Fewer employers, more firms without payroll: US leather and allied product manufacturing, 2004–2023

Preprint, version 1 (October 2026). Not peer reviewed. This is the author’s manuscript, prepared for submission to a peer-reviewed journal. Download: PDF · Word · reproducibility package (zip) · code on GitHub with code, results and the two independent audits.

Cite as: Kalafatis, P. (2026). Fewer employers, more firms without payroll: US leather and allied product manufacturing, 2004–2023 [Preprint]. x9m8.com. https://doi.org/10.5281/zenodo.23982176

Peter Kalafatis (ORCID: https://orcid.org/0009-0000-1637-7142)

Independent researcher, x9m8.com, New York, NY, USA. Correspondence: x9m8x7@gmail.com

Version 1, October 2026.

Abstract

Official counts of US leather and allied product manufacturing (NAICS 316: tanning, footwear, handbags, luggage and other leather goods) are split between two Census Bureau programmes that are rarely read together: County Business Patterns (CBP) for establishments with paid employees, and Nonemployer Statistics (NES) for firms with none. I extracted NAICS 316 from the national public bulk files for 2004–2023 (CBP) and 2004–2024 (NES), recorded the URL, retrieval date and SHA-256 hash of each of 41 source files, documented the 2012 consolidation of the 6-digit codes, and describe the series. From 2004 to 2023 employer establishments fell from 1,475 to 986 (−33%) and March employment from 43,335 to 22,970 (−47%), while nonemployer firms rose from 4,426 to 5,855 (+32%). The nonemployer share of all units rose from 75.0% to 85.6%, but not steadily: +8.0 percentage points by 2012, −0.3 from 2012 to 2021, +2.9 in 2021–2023. All manufacturing rose by a similar amount (47.2% to 58.8%), so the leather rise is not unusual for manufacturing. Employer establishments with 50 or more employees fell by more than half, and the share with fewer than 10 employees rose from 62% to 65% (peaking at 70% in 2015). The two largest nonemployer jumps, 2012 and 2022, fall in NAICS vintage years and I cannot separate them from classification effects. The findings are descriptive; the dataset and code are public domain and reproducible from the listed files.

Keywords: nonemployer firms; County Business Patterns; Nonemployer Statistics; leather manufacturing; NAICS 316; small business; open data

Social media summary: US leather goods makers with payroll fell by a third since 2004; firms with no employees rose by a third. Open Census dataset.

1. Introduction

Counts of businesses in the United States come from more than one source, and the sources count different things. County Business Patterns (CBP) is an annual series of establishments with paid employees, built from the Business Register, with employment, payroll and a breakdown by employment-size class. Nonemployer Statistics (NES) counts businesses with no paid employees, subject to federal income tax, and reports their number and receipts. A question as simple as “is an industry shrinking?” can have two answers depending on which programme is read. The employer-and-nonemployer split was one of the problems addressed when Davis et al. (2007) integrated the two universes to measure the dynamics of young and small businesses, while Haltiwanger et al. (2013) showed that the link between firm size and job creation depends on firm age, and Hurst and Pugsley (2011) argued that many small businesses are not intended to grow. A Census Bureau note reports that, across all industries, nonemployers grew by an average of 2.7% a year from 2012 to 2023 against 1.1% for businesses with paid employees (Census Bureau, 2025), and Conroy et al. (2026) analyse the geography of nonemployer growth across US counties. Katz and Krueger (2019) documented a rise in alternative work arrangements using survey data.

Leather and allied product manufacturing is a small, old industry in which the question is concrete. Import competition is the usual backdrop for US manufacturing decline (Autor et al., 2013, study the effect of Chinese imports on US local labour markets across manufacturing; I make no claim about its role in leather), and the industry includes both factories with payroll and one-person workshops: a bag maker working alone is, in the statistics, a nonemployer. I am not aware of a study that sets the CBP and NES series for this industry side by side. A Crossref search and a web search found none, but a search is not proof of absence.

This paper has three aims. First, to build and document an open dataset: NAICS 316 and its 4- and 6-digit industries from the public bulk files of both programmes, 2004 to the latest year, with the provenance of every source file and an account of the classification changes and disclosure rules that affect comparability. Second, to describe how employer establishments, employment and payroll, and nonemployer firms and receipts, changed. Third, to ask whether the nonemployer share rose, and whether the series support calling it a shift, and to describe the size distribution of employer establishments. The analysis is descriptive. It has no design that could support a causal claim, and I make none.

2. Methods

2.1 Sources

All data are public files on the Census Bureau’s file server (www2.census.gov) that need no key. Employer data: the national County Business Patterns files, cbp<yy>us.txt (2004–2007) or cbp<yy>us.zip (2008–2023), under programs-surveys/cbp/datasets/<year>/. Nonemployer data: the national Nonemployer Statistics files, nonemp<yy>us.txt (2004–2015) or .zip (2016–2024), under programs-surveys/nonemployer-statistics/datasets/<year>/historical-datasets/. File names were taken from the directory listings. CBP 2023 and NES 2024 were the latest files listed on 2 November 2026; no CBP 2024 directory was found. The shell used for the analysis could not reach the server, so each file was fetched with JavaScript in a browser tab on www2.census.gov on 2 November 2026, hashed (SHA-256) in the browser, and only the rows for NAICS 316, manufacturing (31-33) and all industries were extracted and copied out. The URL, byte size, SHA-256 hash, Last-Modified header and retrieval date of every source file are in data/source_files.csv. The extracted rows are identical in the browser and in the package (the SHA-256 of each extract was compared). The full raw files are not redistributed because they are large; they can be re-downloaded from the listed URLs, and the hashes show whether they have changed.

The NES 2007 national file (nonemp07us.txt) returned an HTML “Request Rejected” page on every attempt, although the file is listed, and its Last-Modified value in the listing was 29 October 2026. I did not use it. The 2007 national values for NAICS 316, manufacturing and all industries were instead summed from the 51 state rows of the state file (nonemp07st.txt). The same procedure applied to the 2006 state file reproduces the published 2006 national figures exactly (4,706 firms; $157,045 thousand in receipts in NAICS 316). In 2007 two states (Rhode Island and Vermont) have the value withheld (“D”) in NAICS 316, so the summed 2007 value, 5,148, is a lower bound. Sub-industry values for 2007 were not derived. The 2007 value is plotted with a hollow marker and is not used for any endpoint.

2.2 Variables and definitions

From CBP: number of establishments (EST), employment in the pay period including 12 March (EMP), first-quarter payroll (QP1) and annual payroll (AP), both in $1,000, and establishment counts by employment-size class. From NES: number of nonemployer firms (ESTAB in the file) and receipts (RCPTOT, $1,000). Rows for all legal forms of organisation and, for NES, all receipts-size classes were used. All dollar values are nominal; I did not deflate them, to avoid importing a price series I could not retrieve and verify here.

The units differ. CBP counts establishments (a location), NES counts firms with no paid employees, which are single-location by construction. The “nonemployer share” is nonemployer firms divided by nonemployer firms plus employer establishments. A multi-establishment employer counts several times in the denominator. NES is built from tax records, covers businesses with receipts of $1,000 or more, and CBP excludes some industries (none in NAICS 316) and most government employees (Census Bureau, n.d.-a). I compare leather with all manufacturing (31-33) and all industries as reference series.

2.3 Classification changes

CBP and NES are published under successive NAICS vintages. I assumed the standard five-year cycle with each revision first used in its own reference year (2002 vintage to 2007 reference year; 2007 vintage for 2008–2011; 2012 for 2012–2016; 2017 for 2017–2021; 2022 for 2022 onward). I did not find a Census statement of this mapping on the pages I read, so it is an assumption. The 2012 change is visible in the data: the 6-digit codes for footwear (316211, 316212, 316213, 316214, 316219) become one code (316210), and the leather-goods codes other than handbags (316991, 316993, 316999) become 316998, while 316110 and 316992 persist (Table 5). I built a concordance on that basis and used it to aggregate the 6-digit employer data into four harmonised groups (tanning and finishing; footwear; women’s handbags and purses; other leather goods and luggage). The harmonised groups add to the 3-digit total in every year (largest absolute difference 0, results/check_harmonised_groups_sum.csv), and the 2011 and 2012 values of the combined codes are close (592 establishments for the three 2007-vintage codes in 2011, 604 for 316998 in 2012). For comparisons between CBP and NES I used the 3-digit total (316) and the 4-digit groups 3161, 3162 and 3169, which both programmes publish in every year. To see whether the series move unusually at the first year of a vintage, I compared the year-on-year log change in each series at 2008, 2012, 2017 and 2022 with the median and maximum absolute change in all other years from 2005 to 2023, the range common to both programmes; the NES 2024 value was excluded from this comparison so the nonemployer series was not given an extra data point the employer series lacks. This diagnostic is descriptive: it cannot distinguish a classification effect from a real change that happens to fall in the same year.

Table 5. Six-digit NAICS codes present in the CBP files for NAICS 316, by reference-year block, and the harmonised groups used here.

2.4 Noise infusion and suppression

Since reference year 2007, CBP applies noise infusion: establishment-level employment and payroll are multiplied by random factors before tabulation, and each published value carries a flag showing the size of the resulting change (G, under 2%; H, 2% to under 5%; J, 5% or more), with J cells withheld (“D”) before 2015 (Census Bureau, n.d.-b). Establishment counts are not noise-infused. From reference year 2017 a cell is published only when it has at least three establishments (Census Bureau, n.d.-b), and the files show the code “N” in the size-class count cells. I treated “N” as missing and did not find its meaning in the documentation I read. Where possible I avoided it: the 250-and-over class was derived as the total minus the six smaller classes, which are never “N” for the 316 total, and the classes sum to the establishment total in every year. The employment-size flag (EMPFLAG) in the 2004–2017 files marks 6-digit cells whose employment is withheld; I report how many cells carry it and do not use those cells. The 2014 all-industry row carries the code “r” in that field, whose meaning I did not find. In the NES files, the receipts column in 2008 and later carries a noise flag (G in every 316 row used here). I did not apply any correction for noise; I report bounds in the Results.

2.5 Analysis

All results are descriptive: levels, percent changes, compound annual growth rates (CAGR), percentage-point changes in the share, indices (2004 = 100), a counterfactual decomposition of the share change (holding the nonemployer count, or the employer count, at its 2004 value), and the size-class distribution. There are no standard errors because the files are complete counts (modulo noise and suppression), not samples. No hypothesis tests are reported. As a check on the extraction I recomputed from my all-industry series the Census Bureau’s published all-industry growth rates (Census Bureau, 2025) and compared them (Section 3.6).

2.6 Software, data and AI use

Analyses used Python 3.13 with pandas and NumPy; figures used matplotlib. The study was carried out with the assistance of Claude (Anthropic; model claude-sonnet-5-5) in November 2026, which retrieved the data in a browser, wrote and ran the analysis code, searched the literature and drafted the text under the author’s direction. Independent AI agents are to check every number against the result files and every citation against its source. The author reviewed the work and takes full responsibility for it.

3. Results

3.1 Employers shrank, nonemployers grew

From 2004 to 2023, NAICS 316 employer establishments fell from 1,475 to 986 (−33.2%; −2.1% a year), March employment from 43,335 to 22,970 (−47.0%; −3.3% a year) and nominal annual payroll from $1,306.6 million to $1,037.8 million (−20.6%) (Table 1, Table 2, Figure 1). Over the same years the nonemployer count rose from 4,426 to 5,855 (+32.3%; +1.5% a year) and nominal receipts from $158.9 million to $254.3 million (+60.1%). The latest NES year, 2024, shows 5,869 firms (+32.6% from 2004) and receipts of $245.0 million (+54.2%), so the nonemployer count was flat between 2023 and 2024 and receipts fell.

The path was uneven. Most of the loss in employer jobs came early: from 2004 to 2009 establishments fell 15.5%, employment 33.5% and payroll 31.9%, while nonemployer firms rose 7.5% and receipts fell 5.2% (the 2007 NES value is a derived lower bound and is not an endpoint). From 2009 to 2019 establishments fell a further 6.0% and employment 8.6%, while payroll rose 11.4%, nonemployer firms 8.3% and receipts 37.2%. From 2019 to 2023 establishments fell 15.8% and employment 12.8%, payroll rose 4.8%, nonemployer firms rose 13.6% and receipts 23.0%. Establishments fell 8.4% between 2022 and 2023 alone, against −0.4% in all manufacturing and +0.8% in all industries.

Relative to the reference series, leather employer establishments fell more than all manufacturing (−33.2% against −16.1%) and employment fell much more (−47.0% against −10.8%); employer payroll fell in nominal terms in leather (−20.6%) while it rose in manufacturing (+52.3%) and all industries (+124.6%). Nonemployer firms grew about as fast in leather as in manufacturing (+32.3% against +34.3%) and more slowly than across all industries (+55.9%). Mean employment per establishment fell from 29.4 to 23.3, and nominal payroll per job rose from $30,151 to $45,182. Nominal receipts per nonemployer firm rose from $35,902 to $43,440, which was 79% of the all-industry figure in 2004, 71% in 2012 and 75% in 2023.

Table 1. US leather and allied product manufacturing (NAICS 316): employer establishments, employment and payroll (County Business Patterns, CBP) and nonemployer firms and receipts (Nonemployer Statistics, NES), selected years. Dollar values are nominal. The nonemployer share is nonemployer firms divided by nonemployer firms plus employer establishments. NES 2024 values: 5,869 firms, $245.0 million in receipts.

Table 2. Change from 2004 to 2023 in leather and allied products, all manufacturing and all industries (percent; nonemployer share in percentage points). Dollar values are nominal.

3.2 Did the nonemployer share rise?

Yes, in the sense that the ratio rose. The nonemployer share of units in NAICS 316 went from 75.0% in 2004 to 85.6% in 2023, a gain of 10.6 percentage points, and it rose in 13 of the 19 year-on-year comparisons (including the derived 2007 value) (Figure 2A). The gain was in two steps. Between 2004 and 2012 the share rose 8.0 points (to 83.0%); between 2012 and 2021 it was flat (−0.3 points); between 2021 and 2023 it rose another 2.9 points to its maximum. The counterfactual decomposition shows that the fall in employer establishments alone, with nonemployers held at their 2004 count, would have raised the share by 6.8 points, and the rise in nonemployers alone, with employers held at their 2004 count, by 4.9 points; these are not additive.

Whether this is a “shift” specific to leather depends on the comparison. The share in all manufacturing rose by 11.7 points (47.2% to 58.8%), slightly more than in leather, with 6.5 points by 2012, 2.9 from 2012 to 2021 and 2.3 from 2021 to 2023; across all industries it rose by 5.9 points (72.5% to 78.4%). Leather starts from a higher level, so its share is bounded closer to 100%. The series show a rise in the nonemployer share in leather; they do not show that the rise is larger than in manufacturing as a whole, and the 2012–2021 plateau is a feature of leather that manufacturing does not share.

Within NAICS 316 (Table 3, Figure 2B), the share in footwear rose from 62.7% to 79.8% (+17.1 points; employer establishments −41.6%, nonemployer firms +37.5%), in tanning and finishing from 81.5% to 90.0% (employer −40.2%, nonemployer +22.9%) and in the group that includes luggage, handbags and other leather goods (3169) from 75.7% to 85.3% (employer −27.8%, nonemployer +35.0%). Employer employment fell in all three: −64.1% in tanning and finishing, −49.8% in footwear and −36.5% in 3169. Among the harmonised 6-digit groups, women’s handbags and purses (316992) went from 116 establishments in 2004 to 128 in 2011, 113 in 2012 and 70 in 2023.

Table 3. Employer establishments, nonemployer firms and nonemployer share by industry group within NAICS 316, 2004, 2012 and 2023. The groups are the 4-digit codes published in both programmes (both programmes publish the same 4-digit codes in every year).

3.3 Employer establishments by size

Table 4 and Figure 3 give the size distribution. In 2004, 62.0% of employer establishments had fewer than 10 employees (46.0% in the 1–4 class); in 2023, 65.1% (48.2% in the under-5 class). The share with fewer than 10 employees peaked at 70.0% in 2015 and has since fallen. The share with 20 or more employees fell from 24.7% to 20.1%, and the share with 100 or more from 6.8% to 6.0%. Every class lost establishments between 2004 and 2023. The under-5 class accounted for 41.7% of the net loss of 489 establishments, the 20–99 classes for 25.7%. In proportional terms the largest losses were at 250 or more employees (36 to 15; −58.3%) and 50–99 employees (−57.5%), and the smallest at 10–19 (−25.1%). In manufacturing as a whole the share of establishments with fewer than 10 employees was 53.1% in 2004 and 53.3% in 2023, and in all industries 73.4% and 73.0%; the leather share rose by 3.1 points. These shares describe counts, not employment: I extracted employment only for the total and 6-digit cells, not by size class.

Table 4. Employer establishments in NAICS 316 by employment-size class (employees in the pay period including 12 March), CBP. Counts and percent of establishments. *1-4 in the 2004-2016 files, “under 5” in later files; in every year the classes sum to the establishment total. 250+ is the total minus the smaller classes.

3.4 Classification breaks

Establishment counts show no unusual movement at three of the four vintage boundaries. The year-on-year log changes (×100) at 2008, 2012, 2017 and 2022 were −5.7%, −2.2%, −2.4% and −0.9% (log points ×100), against a median absolute change of 1.9% in other years and a maximum of 8.7% (2023). For employer employment the boundary changes were −11.3%, −5.9%, +6.0% and +4.0%, against a median of 5.0% and a maximum of 12.8% (2009). For nonemployer firms, the changes at 2012 (+9.2%) and 2022 (+9.5%) are as large as any in the series; the median absolute change in other years is 2.4% and the maximum 9.0% (2007, the derived value). The 2012 jump follows a rise of 5.2% in 2011 and sits in a recovery after 2009. The 2022 jump (9.9% as a simple percentage change) should be read with the all-manufacturing rise (+8.0%) and the all-industry rise (+4.7%) in the same year, which indicates a general movement in the NES that year as well as a larger one in leather. The data cannot say how much of these two jumps is reclassification and how much real change.

3.5 Noise and suppression

The 316 total carries a noise flag of G (change under 2%) on employment and payroll in all 17 years from 2007 to 2023 (the 2004–2006 files have no noise flags). If the 2023 employment value of 22,970 were off by 2% in either direction, the 2004–2023 change would lie between −48.1% and −45.9%, so noise does not affect the direction or approximate size of the decline. Employment is withheld for 10 of 128 6-digit employer cells, all in the 2004–2011 files, mostly 316212 (house slippers, in seven of the eight years), with 316219 in two years and 316992 in 2008. In 2017–2023 between 23 and 27 size-class cells in the 316 rows are “N”. All 4-digit and 3-digit totals used for the main series are published. The NES receipts for 316 carry the “G” noise flag in every year from 2008 (no flags in 2004–2006). The derived NES 2007 firm count (5,148) is above the mean of its neighbours (4,782), consistent with the all-industry and manufacturing counts also peaking in 2007 (Figure 1).

3.6 Cross-check against Census Bureau statements

From the all-industry series in the package, the nonemployer count grew at 2.68% a year from 2012 to 2023 and the employer establishment count at 1.08% a year, and nonemployer growth was 4.88% in 2021 and 4.68% in 2022. These agree with the 2.7%, 1.1%, 4.9% and 4.7% reported by the Census Bureau (2025), which supports the extraction. For leather the corresponding annual rates from 2012 to 2023 are +0.42% (nonemployer firms) and −1.33% (employer establishments).

4. Discussion

4.1 What the series show

Between 2004 and 2023 the employer part of US leather and allied product manufacturing contracted by a third in establishments and by nearly half in employment, while the count of firms with no paid employees grew by a third. The nonemployer share rose by 10.6 points. Two features limit how this should be read. First, the rise in share is about the same as in all manufacturing and is therefore not evidence that leather is special; it is consistent with a broad shift of small businesses outside payroll that is also visible across all industries (Census Bureau, 2025). Second, the share did not rise steadily. The path is a rise to 2012, a plateau, and a rise in 2022–2023, and the second rise is accompanied by a one-year fall of 8.4% in employer establishments. A single-year movement of that size could be a real contraction or a feature of the 2023 data release, and I cannot tell which from these files alone.

The employer decline is concentrated in the largest and smallest establishments in different ways. Establishments with 50 or more employees fell the most proportionally; the under-5 class fell the most in number because it is the largest class. The share of establishments with fewer than 10 employees rose over the whole period but peaked in 2015. Both patterns are compatible with several mechanisms (closures of larger plants, downsizing of survivors, firms moving down a size class), which this dataset cannot separate because it counts establishments in each year, not their histories. A longitudinal source of the kind used by Davis et al. (2007) would be needed.

4.2 Employers and nonemployers are not the same firms

An increase in nonemployers and a decrease in employers does not mean that employers became nonemployers. The two counts come from different sources and different units, and no firm is followed from one to the other here. Part of the growth in nonemployers is likely to be new activity unrelated to former employers (the Conroy et al. (2026) finding that growth was largest in lower-income counties, as reported in their abstract, points to differing circumstances behind the aggregate); a one-person workshop selling leather goods online is in the data as a nonemployer, and nothing in the files distinguishes this from a retired employer’s sideline. The paper therefore claims only that the ratio rose.

4.3 Limitations

1. Establishments are not firms: the denominator of the share counts multi-establishment employers more than once, which makes the nonemployer share lower than it would be on a firm basis. This bias is probably small in leather but I have not measured it. (2) The NAICS vintage mapping is an assumption, and the 2012 and 2022 jumps in the nonemployer count may partly reflect classification. (3) The NES 2007 value is derived from the state file, with two states withheld, and the national 2007 file could not be retrieved. (4) Dollar values are nominal; changes in payroll and receipts include inflation. (5) The size-class analysis uses counts only. (6) Noise infusion changes employment and payroll values by amounts the flags bound only loosely (G is under 2%), and the post-2017 “N” codes are not documented in the pages I read. (7) NES depends on tax records, and the Census Bureau’s coverage and methods for nonemployers have their own limits that I have not examined. (8) CBP 2023 and NES 2024 differ in their last year, so the share series ends in 2023. (9) I did not examine whether the 2022 NES release, which also shows a large all-industry jump, involved changes in method. (10) The CBP directory listing was rechecked on 2 November 2026 and still shows no 2024 subdirectory, so the employer series still ends in 2023 as of this writing.

4.4 Use of the dataset

The dataset (Section 6) is small by design. It may serve as a template for any other 3-digit industry: the same extraction applies with a different NAICS prefix. The cross-check against published Census growth rates shows that the extracted all-industry values match the Census narrative figures, which gives some assurance that the extraction is right.

5. Conclusion

In national Census data, US leather and allied product manufacturing lost a third of its employer establishments and nearly half of its employer employment between 2004 and 2023, while the number of nonemployer firms grew by a third and the nonemployer share rose from 75% to 86%. The rise in share was in two steps rather than a steady drift, and a similar rise occurred in all manufacturing, so the data support the claim that the ratio shifted but not that leather is unusual. The larger employer establishments declined proportionally more. The open dataset and code allow others to extend the series, test the vintage assumptions and link them to other sources.

Related on x9m8

• What chrome buys: a meta-analysis of paired chrome-free and chrome leather tests
• Leather’s carbon bill follows the hide price, not the cow

Acknowledgements

I used Claude (Anthropic; model claude-sonnet-5-5) in November 2026 to retrieve the public files, write and run the code, search and check the literature, and draft the text. I reviewed the output and am responsible for it. I thank the Census Bureau for publishing the files.

Author contributions

Peter Kalafatis: conceptualisation, supervision, review and responsibility for the work. AI assistance as described in Section 2.6.

Financial support

This research received no specific grant from any funding agency, commercial or not-for-profit sector.

Conflicts of interest

The author owns Marcellino NY, which makes leather briefcases and buys finished leather.

Research transparency and reproducibility

All data are public files published by the US Census Bureau, which are works of the US government and not subject to copyright in the United States. The extracted NAICS 316, manufacturing and all-industry rows, the derived series and the source-file register (URL, size, SHA-256 hash, last-modified header, retrieval date) are in the reproducibility package linked from the article page and are released into the public domain (CC0 1.0). Code (Python 3.13) regenerates every result, table and figure from the extracted files in data/raw. The full raw Census files are not redistributed; their URLs and hashes are listed. The study was not preregistered; it is a descriptive analysis of public data.

References

Autor, D. H., Dorn, D., & Hanson, G. H. (2013). The China syndrome: Local labor market effects of import competition in the United States. American Economic Review, 103(6), 2121–2168. https://doi.org/10.1257/aer.103.6.2121

Census Bureau. (n.d.-a). About County Business Patterns. U.S. Department of Commerce. https://www.census.gov/programs-surveys/cbp/about.html (retrieved 2 November 2026)

Census Bureau. (n.d.-b). County Business Patterns: Methodology. U.S. Department of Commerce. https://www.census.gov/programs-surveys/cbp/technical-documentation/methodology.html (retrieved 2 November 2026)

Census Bureau. (2025, July 30). Number of U.S. nonemployers grew faster than employer businesses nearly every year from 2012 to 2023 [America Counts story by T. Shoemaker]. U.S. Department of Commerce. https://census.gov/library/stories/2025/07/nonemployer-business-growth.html

Conroy, T., Landini, A., & McDermott, M. (2026). The rise of nonemployer businesses in the United States: An analysis by geography. Small Business Economics, 66(3), 1263–1288. https://doi.org/10.1007/s11187-025-01141-z

Davis, S., Haltiwanger, J., Jarmin, R., Krizan, C. J., Miranda, J., Nucci, A., & Sandusky, K. (2007). Measuring the dynamics of young and small businesses: Integrating the employer and nonemployer universes (NBER Working Paper No. 13226). National Bureau of Economic Research. https://doi.org/10.3386/w13226

Haltiwanger, J., Jarmin, R. S., & Miranda, J. (2013). Who creates jobs? Small versus large versus young. Review of Economics and Statistics, 95(2), 347–361. https://doi.org/10.1162/rest_a_00288

Hurst, E., & Pugsley, B. W. (2011). What do small businesses do? Brookings Papers on Economic Activity, 2011(2), 73–118. https://doi.org/10.1353/eca.2011.0017

Katz, L. F., & Krueger, A. B. (2019). The rise and nature of alternative work arrangements in the United States, 1995–2015. ILR Review, 72(2), 382–416. https://doi.org/10.1177/0019793918820008