Methodological Appendix


Table of Contents

1. Union Membership

2. Industry

3. Census Data

4. Union Density by Decade

5. Strike and Work Stoppage Data

6. National Labor Relations Board Data


1.  Union Membership

We compile yearly data on union memberships, from 1900 to 2015. Following Bureau of Labor Statistics (BLS) reports, we use the requirement of collective bargaining in more than one state as our definition of a national or international union, and we exclude unions that exist for less than five years. We collect union membership data from BLS reports, compilations from other scholars, and union reports. We exclude memberships from outside of the United States. We then aggregate union memberships to the industry level, using industry data from union reports to categorize unions by industry, detailed below.

The major advantage to this approach is that it allows us to assemble a comprehensive, uninterrupted data set of union membership for well over a century. A popular alternative approach to estimate union membership is to use survey data. The main survey used to track union membership is the Current Population Survey (CPS), which is a monthly household survey that asks respondents about their union status in their primary job. The CPS surveys households, asking the “reference” person of the household questions about the work, pay, and union status for the primary jobs of all household members over fifteen years of age. Aggregated data begin in 1973; estimates by state, detailed industry, and occupation begin in 1983. CPS-generated data are commonly used by scholars studying the most recent decades; see Hirsch and MacPherson (2003). For this study, the major limitation of CPS data is that they only document the dis-regulated era.

Union-reported memberships provide the best approach to examining a long arc of U.S. labor dynamics. Early labor scholar Leo Wolman (1936) explains that union membership size is our best estimate of organized labor’s strength and the most reliable and continuous source of its measurement. These membership numbers paint a broad picture of the ebbs and flows of unionization. However, they are just that—broad. They do not capture precise changes in union membership year by year. Several points should be taken into consideration when using this approach. Most importantly, although most unions have collected and reported their membership numbers, they vary in their policy and practice of arriving at them. We aim to use sources that report per capita, dues-paying members. When reports are based on union leaders’ statements, they may be unreliable. Over the century, unions have generally developed more bureaucratic structures that help to run their financial affairs and since the Landrum-Griffin Act of 1959, they are not only subject to periodic audits but also to government reporting. But especially in the early periods and for small unions, membership figures may not exist. Another issue is that we sometimes find large fluctuations in year-to-year membership, which may indicate an error or may truly reflect reality. An example of a factual large yearly membership change is when a substantial strike effort inspires large numbers of workers to sign union membership applications, hence driving up union membership numbers. Upon completion of the strike, especially if it failed to accomplish the desired goals, many of those workers may not pay the deferred initiation fees and union dues, thereby driving the union membership numbers back down. Some unions report little variation in membership and round their numbers when they report members (e.g., reporting 50,000 members for four years in a row). This suggests that membership numbers are estimates, rather than precisely calculated. Further, reports based on union dues payment reflect fluctuations in employment, unemployment, and strikes, as workers who have been laid off, who are underemployed, or who are on strike usually don’t pay dues.

Despite these challenges, relying on figures reported by the major union federations, which, during some periods, account for the vast majority of all union members, overcomes some of these difficulties. Their reports, based on per capita tax paid by affiliated unions, provide “conservative and comparable” (Wolman 1936, 9) membership estimates.

It’s important to note that membership numbers do not capture the full strength of the labor movement. Unions may yield political and economic power and accomplish goals that are not captured by merely reporting membership size. Still, for early years, “the data of membership appear in the long run most faithfully to represent the state of unionism” (Wolman 1936, 4).

There may be disagreement about which estimates most accurately represent union memberships. However, because our strategy requires consistency over the 115 years we study, the best option for this study is per capita membership reports by unions.

Below we list the sources that we consulted to generate yearly union data. In some cases, sources conflicted. Our strategy was to consistently use the same source over multiple years, only drawing on other sources when information was missing or if there was major disagreement between sources. We prioritize sources that use dues-paying members.


Wolman, Leo (1936): Our main source 1900-1934

The main source we use for the unregulated period is Ebb and Flow in Trade Unionism by Leo Wolman (1936), who compiled union-level membership data from 1900 to 1934. Wolman does not specifically articulate his definition of a union. Data are drawn from annual reports of the Executive Council of the AFL, with supplements from individual unions (1936, 14). We compare Wolman’s numbers to those of other sources, which typically cover only a handful of years from that period, including BLS reports starting in 1926.


Troy and Sheflin (1985) and Troy (1965): Our main source 1935-1947

Leo Troy (1965) collected union-level data for 1935–1962. Troy defines membership as dues-paying members and draws on financial reports to federations. When unavailable, he supplemented with other sources, such as letters from unions, reports from union conventions, BLS directories, and monographs of particular unions.

Troy and Sheflin expand and improve on Troy 1965, but they provide less information on small unions. In Union Sourcebook they provide membership based on average annual dues-paying, full-time equivalent membership for every known national and not small union between 1897 and 1983. They derive their membership numbers mainly from financial information on unions derived under the Labor Management Reporting and Disclosure Act of 1959, the Civil Service Reform Act, and annual reports to the Department of Labor’s Office of Labor-Management Standards Enforcement. When necessary, they supplemented missing information with union-supplied information (8 percent of the total) or extrapolation (6 percent of the total). For the former, the annual per capita revenues from local unions is divided by the union’s per capita rate to arrive at the annual membership.

We use Troy and Sheflin (1985) and, if not available, Troy (1965) for membership between 1935 and 1947 (after Wolman [1936] ends and before the BLS begins biannual reporting). Because this is an update on Troy (1965), we use this as our first option; we use Troy (1965) only when when information is missing. We also use Troy/Troy and Sheflin to fill in missing data from Wolman or when the BLS does not record a specific union. We also use it for reference and to provide a check on our own figures.


Bureau of Labor Statistics Reports: Our main source 1948-1979

The BLS reported union membership beginning in 1926, and biannually since 1948. From 1948 to 1980 these BLS reports are our main source for membership.

The earliest reports went by the name Handbook of American Trade-Unions (1926, 1929, 1936). The 1926 edition was an effort to list all of the existing labor organizations of the United States having national entity and significance. The 1929 edition “deals only with bona fide labor organizations functioning nationally in June 1929, and disregards entirely unions which are purely local in character, works councils, and those organizations which are or may be fairly regarded as company unions” (1). The 1936 edition uses membership reported to the BLS. Note that these definitions include Trade Union Unity League unions, which Wolman reports separately. We use these early BLS reports for reference, comparing them to Wolman, Troy, and Troy and Sheflin.

Beginning in 1943, the BLS generated a bi-yearly report of union memberships called the Directory of U.S. Labor Organizations. Earlier years list unions and headquarters, but not membership. Beginning in 1948, the BLS compiled union membership for all unions affiliated with the AFL, CIO, AFL-CIO, and independent unions that had negotiated contracts in more than one state. They state that “All national and international unions known to the Bureau of Labor Statistics are included.”

There are some variations in how the BLS collected this data over time. From 1948 to 1954, unions were asked to report their annual average dues-paying membership or membership in good standing. If the union organization did not report membership, it was left blank. However, for nonreporting AFL, CIO, and AFL-CIO affiliates, paid per capita membership was used (noted by asterisks in source).

From 1955 to 1979, the series included all unions affiliated with the AFL-CIO and independent unions that had negotiated collective bargaining contracts with different employers in more than one state or, for federal government unions, those that had exclusive bargaining rights under Executive Order 11491. During these years, the BLS created a special questionnaire for union reporting of membership that attempted to standardize, as much as possible, union reports of membership numbers. The questionnaire requested the average number of dues-paying members during the first six months of the year (BLS 1950, 2). Later years asked unions to report average annual dues-paying membership for the two previous years (BLS 1979, 20). For nonreporting AFL-CIO affiliated unions, paid per capita membership as reported in the AFL-CIO convention proceedings was used.


Bureau of National Affairs Directories, Courtney Gifford, Ed.: Our main source 1982-2015

From 1982 to 2015 the BLS’s Directory of U.S. Labor Organizations series continued under a private publisher, Bloomberg’s Bureau of National Affairs with Courtney Gifford as editor. The first edition states that the long form list was “prepared by the Bureau of Labor Statistics’ Division of Industrial Relations, with updates by the editor to reflect subsequent mergers. The basic BLS requirement for inclusion in the list is affiliation with the AFL-CIO or, for unaffiliated unions, the existence of collective bargaining agreements with different employers in more than one state.” Gifford includes professional and state associations that engaged in collective bargaining (or representational activities) and claimed membership in more than one state. Gifford also includes organizations operating in more than one city of a state. We, however, exclude unions operating in only one state. Many of the editions include extensive notes regarding reporting, which we take into account.


Other Sources

Some small unions have missing membership information for some years. When our main sources did not provide membership information, we consulted other secondary sources, including Craft and Peck (1998), Spomer (1992), Fink (1977), and Hannan (1988). We also consulted union convention proceedings, websites, or other internal publications to estimate membership.


On Missing Data

Using the sources described above, we were able to compile membership data for all but a handful of small unions. Some small unions continued to have missing data for a series of years, typically at their founding or at the time of merger or death. When this was the case, we estimated membership numbers on a case-by-case basis, given what we know. Typically, we assume that total membership is smaller at the founding date and subsequently grows. Likewise, we assume that their membership declines as unions approach organizational death. A handful of small, usually short-lived unions have no available data. We assign them a small membership of 5,000 workers. These adjustments are our best estimates of likely memberships. Because they only pertain to a handful of small unions (large unions have better recordings), these adjustments do not influence the broad patterns that we report.


Accounting for Non-U.S. members

Because we are interested in U.S. union density, we exclude international members from union membership. Some unions had sizeable numbers of Canadian members, and occasionally members in other countries or territories. Most unions, however, were almost exclusively populated by members working in the United States. We use a series of sources that report both total members and U.S. members: Wolman (1920), BLS (1954), BLS (1966), BLS (1978), Gifford (1992), and Gifford (2002). When we lack information on U.S. membership for unions with international membership, we calculate the percent U.S. members for the dates for which we have information, and then interpolate the nearest known percent between data points.

We do not have international membership estimates prior to 1920 for individual unions. However, Troy (1965) estimates that the total percent of union members that were Canadian in 1900 and 1910 as 1.8 percent. That percent rises to 3.4 percent in 1920. To estimate memberships in years prior to 1920, we take the percent of international members for each union in 1920 and reduce it by the aggregate percentage decline that Troy estimated. This is a rough approximation, given union variation in the percentage of non-U.S. members.

The last data point we have for international union membership is in 2002. We identified unions that were large (over 50,000 members) and had at least 10 percent Canadian members in 2002. For each, we consulted union websites in 2019 to gather the percent of members outside of the United States. We could not find information on a handful of unions, most notably the Amalgamated Transit Union, the Boilermakers, and the Bakery, Confectionary, Tobacco Workers and Grain Millers International Union. If data were missing, we used the last known percentage of international members, which we then subtracted from the total membership. Finally, if no sources indicated non-U.S. members for a union, we coded it as all U.S. workers.



2. Industry

We focus on eleven broad industries. Our long time period prevents us from disaggregating industries further because long-term data sources change how they define subindustries. Our broad industry categories allow us to be relatively consistent over time. We do not present data on all industries, as some have little relevance to our analyses. For example, we do not present information on business service, other professional service, or industry unknown. We combine some industries if that was the standard practice for the major reports that we use from the BLS. The industries are as follows:

1.         Agriculture, forestry, and fisheries. We call this category “agriculture.”

2.         Mining. This includes metal, coal, and nonmetallic mining, and crude petroleum and natural gas extraction.

3.         Construction. Also called building trades.

4.         Manufacturing. We combine durable and nondurable manufacturing for this category.

5.         Transportation, communication, and utilities. We abbreviate this to “TCU.” We combine these three categories because they are frequently combined on BLS reports. Transportation captures multiple categories, including rail, air, buses, and trucking. Communication and utilities include categories such as telephone, sanitary services, water supply, and electric and gas. TCU also includes information and warehousing.

6.         Trade. The trade category combines wholesale and retail trade. This category includes everything from food stores to the sale of motor vehicles.

7.         FIRE. This is finance, insurance, and real estate, and includes banking.

8.         Personal service. This includes categories such as hotels and lodging places, laundering, cleaning, barber and beauty shops, funeral services, dressmaking, shoe repair, and private households. This category was regularly used during the mid-twentieth century, including the 1950 and 1990 census industry classification. However, as the service sector expanded, some agencies began to report their statistics for smaller or larger subsets of this industry. We try to be as consistent as possible, but note that this category became more fluid during the dis-regulated period.

9.         Entertainment. The subsectors of this category includes entertainment, arts, sports, museums, and recreation, which we refer to generally as “entertainment.”

10.       Health services. This includes hospitals, nursing and residential care facilities, other health services, and social assistance.

11.       Public administration and education. We combined public administration and education because other data sources did not always disaggregate these categories. Almost all workers in these categories are in the public sector. However, other industries may also include public sector workers (e.g., public hospitals). Public administration includes federal, state, and local administration as well as postal service. Education includes elementary and secondary schools, colleges and universities, and vocational schools.

We aggregate individual union membership to the industry level. Some unions have substantial memberships in multiple industries, especially after the AFL-CIO merger in 1955 and into the dis-regulated period as unions looked for alternative strategies to defend or expand membership. To account for this, we develop a coding scheme to estimate the percent of members in each industry. We use the best data available; however, the calculations are not exact.

Our baseline data on industries are from the Hannan (1988) data set. This publicly available data set provides information on the main industry that a union represented at the time of the union’s founding. For unions that were founded after the Hannan data set, we code industries based on the main membership of the union.

We then identify unions that have grown to have substantial numbers of members in multiple industries. Little systematic data exists on this topic, so we aim to estimate union memberships by industries for large unions, and those with substantial memberships in at least two industries. We focus on the period after the AFL-CIO merger in 1955, when unions increasingly begin to organize in multiple industries. We use the AFL-CIO reports which list union memberships by industry in the 1960s and 1970s, and we are able to compare to BLS industry estimates. We access secondary sources for unions unaffiliated with the AFL-CIO and for later years. We also check qualitative descriptions provided by Peterson (1945) and Craft and Peck (1998). Many unions fall primarily in one industry category, though it is common for large unions to have members across multiple industries. See Appendix A for a description of how unions managed jurisdictions over the century.

We identify the following unions as large and having substantial numbers of members in multiple industries: Service Employees International Union (SEIU), Teamsters (IBT), American Federation of State, County, and Municipal Employees (AFSCME), United Auto Workers (UAW), International Association of Machinists (IAM), Communication Workers of America (CWA), International Brotherhood of Electrical Workers, United Brotherhood of Carpenters, United Food and Commercial Workers (UFCW), UNITE HERE, and the Laborers International Union of North America (LIUNA). 

For these unions, we aim to code for membership in each industry. This type of information is not well documented. We are able to code for three time points: 1962 or 1969; 1979; and around 2015. Data for 1969 (occasionally 1962) and 1979 have similar formats and are found in “Union Membership and Employment 1969–1979” prepared by the AFL-CIO Department of Research February (1980). These reports list union membership in each industry for unions affiliated with the AFL-CIO. We code the first and last year available: 1969 and 1979.

We supplement this with BLS data on union membership by industry, available yearly from 1962 to 1978. The strengths of the BLS data are that they cover some independent unions and have an earlier beginning time point. However,, because there are large amounts of missing data, we do not use it as our main source. When they report information on the same unions, the AFL-CIO and BLS publications report similar statistics.

Five AFL-CIO affiliates had substantial membership in multiple industries in the earliest time period for which we have AFL-CIO estimates (either 1962 or 1969): SEIU, Machinists, Laborers, IBEW, and Carpenters. To determine membership by industry prior to this period, we do the following. In 1955, we assign each union to 100 percent of the industry indicated by Hannan (1988) as the union’s main industry. Then from the earliest data point (1962 or 1969) on secondary (and other) industries, we extrapolate down to 1955. The AFL and CIO merged in 1955, and this marks a time of many mergers between international unions.

Because they were independent at the time, the Teamsters and the UAW are not included in the AFL-CIO report by industry. To address this, we assign 100 percent of UAW members to manufacturing until the mid-1990s, which is when UAW began to expand to other industries, such as education. We take the proportion of UAW members in each industry in 2015 (as reported on their website) and extrapolate down to 0 in 1995 for all industries, except for manufacturing, which we designate 100 percent of UAW members up until 1995.

The Teamsters has a long history of organizing in multiple industries. They provide little systematic data to the public. However, the Teamsters’ website accessed in 2017 breaks down membership by industry. At that point, over 50 percent were in transportation/communication, roughly 25 percent were in manufacturing, and the rest were scattered across various industries. We extend these percentages back to 1969. Because there is no systematic data on the Teamsters, we use secondary sources to roughly estimate industry memberships back to the early 1900s.

Beyond the earlier AFL-CIO reports on industry breakdowns, there is no systematic information on union membership by industry in the 2000s. Therefore, in 2017 we examined union websites and/or directly corresponded with the unions to obtain this information. These numbers reflect membership in the previous one or two years (2016 or 2015). There is likely little difference between 2015 and 2016. We assign these estimates to 2015, which is our last year of data for the project. We were unable to obtain estimates of membership by industry for several unions: the Carpenters, UFCW, and Laborers. To address this, we took the same percentage breakdown by industry that the union listed in 1979 for those three unions.

The merger between UNITE and HERE in 2004 requires additional attention. At the time of merger in 2004, HERE had 239,041 and UNITE had 203,411 members. The 2017 website reports UNITE-HERE membership at 270,000, but members of the divisions have members totaling 465,000 (100,000 in each of the following: Hotels, Food Service, and Gaming; 45,000 in Airports; 20,000 in Transportation, but the last number is from when the union constituting this division previously merged with HERE). UNITE affiliates did not report membership. Using the fact that UNITE and HERE represented approximately the same number of members at the merger date, we attribute 100,000 to Textile. Using the total of the reported members in the industries (465,000) as the base, we estimate the proportions as follows: Hotels, 21.5 percent; Airports, 9.7 percent; Food Service, 21.5 percent; Gaming, 21.5 percent; Textile, Manufacturing, and Distribution, 21.5 percent; Transportation, 4.3 percent.

Finally, the AFL-CIO data from 1969 and 1979 only list “services” generally. We are interested in more detailed measures of service, including trade, health services, education, personal services, and entertainment. However, Unions that list their membership by industry on their websites tend to distinguish the type of service work. We extend 2015 estimates of various service types to the earlier time points. Again, we do not have website information for the Carpenters, UFCW, and Laborers. Of these, only the UFCW has substantial service sector members, which we designate as wholesale/retail trade services.



3. Census data: employment, race, gender, and occupation

Employment data are drawn from the Integrated Public Use Microdata Series (IPUMS) (Ruggles and Sobek 1997), which is composed of census samples. The census provides the most consistent source of quantitative data on employment, with industry information beginning in 1910. The census also has information by industry on other key variables, including occupation, race, and gender.

One challenge to using employment data over a long time period is that the questions and sampling design of the U.S. Census changed over time. We follow the methodology established by Sobek (2001), who worked with the census to construct a method to create consistent employment data over time by industry. Sobek does the following. Employment numbers are limited to employed non-institutionalized civilians age sixteen or older in the labor force (1940–2015) or with an occupation (1910–1920). Industry classifications were introduced to the census in 1910. The IPUMS infers industries from occupations prior to 1910, and therefore industries are not comparable prior to 1910. All of our industry data, therefore, start in 1910. Industries are based on the 1950 industrial classification system. Likewise, occupations are based on the 1950 occupational classifications. It is not possible to determine employment status prior to 1940 and therefore we include only the employed. From 1940 and on, occupation statistics are limited to currently employed people, which matches the criteria set by the modern BLS. Sobek also details the participation of females. Due to the wording of the census, scholars agree that the estimated female participation rate is artificially low prior to 1940. However, scholars disagree over the magnitude of the effect. Sobek does not modify numbers of females in the labor force prior to 1940. It is also possible that the 1910 census overcounted women. Sobek presents 1910 data but cautions users of potential issues.

We opt to use the decennial census because it covers the longest period of time with key indicators on race, gender, occupation, and industry. While we lose some yearly precision with this approach, we gain consistency in measurement over time.

Three main alternative measures provide employment by industry. Each has limitations, and none cover the full century. We compare our census-based method to these alternatives. There are differences in employment numbers between all of these measures, which the Bureau of Economic Analyses (BEA) describes. However, these data sets are similar in their general employment trends.

First, the Bureau of Economic Analyses provides yearly employment by industry beginning in 1929. The early decades do not include key industries for our analyses, including education, health, entertainment, and personal services. It also does not provide demographic data by industry. It does, however, include information on wages, corporate profits, hours worked, and other industry-level information. The BEA makes adjustments to account for employment not fully covered by the state unemployment insurance programs. Still, its employment counts differ from others.

Second, the Current Employment Statistics (CES) is an office of the BLS. It includes employment data for non-farm industries. Broad, aggregate measures begin in 1939; more detailed data are available toward the end of the twentieth century. The CES Program is a federal-state cooperative program. They write “The CES survey is based on approximately 149,000 businesses and government agencies representing approximately 651,000 worksites throughout the United States.” One downside is that the CES industry groupings are aggregated in early decades (e.g., education and health are lumped together) and agriculture is excluded. It also does not include demographic information on workers by industry. The CEW is commonly used by economists and policymakers to track shifts in employment. Its employment figures are sometimes higher or lower than estimates from the census or the Current Population Survey (CPS).

Third, the CPS is a monthly survey of households conducted by the Bureau of Census for the BLS, with industry-specific questions beginning in 1983. CPS data are the most common way that scholars calculate union density (using both its union and employment information to calculate density). The CPS uses wage and salary workers as its base and therefore does not include independent contractors. Because the CPS only began in 1983, it not suitable for this project. We do use it as a check by comparing CPS union density and employment numbers to our estimates.



4. Union Density by Decade

Union density is the ratio of union membership to employment. We are interested in broad patterns of density over time. Because we use census data for employment, we typically have employment on the decennial, or every ten years. Therefore, we report estimates of industry union density by decade. This approach is also consistent with the nature of our membership estimates, which paint a broad picture of ebbs and flows of unionization rather than a precise accounting by year. We average membership for each decade, which we then divide by average employment. Because our analysis does not end on an even decade, we average 2000–2015 for our last time point.

We compare our estimates to other research which presents industry-level union density, including Wolman (1936) and through our own estimates of CPS data starting in 1983. There are some differences, which stem from the different numerators (the union membership data source) and the denominator (the employment source).

Union-reported memberships differ from CPS survey estimates of union membership. Union reported memberships (based on per capita dues) tend to be higher than CPS survey indicators (based on household reporting of the union status for household members’ primary jobs). For example, using the CPS survey method, Hirsch and Macpherson report around 14.8 million union members and 16.4 million workers covered by union contracts in 2015. Using our methods (collecting membership numbers for all unions and then summing them for an aggregate number), we find around 18.8 million members. This number is smaller than what the major federations report themselves on their websites: the AFL-CIO reports around 12.5 million members, CTW around 4.5 million, and the National Education Association (NEA), a large independent union, reports close to 3 million members—for a total of around 20 million union workers.

These differences in union density estimates are most apparent in industry-level density variation by source. The time trends within industry are similar across sources. For example, in the dis-regulated era, our approach (membership reports divided by census employment, averages over ten years) produces higher-density numbers for manufacturing industry than the CPS approach (household surveys). These differences could stem from a number of factors, including how unions reported their membership by industry (which is how we constructed union member by industry), and how CPS respondents reported their family members’ union status and industry. However, both approaches show similar trends within manufacturing: union density declines within manufacturing during the dis-regulated period.



5. Strike and Work Stoppage Data

There is no data set on strikes that covers a long period of time. We draw on multiple sources, with varying definitions, methods, and periods covered.


Peterson, Florence. 1937. “Strikes in the United States, 1880-1936” Bureau of Labor Statistics, Bulletin No. 651.

The federal government began collecting data on strikes in conjunction with the 1880 census. This report contains major statistical data on strikes and lockouts from the earliest recorded date (1880) through 1936. Between 1881 and 1905, it conducted “periodic surveys” of strikes, but it collected less data from 1906 to 1913. In 1914, the BLS began its regular reporting of the number of strikes. In 1927, it made another major advance, by including “man-days” idle, detailed industry descriptions, and more information on strike outcomes. As of 1938, the Bureau’s information on strikes and lockouts came from “700 daily, weekly and monthly papers and journals” and reports from government labor boards (Peterson 1937, 170). BLS staff members followed up on these newspaper reports with questionnaires to firms and unions engaged in the disputes in order to gather more systematic information. In her 1938 publication, the BLS’s Florence Peterson stated that “it is believed that few, if any, strikes escape the Bureau’s attention.”

Prior to 1922, the BLS tracked strikes and lockouts separately. After, the BLS did not distinguish between the two, stating that the “industrial causes and effects are practically the same” and that the number of lockouts was relatively insignificant, comprising only 1–2 percent of all disputes (p. 4). All labor disputes were referred to with the generic term “strike” beginning in 1935. The BLS reports information on strikes that involve at least six workers and that lasted at least one day. Strikes statistics include number of strikes and number of workers involved. There is some information on a range of issues, including cause, outcome, sex of workers, establishment size, and state. Industry-level information is sparse from 1900 to 1926. Beginning in 1927, data collection was improved, including more systematic data by industry.


American Federation of Labor. Reports of Conventions Proceedings. Yearly 1900-1926.

The AFL convention reports from 1900 to 1921 contain details on member unions, including number of charters, membership, and number of strikes won, compromised, and lost. We use the strike information to estimate AFL strikes by industry and their success.


IWW History Project: Industrial Workers of the World 1905-1935. University of Washington.

Co-directed by Professor James Gregory and Labor Archivist Conor Casey from the University of Washington, this project explores the history of the IWW. It includes a yearbook of IWW strikes and other events. Specifically, it identifies nearly four hundred strikes that took place between 1905 and 1920 which involved IWW presence, but not systematically by industry or number of workers involved. Some of these strikes may be “double counted” with the AFL strikes numbers, since some of these strikes also involved members affiliated with the AFL.


Bureau of Labor Statistics. Analysis of Strikes and Lockouts (various name changes). Yearly 1934-1980 (excludes 1935).

These yearly reports compile data on all work stoppages, and data are reported by industry, not by union. All of these reports combine strikes and lockouts, and depending on the year they use the terms “work stoppages,” “strikes and lockouts,” or “strikes” (all of which refer to the same phenomena—we use the term “strikes”). Strikes are defined as involving at least six workers and lasting at least one day. Smaller or shorter strikes are not counted. The BLS collects information on strikes principally through the news, union publications, and trade journals. For some years when the BLS received notice of a strike, the BLS sent report forms to representatives of employers and unions to receive more detailed information. It is possible that some strikes escaped attention, especially small, short strikes or those with no union representation.

The reports include both aggregate data on national trends and data by industry. Strike statistics include the number of strikes, number of workers involved, and days idle. Some years also contain more detailed information, including reason for the stoppage and contract status (e.g., was the strike issue over a first contract, etc.). These reports do not systematically report strike outcomes for most strikes. However, they do include additional information on very large strikes, those involving over 10,000 workers. For these very large strikes, the reports include information on the major terms of settlement, which we code to determine general outcome (predominantly a win, compromise, or loss).

The BLS delivered its historical files to the National Archives, which now holds several electronic historical work stoppages files: 1953–1977, with information on number of workers involved, state, SIC, major issue, contract status, beginning and ending dates, and days idle. The 1978–1981 file adds several fields, including union and employer involved, violence, mediation, and settlement. The yearly strike reports have been digitized and are now available online.


Bureau of Labor Statistics. Major Work Stoppages Data.

The BLS ceased publishing work stoppage reports in 1981. However, it did continue to publish information on “major” work stoppages, which it defines as involving over 1,000 workers and lasting at least one full shift. The BLS obtains information on strikes from major media sources, reports from the Federal Mediation and Conciliation Service (FMCS), state labor market information offices, and BLS Strike Reports from the Office of Employment and Unemployment Statistics. We use FMCS data for our main strike analysis since it includes smaller strikes.

The BLS also reported information on large strikes, or those involving over 10,000 workers. The BLS only describes the major terms of settlement up until 1980. Since the outcome was not reported in later years, we code the outcome for strikes involving 10,000 workers. After 1980, we construct the list of very large strikes (10,000 or more) from the BLS publication (U.S. BLS 1993) on major work stoppages. We consult news reports and union publications to code for whether these strikes resulted predominantly in a win, compromise, or loss.


Federal Mediation and Conciliation Service. Work Stoppages Data. Yearly 1984-2014.

The FMCS collected data on nearly every authorized strike beginning in 1984, which is the data we primarily use to analyze strikes in the dis-regulated period.

The FMCS reports strikes by bargaining unit and includes industry information, size, and union affiliation, which we then aggregate to the industry level. The FMCS does not include strikes associated with new organizing efforts. Although strikes for recognition were common in earlier decades, they had become relatively rare during the dis-regulated period (Martin 2008), suggesting that their omission from the FMCS would not substantially change our analyses. The FMCS identifies strikes by bargaining unit, rather than firm or union, meaning that strikes by two unions at the same firm are counted twice. Even with this potential for overcounting, the strike rate is substantially lower in the dis-regulated period. Similar to other sources, the FMCS does not differentiate between strikes and lockouts, but evidence suggests that, like in the earlier periods, the rate of lockouts is small (Martin and Dixon 2010, fn 9).


6. National Labor Relations Board Data

The National Labor Relations Board (NLRB) published detailed yearly reports from 1936 to 2010. These reports have several types of information relevant to this project. These detailed annual reports were discontinued in 2011; however, aggregate measures on elections and decertifications are still available through the NLRB website.

The NLRB reports include the number of unfair labor practices cases filed by employers and unions. Only unions filed ULPs initially; employers began in 1948. We do not present ULP trends for industries that are largely uncovered by the NLRB, including public administration and education and agriculture, forestry, and fisheries. In 1964, the NLRB changed several aspects of its yearly reports, including adding several industries (entertainment, personal services, and health services).

The NLRB reports also include aggregate and industry-specific data on NLRB elections, including the number of elections and their outcome, as well as the number of workers eligible to vote and their outcome. The reports also include information on decertifications starting in 1948. Beginning in 2009, the reports use fiscal, rather than calendar, year.

Each yearly NLRB report has a table that breaks down elections by industry, which is the main table we use for this project. This table reports on the numbers of cases closed in any given year (rather than petitions filed) and outcome. The NLRB holds several different types of elections and their reporting methods change over time, especially on how they report on decertification elections. Until 1963, the industry table on elections excluded decertification elections. Beginning in 1964, the industry table reported on all elections combined, including decertifications, RC elections, and RM elections. As elections over new bargaining units, RC elections are filed by employees or unions and RM elections are filed by employers. RC cases make up the vast majority of all elections.

Decertification elections were a small fraction of overall elections. For example, in 1964 of the 7,529 elections held, 220 were decertification elections. In 2015, of the 1,687 elections held, 21 were decertifications. The industry-level tables do not indicate the type of elections; however, because decertifications comprise only a small number of overall elections, most of the trends can be attributed to new organizing (RC cases).