Official Statistics

Trade union membership, UK, 1995 to 2024: technical information

Published 22 May 2025

Trade union questions in the Labour Force Survey (LFS)

A question on the LFS on trade union membership has been asked annually since 1989 of all individuals in employment. Questions on trade union presence and recognition were introduced in 1993 and the question on collective agreements was introduced in 1996. The questions relating to trade union membership were reordered and reworded in 1999; these changes affect the time-series for trade union presence and collective agreements. The wording of the questions pre-1999 and post-1999 can be found in the 2023 bulletin.

The wording of the question that asks respondents whether they are a member of a trade union remained the same, only its place in the sequence changed.

The question that asks whether any of the people at the respondent’s place of work are members of a trade union or staff association is designed to measure trade union presence. The wording, routing and sequence of this question changed in 1999. Prior to that, it was asked of all in employment; but from 1999, it has only been asked to those who say that they are not union members.

Before 1999, the question on whether the respondent’s pay and conditions were directly affected by collective agreements between their employer and a union (LFS variable: TUCOV) was only asked where the respondent first identified unions as being present at the workplace (LFS variable: TUPRES), and then whether it was recognised (LFS variable: TUREC). This meant that the number of people whose pay and conditions were affected by collective agreement was an underestimate. For this reason, the routing of the question was changed in the 1999 LFS and is now asked of all in employment. Users must therefore be aware that data derived from the TUCOV LFS variable in the 1999 dataset are not directly comparable with those of previous years due to the change in the question’s coverage.

It is possible that some non-sampling error arises in the series of questions on trade unions because of measurement problems. Around a third of the sample are proxy respondents (other household members answering on behalf of the individual), and the data shows that this group are less likely to be identified as union members than those responding on their own behalf. This issue was considered in more detail in a 2013 paper.

The Annual Survey of Hours and Earnings (ASHE), conducted by the Office for National Statistics (ONS), collects information on the number of employee jobs where pay is set with reference to an agreement affecting more than one employee (such as agreed collectively by a trade union or a workers’ committee). The question asked is therefore broader than that asked in the LFS. As the information is collected from employers it may give a more accurate picture of the coverage of such agreements. The achieved sample in 2024 was around 173,000.

In this bulletin, the ASHE figures relating to collective agreements have been included in the main tables and are used as the main measure of collective agreement coverage. We expect the figures more reliably reflect the level of coverage in the UK.

It is known from surveys of employers that only a small proportion of public sector workplaces are not covered by collective bargaining, and that these arrangements are generally made at head office level or across many organisations. It is therefore likely that some employees who are not union members and/or who work in small workplaces in the public sector may be unaware that collective bargaining arrangements apply to their organisation.

There is potentially a similar effect among workers in the private sector. This may explain why the LFS estimate of collective bargaining coverage is well below the expected level, especially for the public sector. In addition, because sector is self-reported in the LFS, there may be a number of respondents wrongly classifying themselves as public sector workers. Consequently, there may be a downward bias to this measure.

The ASHE question references collective agreements agreed by a trade union or worker’s committee, so possibly some non-union collective agreements may also be included (though we would not expect these to be widespread). The question in the LFS specifically refers to negotiations between the employer and a trade union that affect an employee’s pay and conditions. The LFS collective agreements statistics are provided in the annex tables.

Northern Ireland 1997 data issue

In Autumn 1997, incorrect routing of Northern Ireland respondents meant that only those who answered yes to TUPRES were asked UNION. Therefore, there was under-response of unknown size to UNION. Those who should have been asked UNION were instead coded as not having answered the question. DBT analysis indicates this may mean that there is an undercount of up to 10,000 in 1997 levels or 0.1 percentage points in 1997 proportions.

Variables in the LFS

Most of the classifications used to place respondents in different categories are based on a direct question relying on the person’s self-assessment of their circumstances. Some are based on a combination of more than one question, and others are coded by the ONS based on standard conventions.

The classifications used in this publication are generally self-defined. This includes gender, age and ethnic group. Two types of employment status – full-time or part-time, and permanent or temporary – are based on direct questions and do not rely on any set criteria (for example, number of hours worked). The classification for flexible working hours only includes those who work under such arrangements. Region and whether an individual is an employee are both self-defined variables. Highest qualification is principally based on a question asking individuals to nominate what qualifications they have from a list of around 50 categories.

These have then been aggregated for the analytical purposes. The categories for ethnic groups have been changed in the LFS in both 2001 and 2011 to be consistent with those used in the 2001 and 2011 Censuses, respectively. The estimates for employees with a disability use the DISEA variable based on whether an individual has a disability as defined in the Equality Act 2010.

Industry and occupation variables in the LFS are produced by coding textual open-ended survey answers to standard classifications, the Standard Industrial Classification and Standard Occupation Classification (SOC).

The occupation figures for 2021 are based on SOC20, which was introduced to the LFS in 2021. The revised occupation codes primarily move occupations into the Professional Occupation major group. This reflects ONS findings that “an increasing number of occupations require the application of knowledge and expertise that is associated with a tertiary level of education”.

Defining whether people work in the public or private sector is based on 2 questions first introduced in 1993. These ask, first, if they worked in a private firm or business, a limited company, or some other kind of organisation; and second, if other, what kind of non-private organisation. The LFS defines public sector as that owned, funded or run by central or local government and the private sector is everything else.

Some respondents may not know whether their industry is in the public or private sector, and, for certain types of activity, there may be no clear answer and the interviewers are given some guidelines to help sort out some common confusion. There is no indication that the reclassification of some banks to the public sector for a period after the 2008 financial crisis by the ONS in its Public Sector Employment statistics had any impact on how people working in these organisations classified their sector in the LFS.

The estimates produced in this bulletin from the LFS data show a higher level of employment in the public sector, and a lower level in the private sector, than the ONS Public Sector Employment statistics.

Sample size and standard errors

The cell sizes in some tables in this publication are too small to provide reliable estimates and have been marked with an asterisk. In this publication, estimates fewer than 10,000 have not been published because standard errors are likely to be larger than the estimates themselves. Although the ONS has lifted its protocol of not releasing data under the 10,000 threshold, it is now the statistician’s discretion whether to release anything under this threshold. For wage related data, the threshold used is 30,000.

As an indication of the standard errors in the trade union estimates, the standard errors around the total employment and employee union membership proportions are likely to be around 0.3 percentage points for 2024. Standard errors for union membership proportions by gender are likely to be around 0.5 percentage points. Standard errors for union membership proportions by region average around 1.1 percentage points.

The 95% confidence intervals for trade union membership, presence in the workplace and collective agreements and average wages are provided alongside the tables. These have been calculated using the survey package in R following ONS guidance on calculating confidence intervals for complex survey designs such as the LFS.^{[footnote 1]}

Following the guidance provided by the LFS methodology team, the person identifier (CASENO) is used as the clustering variable, unitary authority/local authority (UALA) as the low-level geography stratum, and the person weight (PWT) and income weight (PIWT) as the weighting variable where applicable. As with the tables, the domain is across employees or those in employment that are not in a military occupation.

Switch from seasonal to calendar quarters

In 2006, the structure of the LFS switched from a seasonal quarter basis to a calendar quarter basis. The last set of published LFS seasonal results covered December 2004 to February 2005. In accordance with European Union regulations, all subsequent quarters have been published on a calendar quarter basis. In 2008, the ONS carried out an extensive reweighting programme and all quarterly LFS data are now published on a calendar quarter basis back to 1992.

This has eliminated the structural break into the trade union membership time series associated with the change from seasonal to calendar quarters. Analysis was conducted (see the 2023 bulletin) to demonstrate that the reweighting of historic LFS datasets did not materially changed the trade union densities from those that were previously published.

Estimating union membership levels

Each household agreeing to take part in the LFS is interviewed on 5 consecutive quarters (or waves). However, there are cases on subsequent waves when households are not contactable or do not agree to continue taking part so further responses cannot be recorded. When this happens, current data may be imputed by carrying forward answers given by them in the previous interview. However, for questions that do not appear every quarter, as with trade union membership, there is no previous response to carry forward, and a ‘does not apply’ (DNA) response is therefore recorded.

There are 2 possible methods of estimating union membership levels. One method relies on calculating an adjusted weight, whereby cases with missing data are assumed to have the same distribution as valid responses. Consequently, missing data is included in the estimates by allocation of their weight along the same distribution as valid respondents.

Unfortunately, due to the restructuring of the LFS to calendar quarters (described earlier) this produced too much missing data in the seasonal quarter series to produce reliable estimates. Information was lost for September in converting seasonal to calendar quarters and no information was available to be brought forward from December. Therefore, it was not possible to display a consistent time series including the data originally collected on a seasonal quarter basis using the adjusted weight method.

In order to mitigate these issues a second method was developed to closely approximate the results from an adjusted weights approach (see Table A2) but also allow a consistent time series with the calendar quarter information to be estimated. It consists of union density multiplied by the population (as estimated by the LFS).

The method is improved by making the same calculation but by detailed age, gender and regional disaggregation and then aggregating them back up to national and regional levels. Age, gender and region were specifically chosen as they are the basis on which the LFS is weighted (see the Labour Survey User Guide Vol 1.).

This approach is further supported by the fact that union density figures are produced on a valid response basis and do not consider any missing data. Furthermore, union densities show a great deal of robustness under change. These estimates of union membership levels are now provided as the headline figures in this publication (see Tables 1.2a, 1.3a, 2.1a, 4.2).

To produce trade union membership broken down by public and private sector, a further step was applied. It consists of union density split by public and private sector multiplied by the union membership population (estimated as above). The calculations involved are similar-to that of the second method described above, apart from the fact that a further breakdown by sector was used.

A comparable method was used to calculate membership levels by industry with a breakdown by industry used rather than a breakdown by public and private sector. The table below (produced in 2011) gives a comparison of the estimates from the 2 methods between 2006 and 2010 by gender and employment status.

Table A2a: estimated trade union membership levels, United Kingdom, 2006 to 2010, not seasonally adjusted (thousands)

Year	All employees, union density method	All employees, adjusted weights method	All employees, difference between methods	All in employment, union density method	All in employment, adjusted weights method	All in employment, difference between methods
2006	7,021	6,992	29	7,359	7,315	44
2007	7,005	6,978	27	7,334	7,292	42
2008	6,878	6,857	21	7,219	7,188	31
2009	6,715	6,696	19	7,054	7,017	37
2010	6,536	6,530	6	6,854	6,818	36

Table A2b: estimated trade union membership levels, United Kingdom, 2006 to 2010, male employees and in employment, not seasonally adjusted (thousands)

Year	Male employees, union density method	Male employees, adjusted weights method	Male employees, difference between methods	Males in employment, union density method	Males in employment, adjusted weights method	Males in employment, difference between methods
2006	3,430	3,400	30	3,649	3,611	38
2007	3,379	3,355	24	3,606	3,574	32
2008	3,278	3,254	24	3,522	3,487	35
2009	3,121	3,099	22	3,344	3,311	33
2010	2,976	2,967	9	3,188	3,162	26

Table A2c: estimated trade union membership levels, United Kingdom, 2006 to 2010, female employees and in employment, not seasonally adjusted (thousands)

Year	Female employees, union density method	Female employees, adjusted weights method	Female employees, difference between methods	Females in employment, union density method	Females in employment, adjusted weights method	Females in employment, difference between methods
2006	3,591	3,591	0	3,710	3,702	8
2007	3,625	3,621	4	3,728	3,716	12
2008	3,599	3,601	-2	3,697	3,697	0
2009	3,593	3,596	-3	3,709	3,704	5
2010	3,560	3,562	-2	3,666	3,654	12

To provide the levels estimates in these statistics, DBT uses the GOVTOR LFS variable for regions. If this variable is not available to analysts trying to replicate the results, then GOVTOF, or URESMC would provide a reasonable alternative. However, it would produce marginally different results.

The levels figures presented in the bulletin are rounded to the nearest thousand and the percentage figures to one decimal place. The year-on-year changes reported are calculated on the unrounded data (so may differ slightly from the difference between 2 years calculated from the published tables).

Differences between administrative data and LFS estimates

Another official source of trade union membership data is provided by the Certification Officer (CO) and can be seen in Table 1.1. Data collected annually from trade unions by the government, including the CO from 1974, provide a long and consistent back series of the number of trade unions and the number of union members. The LFS has a shorter back series, from 1989 onwards, but can provide extensive information on the respondent’s individual and workplace characteristics, allowing more detailed analysis.

There are differences in how the 2 sources report union membership. For example, the CO membership count includes all members of unions having their head office in Great Britain (unions that are listed or scheduled by the CO), including those members in Northern Ireland, the Irish Republic and ‘elsewhere abroad’. These figures may also include union members who are unemployed, inactive or retired. The LFS asks questions on the union status of all those in employment, thus excluding the unemployed and retired. The LFS also only covers those who are resident in the UK (or in Great Britain for the years 1989 to 1994).

The LFS union questions have United Kingdom coverage from 1995 onwards. The LFS estimates the number of individuals who are union members, rather than the individual memberships. Hence those belonging to 2 unions would appear twice in the CO data, but only once in the LFS data.

The administrative data figures for 2021 to 2022 and 2022 to 2023 combine figures from Unite the Union’s partial annual returns with the figures published in the Certification Officer’s Annual Report for 2023 to 2024.

Transformed Labour Force Survey (TLFS)

Due to quality issues with the LFS linked to response rates and achieved sample which fell substantially during the COVID-19 pandemic of 2020, ONS has been developing a Transformed Labour Force Survey which will be an online first version of the LFS, with a much larger initial sample. It has subsequently been split into a shorter ‘core’ survey and a longer ‘plus’ version.

A decision on moving to the TLFS as the government’s main labour market survey in November 2026 will be taken in July 2026. The trade union questions will be on the ‘plus’ survey. Find out more information on the TLFS.

1. ONS (2016) ONS methodology working paper series no. 9 – Guide to calculating standard errors for ONS Social Surveys ↩