative data, such as the IRS, do not share the Census
Bureau’s need for establishment level data. Multi unit enterprises report to the
IRS under at least one EIN. The Census Bureau must then break the enterprise
and its EINs into their constituent establishments. This is done primarily via the
Economic Census and the annual Company Organization Survey (COS).

5
Processing issues with the SSEL

There are a number of processing features of the SSEL that affect how
accurately it describes the universe of employer establishments in the U.S. The
primary purpose of the SSEL is to provide a mail-out frame for the Economic
Census. Understandably, the Census Bureau devotes more resources to the
SSEL when it is preparing to do an Economic Census and when it uses the
results from the census to update the SSEL. This introduces a 5-year cycle that
manifests itself in the data in several ways that users of the LBD should be
aware.

First, the quality of the single/multi-unit breakout declines after an Economic
Census and then improves again with the next census. Administrative data
describe EI units. New EI units will enter as single units. Most of these will, in
fact, be single units, but some will not. The Census Bureau often does not learn
this until an Economic Census. This results in spikes in the number of new multi-
unit establishments in Economic Census years.

Also, the coverage of the COS varies over the SSEL processing cycle and due to
budgetary considerations. This effects how well information on multi-unit
establishments is updated between Economic Censuses. This impacts both the
number of multi-unit establishments, and updates to their data items, such
payroll, employment, industrial activity and firm affiliation.

III. Creation of the LBD linkages

The core of the LBD is a set of links that describe how establishments in a
particular annual SSEL file relate to those in the preceding year. These links flag
establishment records as births, deaths or continuers. We construct these
linkages using numeric establishment identifiers along with name, address and
other information. Before creating the links, however, we tried to ensure that the
SSEL files we used were as complete and accurate as possible.

Preparation of SSEL files

1. Assembling the files

As mentioned above, the SSEL is constantly updated. The SSEL files
maintained at CES, however, are annual snap-shots. In recent years, CES
obtains files for a given reference year in fall of the following year, once COS
processing is complete. These files are obtained by CES from the Economic
Statistical Methods and Programming Division (ESMPD), which is responsible for
maintaining the SSEL file electronically. For each year, CES data staff creates
SAS® datasets for both the multi and single unit files.

6
Prior to 1995, CES did not directly acquire annual snapshot files from ESMPD.
For the years 1974 through 1994, CES obtained annual SSEL datasets from tape
backups or County Business Patterns (CBP) micro-data files. This work was
done in the mid nineties, by CES researcher Al Nucci, and involved translating
the data from a proprietary Census Bureau binary format to ASCII. To our
knowledge, the original archived data are no longer available.

Once at CES, the data were stored in compressed ASCII format, in as many as
50 files per year. Due to space limitations, the data were stored and used on
various media and computers. To facilitate the use of these files for this and
other research projects, CES data staff recently assembled all the data together
on a single computer and in SAS® datasets.

There were problems with several of the files that required attention before we
could begin matching. For example, there were tapes missing when the 1978
Single Unit data were brought to CES. Therefore, we are missing those
establishments whose data was stored on the missing tapes. We are also
missing the entire 1988 and part of the 1989 Multi Unit files. For these, we were
able to supplement with data from County Business Patterns files. Several other
fixes had to be made to other files before we could begin constructing links.
These are documented in Miranda (2002a).

2. Defining Active records

Not all records in the SSEL pertain to active establishments. There are several
reasons why this is so. SSEL records can represent non-establishment entities
or recently closed or otherwise defunct establishments (e.g. for single-units, the
Census Bureau keeps records up to 9 quarters after establishments stop
reporting data to the IRS). In addition, there are typically records on both the
single and multi-unit files that pertain to the same EIN. Before successive years
of the SSEL could be matched, we excluded inactive and duplicate records.

The procedures we used to arrive at the final annual SSEL datasets used to
construct the LBD are based on Foster (1999)6. These are somewhat
complicated and would involve discussing Census confidential material. Details
on the procedures used to delete inactive and duplicate records are available in
Miranda (2002a). The following is a summary of those procedures.

• We drop all records with zero annual payroll.
• We drop all records with flags indicating inactivity or that the record
does not pertain to an establishment.8

6
Foster (1999) worked in close consultation with the Economic Planning and Coordination
Division, which is responsible for the construction and maintenance of the SSEL.
8
An exception to this is out of scope EIN level records that represent multi location entities.
These occur primarily in public administration.

7
• We unduplicated records pertaining to the same EIN across both the
single and multi unit files. These usually arise when a multi-unit firm
reverts back to a single unit. In this case, we keep the MU record as
the information is more reliable than the administrative data on the
single unit file.
Table 1. SSEL Establishment Counts
By Year and Single / Multi-Unit Status
All All Active Active Total
Year Single-Units Multi-Units Total Single-Units Multi-Units Active
74 7,537,208 NA
75 7,777,714 1,119,935 8,897,649 3,866,226 822,906 4,689,132
76 6,335,307 984,980 7,320,287 4,120,205 854,077 4,974,282
77 6,580,974 1,221,155 7,802,129 4,124,203 1,029,573 5,153,776
78 5,653,200 1,318,752 6,971,952 3,516,614 1,032,373 4,548,987
79 8,193,678 1,429,334 9,623,012 4,311,722 1,051,010 5,362,732
80 8,519,356 1,531,623 10,050,979 4,252,325 1,059,400 5,311,725
81 4,672,851 1,634,420 6,307,271 4,200,275 1,072,311 5,272,586
82 5,227,582 1,543,325 6,770,907 4,154,743 1,165,171 5,319,914
83 4,958,975 1,564,158 6,523,133 4,110,086 1,168,247 5,278,333
84 5,282,458 1,694,606 6,977,064 4,367,954 1,194,318 5,562,272
85 5,738,143 1,279,839 7,017,982 4,734,395 1,169,512 5,903,907
86 5,791,491 1,424,844 7,216,335 4,789,675 1,197,080 5,986,755
87 5,974,577 1,909,018 7,883,595 4,860,589 1,336,888 6,197,477
88 6,058,104 1,353,827 7,411,931 4,918,356 1,327,763 6,246,119
89 6,470,451 1,665,868 8,136,319 4,996,037 1,306,561 6,302,598
90 8,724,368 1,860,663 10,585,031 5,309,701 1,354,733 6,664,434
91 9,061,948 1,634,273 10,696,221 5,302,086 1,351,572 6,653,658
92 9,605,164 1,958,000 11,563,164 5,293,965 1,493,705 6,787,670
93 10,561,452 2,019,632 12,581,084 5,401,127 1,471,281 6,872,408
94 10,733,629 2,231,134 12,964,763 5,498,328 1,498,450 6,996,778
95 11,430,782 2,356,256 13,787,038 5,584,717 1,515,521 7,100,238
96 10,231,525 1,781,654 12,013,179 5,705,858 1,484,490 7,190,348
97 10,656,372 2,150,213 12,806,585 5,728,271 1,604,417 7,332,688
98 11,701,778 2,324,255 14,026,033 5,760,682 1,612,765 7,373,447
99 12,778,708 2,549,522 15,328,230 5,791,404 1,655,470 7,446,874

We do not impose any industry or geographic scope restrictions. The Census
Bureau’s County Business Patterns program, which also uses the SSEL as its
primary data source, makes several restrictions of this nature. Therefore, the
LBD universe is larger than the CBP universe and our establishment counts will
exceed those in the CBP based BITS/LEEM file.

Table 1 lists the establishment counts for the SSEL. We list both the total
records on the CES SSEL files and the number of “active” establishments. The
total number of records on the SSEL fluctuates considerably from year to year.
However, the number of active establishments is much more stable and trends
up over the period.

8
It is possible to see the effects of missing SSEL data in table 1. This is especially
true for 1978. Fortunately, we are able to exploit the longitudinal properties of
the data to fill most of these gaps. This will be discussed further below.

Creating the Link Files

Serious work on constructing the LBD began at CES in 1999. Krizan (1999)
developed a methodology for linking adjacent years of the SSEL that was based
on Trager and Moore (1995). We have further modified the Krizan methodology
here (see Miranda 2002b and 2001).

The goal is to construct link, or pointer, files that describe how a record in one
year is related to records in the previous year. For each year of the LBD, we
want to be able to flag a record as a:

• Birth – a record in the current year that does not match to a record in the
preceding year,
• Death – a record in the preceding year that does not match to a record in
the current year, or
• Continuer – a record in the current year that matched to a record in the
preceding year.10

The Census Bureau assigns numeric establishment identifiers that make creating
longitudinal linkages a straightforward task for the majority of establishments.
However, for several years we have only a subset of these identifiers. In
addition, there are problems with the longitudinal characteristics of the Census
assigned numeric identifiers for a significant number of cases. These problems
necessitate augmenting the numeric matches with character-based matches
using information, such as establishment name and address. We used
Automatch®, sophisticated commercially available software originally developed
at the Census Bureau, to perform statistical record linkage using character based
information. Like Trager and Moore (1995) and Krizan (1999), we first matched
using numeric identifiers and then submitted the unmatched residual records to a
name and address matching procedure using Automatch®.

1. Numeric Matching

The Census Bureau maintains several numeric identifiers that can be used for
linking establishments longitudinally. These identifiers have different purposes
and characteristics that affect their usefulness for creating longitudinal linkages.

10
The LBD actually has a much richer set of flags that detail the nature of the linkage across two
years. For more information on how the flags were constructed and what information they convey
about the linkages, see Jarmin (2002a).

9
The following is a brief discussion of these characteristics and more detailed
explanations can be can be found by referring to the SSEL Glossary and Chapter
3 of the LRD Documentation (U.S. Census Bureau, 1998 and Center for
Economic Studies 2002).

• CFN: Census File Number. This is the primary processing ID used by the
Census Bureau to identify establishments in Economic Censuses and
surveys. The CFN for a given establishment can change over time for
several reasons (including ownership changes, changes in single and
multi unit status and changes in legal form of organization). Because it
can change over time, the CFN has limited usefulness as a longitudinal
identifier. The CFN is available over the entire period covered by the LBD.

• PPN: Permanent Plant Number. The PPN was introduced to the SSEL in
1982 to facilitate longitudinal linkages. As it does with the CFN, the
Census Bureau assigns each establishment a unique PPN. Unlike the
CFN, however, the PPN for a given establishment is designed to remain
unchanged as long as the establishment remains active at the same
location. The PPN is the best available longitudinal identifier on the SSEL.
Unfortunately, it is not available for all years and sectors on the LBD. It
was not introduced to the business register until 1982. PPNs exist prior to
1982 only for manufacturing establishments on the LRD. In addition, there
was a change in the method by which PPNs were assigned in 1985.

• EIN: Employer Identification Number. This is the establishment’s taxpayer
ID assigned by the IRS. For single unit establishments, the EIN is a
unique identifier and is equal to the CFN (less a leading 0 attached to
create the 10-digit CFN from the 9-digit EIN). This is not the case,
however, for multi-unit establishments. A multi-unit firm will have at least
one EIN. Each EIN that a multi-unit firm reports under will be contained
on the Single Unit file and flagged as a “sub master” record.

• NEWID/OLDID: The SSEL tracks one change in the CFN. The OLDID is
the predecessor CFN and NEWID is the current CFN.

We use the same numerical linking methodology as Krizan (1999). When they
were available, we matched first by PPN, flagged the matches and set aside the
residuals. If PPNs were not available, we matched by CFN first. After CFN, we
matched by EIN and then finally by OLDID/NEWID.

2. Name and Address Matching.

There are two reasons why we are interested in augmenting the numeric
matches with name and address matches. First, the PPN is the only true
longitudinal identifier, and it is missing for the early years of the LBD. Second,
numeric identifiers, including the PPN, are subject to errors. We want to make

10
sure we are finding as many valid year-to-year establishment linkages as
possible, without picking up erroneous ones. Missing matches inflate the number
of establishment births and deaths.

The first problem is pretty straightforward and it’s easy to see why we would want
an additional matching method when PPNs are not available. In years where
PPNs are missing we simply pass the unmatched records from the two adjacent
years through the statistical matching algorithm we developed in Automatch®.

Dealing with PPN errors is more complicated and, therefore, requires a little more
discussion. Although the PPN for an establishment is not supposed to change
as long as business is conducted continuously at a given location, there are
reasons why a PPN might change. First, there is the valid reason of splitters and
combined reports. In the case of splitters, the Census Bureau may ask an
establishment to break out reporting at a single establishment into two or more
establishments. Likewise two adjacent establishments may be merged into one
combined report. In both cases, no establishment entry or exit has occurred, but
the number of records and, hence, PPNs on the SSEL changes. These types of
changes are rare, however.

The most common reason why PPN linkages are broken is due to processing
errors triggered by changes in establishment status. These are caused by
events such as changes in ownership due to mergers and acquisitions, changes
in the establishment’s legal form of organization, and changes in single/multi-unit
status. Trager and Moore (1995) term these “reorganizations.” Depending on
when the reorganization is coded to the SSEL, there are different ways of dealing
with this problem.

The simplest cases to handle are year-to-year reorganizations. These arise
when the break in the PPN and other numeric identifiers occurs across two
annual versions of the SSEL. That is, continuing establishments in the year t
SSEL will not link numerically to records in the year t+1 file. We would record
false deaths between year t and t+1. Likewise, continuing establishments in year
t+1 would not link numerically to records in the year t file and would be incorrectly
identified as births.

The remaining classes of reorganizations are more complicated. Mid-year and
birth reorganizations arise when the break in the numeric linkages occurs within
an annual SSEL file. In the case of mid-year reorganizations, an establishment
continuing from t-1 through t+1 would undergo some change in status during
year t that causes a break in the numerical linkages. This change, however,
results in duplicate records for the establishment on the year t SSEL. One of
these records links numerically to year t-1, but not to year t+1 and the other links
numerically to year t+1 and not year t-1. Likewise birth reorganizations occur
when a new establishment undergoes a change in status in its initial year. Again
there will be duplicate records for the establishment in year t, but only one of

11
them will link numerically to a record in the subsequent year. Finally, in addition
to splitters/combined reports and reorganizations, numeric establishment links
can be broken through keying and coding errors.

Name and address information on the SSEL can be used to repair broken
numerical linkages. However, these fields are difficult to compare using standard
computer matching techniques. Matching fields in SAS® or other commonly used
statistical or database software requires an exact match. There are many ways
that names and addresses can vary across two lists and still contain the same
information. To the human eye a match may be obvious, but to a computer if the
fields being matched differ even slightly, no match will be made.

The problem of linking records when exact matching is not possible has received
considerable attention by a small cadre of researchers (see Internal Revenue
Service 1985). The state of the art in statistical record linkage has advanced
such that there is now commercially available statistical matching software. We
use Automatch®, a software package based on code originally developed at the
Census Bureau. 12 A key feature of the software is its standardization routines.
These take the name and address fields that we supply from the SSEL, remove
extraneous information and output standardized fields for matching. The
software assigns weights to potential matches. Those above a user-defined
cutoff are matches.

We did considerable testing before settling on the Automatch® algorithm we
employed to repair broken numerical linkages for the LBD. A summary of these
tests and a description of the final algorithm are available in Miranda (2002b).

3. Post-Matching Edits and Coping with Extended Periods of Establishment
Inactivity

Discussion of our longitudinal linking algorithms so far has focused solely on how
active establishment records in a particular annual snap-shot of the SSEL link to
active records in the previous annual snap-shot. There are two situations where
we need to link records over longer time periods: i) to fill in missing source data,
and ii) to allow for extended periods of establishment inactivity.

In the first case, we simply exploit the longitudinal nature of the LBD to infer
establishment records in cases where we know we are missing SSEL source
data. This occurs for the years 78, 83-86, 89 and 91 (see Miranda 2002a,
2002c). To see how we fill “holes” in the LBD, imagine an active establishment
whose record is on one of the missing 1978 SSEL tapes. We simply match 78
deaths (active in 77 but not in 78) to 79 births (active in 79 but in 78). If a match

12
Automatch® is now available under the trade name Vailty® (see www.vality.com). For details
on Automatch® see MatchWare Technologies (1997).

12
is made, we infer that the establishment was active in 78. We also can do this
for missing 78 births. Namely, we look positive prior year payroll on 79 births that
were not active in 77. The vast majority of the cases that required post-matching
linkage flag edits, were cases where we needed to correct for missing source
data. In a smaller number of cases, we needed to fix incorrectly assigned
linkage flags. All post-match edits are flagged accordingly to give researchers
more information about the nature of the longitudinal links. More details are
available in Jarmin (2002a).

In the second case, we consider establishments that undergo periods, in excess
of a year, with no payroll for reasons other than missing source data. For
example, consider an establishment that was active from 1988 to 1994, then was
inactive (no payroll) for two years and then became active once more in 1997.
The linking algorithms we have described thus far would classify the
establishment as a death in 1995 (active in ’94 but not in ’95) and as a birth in
1997 (active in ‘97 but in ’96) with a new longitudinal identifier. That is, the LBD
would contain two records for the same physical establishment.

Following the methodology we have outlined thus far, we constructed a prototype
LBD that did not allow establishments to undergo extended periods of inactivity.
Jarmin (2002b) matched this first prototype LBD to a longitudinal version of the
Annual Survey of Manufactures (ASM) and found that these periods of inactivity
are much more prevalent than we had previously thought. Many establishments
leave the active SSEL universe for two or more years and re-enter later, often
with the same numeric IDs.

There are several explanations for why we might observe this type of behavior in
the data. At this point, we have not done a detailed analysis, but a likely
explanation is that we are observing businesses that transition back and forth
between the employer (positive payroll) and non-employer universes. Other
explanations include processing errors and re-tooling.

We have modified our linking algorithms to take account of establishments with
extended periods of inactivity. The details of this are given in Jarmin and
Miranda (2002). When we compare the revised LBD with the first prototype, we
find that accounting for extended periods of establishment inactivity reduces the
number of unique establishments we have in the file by over 1.2 Million. This
reduces the number of births and deaths in the current version of the LBD by
over 40,000 per year as compared to the first prototype. Because we are still not
sure of the status of many of these establishments we have separately identified
them in the LBD so that researchers can treat them differently if they choose.

4. Matching Results

The current version of the LBD contains data for all employer establishments
from 1975 through 1999. We use data from the 2000 SSEL, but the nature of the

13
algorithms we used to deal with reorganizations means we can’t compute links
for the last year with available data. The first prototype LBD contained data only
up to 1998. Thus, we have already updated the file once and will continue to do
so on an annual basis when we receive new SSEL data.

Table 2 summarizes the results from our numeric and name and address
matching. It also shows the results from post-matching fixes we implemented to
correct for missing data and miscoded linkages and to accommodate temporarily
inactive establishments.

The first two columns in table 2 present, by year, the number of active
establishments on the archived SSEL files at CES (see table 1), and the number
of active establishments on the LBD after all matching and editing was
completed. For most years the number LBD establishments is lower than the
number of SSEL establishment because we eliminate duplicate records
introduced by reorganizations. For the years 78, 83-86, 89 and 91, we use the
longitudinal nature of the LBD to correct for missing source data on the SSEL
(see Jarmin, 2002a for more details). Thus, the LBD has more records than the
(active) SSEL for these years.

The next three columns break out the LBD by continuers, births and deaths.
Recall that the linkages on the LBD for a given year describe how establishments
link to records in the prior year. Therefore, when interpreting the results in table
2, one should remember the number of continuers in, for example, 1994 is the
number that were present in both 1993 and 1994. Likewise the number of births
in 1994 refers to establishments that were active in 1994, but not in 1993, and
the number of 1994 deaths refers to establishments that were active in 1993 and
not in 1994. We list “true” births, deaths and continuers. That is we exclude
from deaths those establishments that exit the active employer establishment
universe only to re-enter at some later point. Likewise, we exclude from births
those establishments that are re-entering after an extended period of inactivity.
The numbers of establishments that enter and exit the active employer universe
due to extended periods of inactivity are listed separately in columns under the
“Temporarily Inactive” header.

14
Table 2: Match Results
Establishment Counts Nature of Linkages13

Temporarily Inactive Name &
Active SSEL Active LBD Continuing Births Deaths Address
Year Exit Enter Matches
76 4,974,282 4,945,528 4,120,965 824,563 580,697 56,238 47,538
77 5,153,776 5,125,942 4,244,170 844,422 658,862 42,496 37,350 47,808
78 4,548,987 5,152,243 4,437,114 683,598 660,576 28,252 31,531 39,204
79 5,362,732 5,330,266 4,633,798 681,813 480,788 37,657 14,655 48,625
80 5,311,725 5,283,897 4,632,827 610,991 621,769 75,670 40,079 45,309
81 5,272,586 5,244,139 4,526,994 649,292 690,877 66,026 67,853 56,296
82 5,319,914 5,294,765 4,570,950 702,036 579,573 93,616 21,779 44,832
83 5,278,333 5,586,606 4,711,849 755,528 548,637 34,279 119,229 31,575
84 5,562,272 5,833,945 5,012,940 779,039 550,405 23,261 41,966 32,831
85 5,903,907 5,981,692 5,181,467 771,830 618,645 33,833 28,395 41,532
86 5,986,755 6,098,536 5,305,934 763,103 635,823 39,935 29,499 41,862
87 6,197,477 6,174,220 5,298,607 851,033 729,996 69,935 24,580 39,299
88 6,246,119 6,228,218 5,442,626 717,030 613,470 118,129 68,562 35,206
89 6,302,598 6,388,877 5,482,319 797,117 709,458 36,443 109,441 36,715
90 6,664,434 6,645,560 5,673,200 933,622 643,935 71,751 38,738 32,611
91 6,653,658 6,729,082 5,853,353 799,454 745,541 46,670 76,286 34,016
92 6,787,670 6,759,906 5,939,778 787,850 704,433 84,872 32,289 43,189
93 6,872,408 6,860,000 6,026,126 746,635 654,895 78,918 87,239 31,970
94 6,996,778 6,973,457 6,128,377 760,594 661,737 69,912 84,486 39,509
95 7,100,238 7,077,456 6,245,314 754,795 658,473 69,695 77,347 39,057
96 7,190,348 7,167,943 6,314,604 766,265 705,050 57,867 87,075 38,942
97 7,332,688 7,305,127 6,330,006 894,978 790,222 47,896 80,146 44,380
98 7,373,447 7,351,196 6,511,988 754,708 793,139 84,799 20,073
99 7,446,874 7,405,245 6,577,081 828,164 774,793 16,427
Total 147,840,006 148,943,846 129,202,387 18,458,460 15,811,794 1,283,351 1,283,324 928,806

The 1976 through 1999 LBD (which also includes data, but not links for 1975)
contains 166,087,537 longitudinal (active LBD + deaths) linkages for 23,259,023
unique establishments. The last column of table 2 lists the number of name and
address matched by year. Name and address linkages account for a small
portion of the total number of linkages. However, at some point in their tenure in
the LBD, approximately 3.68% of the establishments required name and address
matching, at least once, in order to preserve their longitudinal linkages.

13
Note that the categories “births”, “deaths” and “continuers” are agglomerations of more refined linkage
flags contained in the LBD linkage file. Thus, the information contained in the linkage flags is more
extensive than what in presented here. Jarmin (2002a) and Jarmin and Miranda (2002b) provide more
details.

15
A key concern is to separate true births and deaths from spurious ones
generated by the processing systems that generate the data underlying the LBD.
We have tried to ensure that the data in the LBD are as accurate as possible.
We have used both numeric identifiers and name and address matching to try to
find all the linkages between records on successive years of the SSEL. We have
also used the longitudinal nature of the LBD to fill in for missing SSEL source
data and account for establishments that enter and exit the LBD due to periods of
inactivity. The impact of this “post-match” processing can be seen in figures 1
and 2. Figure 1 shows the births and deaths obtained after linking the SSEL
files, as they exist at CES. The most striking feature of figure 1 is the huge spike
in deaths in 1978 (i.e., establishments active in 1977 and not in 1978) followed
by a similar spike in births in 1979 (i.e., establishment active in 1979 and not in
1978). This results from the fact that the 1978 SSEL file at CES is incomplete.

Missing source data for 1978 and other years, miscoded LBD linkages and
temporary periods of inactivity create problems when computing establishment
birth and death rates in the LBD. That is why we go to such care to correct, or at

Figure 1
LBD Establishment Births and Deaths Before Post-Match Fixes
1,400,000
1,200,000

Number of
1,000,000
800,000 Death
600,000 Total Births

Establishments 400,000
200,000
0
74 76 78 80 82 84 86 88 90 92 94 96 98
Year

least take account of, these problems. Figure 2 shows the total numbers of
establishment births and deaths after we perform our post-matching edits.15 The
figure clearly shows that the post-match fixes smooth the birth and death series’
considerably. We juxtaposed the LBD birth and death series’ in figure 2 with
GDP growth. As expected, net entry (the difference between births and deaths)
appears to be pro-cyclical. However, we are still concerned that processing
drives some of the results. We are suspicious of the birth spikes in 1987 and

15
“Total Births” is births plus re-entry of temporarily inactive establishment. Likewise, “Total
Deaths” is deaths plus temporary exits. We report totals since the algorithms understate the
number of re-entrants towards the end of the LBD since we do not yet know whether
establishments that exit in the last few years, might yet re-enter.

16
1997. The spike in births in 1990 can be attributed to changes in the processing
of agricultural establishments.

Figure 2
LBDv2 Establishment Births and Deaths
After Post-Match Fixes
1,200,000 8.0 Total Births

1,000,000 6.0

800,000 4.0
Total Deaths
600,000 2.0

400,000 0.0
GDP percent
200,000 -2.0 change based on
chained 1996
0 -4.0 dollars
76 78 80 82 84 86 88 90 92 94 96 98

IV. The LBD

The LBD covers over 23 million establishments between 1975 and 1999 and
contains limited information on their characteristics and activities. Additional
establishment level information is available, but we chose only to include what
information was available consistently over time and across different sectors.
The limited detail on the LBD is mitigated by the ease with which it can be linked
to other establishment and firm data from the Census Bureau and other sources.
In this section, we discuss several of the data items we have included on the
LBD.

Establishment Age and Tenure

One of the big advantages of the LBD is better establishment age and tenure
information. Static datasets based on Economic Census or survey files typically
contain no information on establishment age.17 The LRD can be used to
measure the age of manufacturing plants opened after 1963. Researchers using
the BITS file have utilized a variable (syr) that indicates the year the record was
added to the SSEL (Robb 1999; Acs and Armington 1998; Nucci 1999).
However, that variable is not completely reliable for the same reasons that the
PPN is not.

17
The exception is the 1975 And 1981 ASMs where samples of manufacturing plants were asked
to list when the plant began operations at the current location.

17
We can easily compute establishment tenure in the LBD. For all but those
establishments active in 1975, we can compute age as well. Establishment
tenure is just the difference between the last and first year for which an
establishment has non-missing linkages in the LBD. For example, an
establishment born in 1978 and dying in 1988 would have tenure of 11 years
since it would have valid linkage flags from 1978 (birth) to 1989 (death, recall that
the flags describe the nature of the linkage to the prior year). Computing age is
similar. However, we can’t compute a true age for establishments with left
censored data (i.e., those in the LBD in 1975). This problem is most acute in the
early years of the LBD and steadily diminishes over time as the 75 cohort thins
and is replaced by new entrants. For example, we can compute age for only
17% of the establishments active in 1977 and for 88% in 1998.

Figure 3 shows the tenure profile for establishments in the LBD. Clearly, most
establishments are in the LBD for only a short time. Tenure is the same as age
for all but the 75 cohort, and, as expected, we see a monotonic decrease in
tenure. The blip at tenure equal to 24 years represents establishments active in

Figure 3
LBD Establishment Tenure

5000000

Number of
4000000

3000000
Number

Estabslishments
2000000

1000000

0
0 2 4 6 8 10 12 14 16 18 20 22 24
Years Active in the LBD

every year of the LBD. Note that establishments with tenure equal to 0 are 1999
entrants and, therefore, have the same initial and terminal year in the current
version of the LBD.

We are concerned that data quality may be poor for establishments with few
observations in the LBD. This is especially the case for establishments never
surveyed by the Census Bureau, and for which all the information we have is
derived from administrative sources. While this is clearly a problem when looking
at individual establishments, it is less so when looking at the number of plant-
year observations in the LBD. Figure 4 shows the cumulative tenure distributions
for establishments, and for establishments weighted by the number of years we
observe then in the LBD. Whereas 50% of establishments are in the LBD for
three years or less, more than 75% of the plant-year observations are contributed

18
by establishments that are active in the LBD for 5 or more years. These
establishments are much more likely to have been canvassed by the Economic
Census or some other Census Bureau collection and, therefore, have better
industry, geographic, employment and other data. We will be able to test this
more precisely once we begin testing the linkages between the LBD and census
and survey data. The LBD will be an excellent tool to assessing whether current
data collections are of sufficient detail and frequency to accurately measure the
economy.

Figure 4
Cumulative Tenure Distributions
100.00%

75.00%

Cumulative %
Cumulative % of
50.00% Establishments
Cumulative % of
25.00% Observations

0.00%
0 2 4 6 8 10 12 14 16 18 20 22 24
Tenure (years active in the LBD)

Industry and Geographic Classifications on the LBD

The LBD contains industry and geographic classifications similar to those on
most Census Bureau establishment data sets. Currently, however, these are
limited to contemporaneous codes and they can be missing or inconsistent over
time. We are currently working on improving both the geographic and industry
data contained in the LBD. One of our biggest challenges is to provide
consistent industry coding over the entire history of the LBD. This is complicated
by the fact that the period of the LBD spans two major SIC (Standard Industrial
Classification) regimes and the switch to NAICS (North American Industrial
Classification System). Since these classifications (especially geographic) vary
little within establishments over time, we can exploit the longitudinal nature of the
LBD to fix missing and inconsistent codes.

Basic Economic Variables in the LBD

In addition to geographic and industry information, the LBD contains basic data
on payroll and employment. These provide researchers with basic measures of
the scale of economic activity taking place at establishments. In the current
prototype LBD, these items come from the SSEL. Once we link the LBD to the
Economic Censuses and other surveys, we will edit the payroll and employment
data in the LBD to reflect the best available information.

19
Unlike industry and geography codes, payroll and employment data are much
more sensitive to data quality problems, since they vary more within
establishments over time. This is particularly important in reference to missing
source data and SSEL processing cycles.

To get an idea of the scope and quality of our current employment measures we
compare LBD employment with numbers from CBP. Figures 5 and 6 plot
employment, by year in major industry groups, and compare CBP numbers (red)
with LBD employment numbers (green) and with the LBD using CBP type in-
scope restrictions (brown). We see that, especially in agriculture, TCU, FIRE and
services, the scope of the LBD is larger than the scope of CBP resulting in higher
LBD employment for these sectors. On the whole, however, the LBD tracks CBP
employment very well for most sectors once we restrict attention to
establishments that are in scope for the CBP.

Firm Affiliation

Currently there is basic firm affiliation information for all establishments in the
prototype LBD. This comes from the CFN. We intend to make several
improvements and enhancements to the information on firm affiliation in the LBD.
We want to track ownership changes and create flags that describe the nature of
the linkages between the ownership statuses of establishments over time.

V. Summary

This paper documents recent efforts at the U.S. Census Bureau to construct a
new longitudinal establishment data set. Several aspects of the LBD set it apart
from existing longitudinal establishment datasets. First, it covers all industrial
sectors of the economy. Widely used longitudinal establishment files, such as the
LRD, have been limited to the manufacturing sector. However, this sector is
declining in importance and it may not be representative of other industries. The
LBD will allow research into expanding sectors of the U.S. economy. Second, the
LBD covers all employer establishments (with a minimum of one employee). The
wide coverage offers possibilities for research into the life cycle of small firms.
Third, it contains detailed geographic data allowing research on the dynamics of
business location and structure. It will be possible to explore the interactions
between local communities and businesses and also amongst types of
businesses. Fourth, establishment data go back 25 years and covers multiple
business cycles. Other establishment data sets are typically limited to the 1990's.
Finally, linking the LBD to other Census Bureau establishment data is relatively
trivial making it possible to considerably enhance the scope and depth of
information for LBD establishments. The new data has the potential to enhance
our understanding of such topics as job creation and destruction, firm turnover,
the life cycle of establishments and changes in the structure of the U.S. economy
amongst others.

20
LBD: Employment LBD with CBP Exclusions LBD: Employment LBD with CBP Exclusions LBD: Employment LBD with CBP Exclus
CBP: Employment CBP: Employment CBP: Employment
1200
1000 6000

1000
5000
500
800
4000

0 600 3000
76 78 80 82 84 86 88 90 92 94 96 9899 76 78 80 82 84 86 88 90 92 94 96 9899 76 78 80 82 84 86 88 90 92 94 96 9899
yr yr yr
Agriculture, Forestry and Fishing Mineral Industries Construction Industries
LBD: Employment LBD with CBP Exclusions LBD: Employment LBD with CBP Exclusions
CBP: Employment CBP: Employment
22000 7000

21000
6000
20000

5000
19000

18000 4000
76 78 80 82 84 86 88 90 92 94 96 9899 76 78 80 82 84 86 88 90 92 94 96 9899
yr yr
Manufacturing TCU

Figure 5 Employment (x1000): by SIC 87 Major Division

21
LBD: Employment LBD with CBP Exclusions LBD: Employment LBD with CBP Exclusions LBD: Employment LBD with CBP Exclus
CBP: Employment CBP: Employment CBP: Employment
7000 25000 10000

6000 20000 8000

5000 15000 6000

4000 10000 4000
76 78 80 82 84 86 88 90 92 94 96 9899 76 78 80 82 84 86 88 90 92 94 96 9899 76 78 80 82 84 86 88 90 92 94 96 9899
yr yr yr
Wholesale Trade Retail Trade FIRE
LBD: Employment LBD with CBP Exclusions LBD: Employment LBD with CBP Exclusions
CBP: Employment CBP: Employment
50000 4000

40000
3500
30000
3000
20000

10000 2500
76 78 80 82 84 86 88 90 92 94 96 9899 76 78 80 82 84 86 88 90 92 94 96 9899
yr yr
Service Industries Public Administration

Figure 6: Employment (x1000): By SIC 87 Major Division

22
References

Acs, Z., and C. Armington, (1998), “Longitudinal Establishment and Enterprise
Microdata (LEEM) Documentation,” Center for Economic Studies Working
Paper CES 98-1.

Bartelsman, E.J, and M. E. Doms (2000), “Understanding Productivity: Lessons
from Longitudinal Microdata,” Journal of Economic Literature, 38(3), pp.
569-94.

Caves, R. E., (1998), “Industrial Organization and New Findings on the Turnover
and Mobility of Firms,” Journal of Economic Literature, 36, pp. 1947-82.

Center for Economic Studies, (2002), “Establishment and Firm Definitions,” in
Longitudinal Research Database Documentation Manual, Ch3.

Dunne, T., (1992), “Documentation Note of Corrections made to PPN’s During
Correction Process on Plant Linkage Errors on the LRD.

Foster, L., (1999), “Documentation for the ES202/SSEL Joint Project,” internal
memo, Center for Economic Studies.

Internal Revenue Service, (1985), Record Linkage Techniques – 1985:
Proceeding of the Workshop of Exact Matching Methodologies, Statistics
of Income Division, Internal Revenue Service, Publication 1299 (2-86).

Jarmin, R., (2002a), “Post-Matching Linkage Edits on the Longitudinal Business
Database,” Center for Economic Studies Technical Note, CES-TN-2002-
03.

Jarmin, R., (2002b), “Improved Age Information for ASM Establishments,” Center
for Economic Studies Technical Note, CES-TN-2002 [Forthcoming].

Jarmin, R. and Miranda, J., (2002), “LBD Documentation: Broken Longitudinal
Links from Missed Reorganizations and Inactive Spells,” Center for
Economic Studies Technical Note, CES-TN-2002 [Forthcoming].

Krizan, C.J., (1999), “SSEL Linkage Project Documentation,” Internal Document,
Center for Economic Studies, U.S. Census Bureau.

MatchWare Technologies, (1997), Automatch, Generalized Record Linkage
System: User’s Manual, Burtonsville, MD.

McGuckin, R., and G. Pascoe, (1988), “The Longitudinal Research Database
(LRD): Status and Research Possibilities,” Survey of Current Business,
Nov. 1988, pp. 30-37.
Miranda, J., (2002a), “LBD Documentation: Defining Active Establishments and
Other Data Issues”. Center for Economic Studies Technical Note, CES-
TN-2002-04.

Miranda, J., (2002b), “LBD Documentation: Evaluation of Probabilistic Linking
Algorithms in the Auto Repair Shop Industry”. Center for Economic
Studies Technical Note, CES-TN-2002-02.

Miranda, J., (2002c), “LBD Documentation: Geography”. Center for Economic
Studies internal document.

Miranda, J., (2001), “Statistical Linking With AUTOMATCH®: Diagnosis of the
SSEL 1996 to 1997 Full Linking Procedure”. Center for Economic Studies,
Internal Memo.

Monahan, J., (1992) “Longitudinal Research Database (LRD) Technical
Documentation Manual,” Internal Document, Center for Economic Studies,
available at http://weasel.ces.census.gov.

Nucci, A., (1993), “Longitudinal Business Database: Project Proposal,” Internal
Document, Center for Economic Studies.

Nucci, A., (1999), “The Demography of Business Closings,” Small Business
Economics, 12, pp. 25-39.

Pivetz, T., M. Searson and J. Spletzer, (2001), “Measuring Job and
Establishment Flows with BLS Longitudinal Microdata,” Monthly Labor
Review, 124(4), pp. 13-20.

Robb, A. (1999), “New Data for Dynamic Analysis: The Longitudinal
Establishment and Enterprise Microdata (LEEM) File,” Center for
Economic Studies, CES 99-18.

Spletzer, J., (1997), “Longitudinal Establishment Microdata at the Bureau of
Labor Statistics: Development, Uses, and Access,” Proceedings of the
Section on Survey Research Methods, American Statistical Association.

Trager, M. L., and R. A. Moore, (1995), “Development of a Longitudinally-Linked
Establishment Bases Register, March, 1993 Through April, 1995, mimeo,
U.S. Census Bureau.

U.S. Census Bureau, (1979), “The Standard Statistical Establishment List
Program,” Technical Paper #44.

24
U.S. Census Bureau, (1998), “Business Register: Glossary of Data Elements,”
Internal Document.

Walker, E., (1997), “The Census Bureau’s Business Register: Basic Features
and Quality Issues,” paper presented at the Joint Statistical Meetings,
Anaheim, CA.

25
Appendix

Table A.1: LBD files and variables
File Format Data Items
LBD_LINK7599_V2 SAS 8 Dataset LBDNUM; CFN75-CFN99; MU75-MU99;
RECNUM75-RECNUM99; FLAGA75-
FLAGS99
LBD_PAY_EMP_7599 SAS 8 Dataset LBDNUM; EMP75-EMP99; PAY75-
PAY99
LBD_GEO SAS 8 Dataset LBDNUM; STGEO75-STGEO98;
STGEO275-STGEO298; CTYGEO75-
CTYGEO98; CTYGEO275-
CTYGEO298; ZIP75-ZIP98
LBD_GEOCTE SAS 8 Dataset LBDNUM; STATE; COUNTY, FLAG,
FLAG2
LBD_LINK7599_V2_IND SAS 8 Dataset LBDNUM; YR; RECNUM; MU; SICRL;
SICRL87; SICYR; SICYR87; SIC; SIC72;
SIC87

File and source

File: covid-corporate-financing-facility-ccff-information-for-those-seeking-to-participate-in-the-scheme.pdf
Size: 580,691 bytes
SHA-256: f5aa9958f7dc3beb624e72f0c68c2a0490f7a79d1ba6e3092198af386de8721b
Our copy: covid-corporate-financing-facility-ccff-information-for-those-seeking-to-participate-in-the-scheme.pdf
Original: www2.census.gov