Under development: This book is still under active development, its content and structure are subject to change.

Introduction

The Civil Service People Survey is is the annual employee attitudes survey that has carried out across the UK Civil Service since 2009. It is an important management tool for managers and leaders working within in the Civil Service. It is also a useful tool for external parties that hold senior officials and government ministers to account for the leadership and management of the Civil Service.

This introduction covers: who this companion is for; a brief explanation of how the People Survey is run; key underlying concepts in the structure of survey’s data; and, an outline of the harmonised concepts and identifiers and the structure of the harmonised datasets.

This companion uses the terms ‘Civil Service People Survey’ and ‘People Survey’ interchangeably; the acronym ‘CSPS’ is used sparingly, usually were space is at a premium. A more complete guide to terminology is also available.

Who this companion is for

This companion mainly acts as documentation for the data processing in the csps-data repository. It has been called a ‘companion’ because it takes a narrative approach to documentation that also includes wider discussion of the survey’s background and operation, and explains the rationale for behind decisions taken to harmonise the data over time.

The main audience for this companion is those who wish to reuse the harmonised data for their own analysis. It also serves as a record of decisions made in regards to harmonising aspects of the survey’s data structure that change over time. This companion may also be useful for those working inside the Civil Service that are working with multi-year data and/or interested in comparing results between organisations or to Civil Service benchmarks.

How the People Survey is run

The Civil Service People Survey is coordinated by the People Survey Team based in the Cabinet Office, they are responsible for the overall management and operation of the survey and publishing of the results. The survey’s fieldwork and internal reporting is provided by an external third-party contractor, currently Qualtrics. Alongside the central team and external contractor, participating organisations have a ‘survey manager’ that acts as the liaison point for each organisation, survey managers are responsible for collating crucial ‘local’ information necessary for the running of the survey (such as the internal team hierarchy), publicising the survey and encouraging participation during the fieldwork period, and disseminating the results within their organisation.

The need for a harmonised dataset

While data and results from the survey have been published each year, there is no single dataset or API that makes it easy to access the data. The range of data published each year varies, as do aspects of the survey content and coverage.

Although there is some degree of consistency and stability in the survey’s questionnaire over time there are additions and removals of questions that have occurred over the past 16 years. Organisational coverage varies over time as organisations are established, merged, abolished or changed in other ways. The range of demographic categories either asked about or published has also changed over time.

These challenges prevent easy re-use of the data, including by individuals within the UK Civil Service. The aim of the data processing documented in this companion is to provide a public resource that makes it easy for others to re-use the data.

Original survey data concepts

Before any processing we must first define core concepts in the survey’s data structures and methodology.

Attitudinal questions and measures

The main body of the survey’s questionnaire is made up of individual questions that measure individual’s attitudes and experiences of working in their civil service organisation over the past year. These questions are often rated through a five-point Likert scale of ‘strongly disagree’ to ‘strongly agree’, see the section on response categories for more details. Most of these attitudinal questions are also used to derive composite measures that are used as headline results for the survey, see the section on indexes and theme scores for more details. In the context of the data processing, together, these are referred to as attitudinal questions and measures. In programming code the term qm has been typically used to refer to these data. See !!! for more details.

Organisations

Respondents to the survey participate as part of an organisation, typically (but not always) the government department or agency they work for. Questions in the survey typically ask respondents to consider their responses in relation to this organisational unit. In programming code the term org has typically been used to refer to organisations, and dept_group is used to refer to groups of organisations related to a specific government department. See !!! for more details.

Civil Service benchmark and all respondents results

The headline cross-government results from the survey are the Civil Service benchmark results, this is calculated for each question and measure as the median score of participating organisations.

The results for all respondents to the survey are also published but these scores are largely determined by the experiences of respondents based in the 5-6 largest civil service organisations. These scores may be referred to as either all respondents, all civil servants or mean scores.

Demographic questions

Alongside the attitudinal questions the People Survey asks a range of demographic questions relating to both their job and personal characteristics that are used to understand the experiences of different groups of civil servants. The demographics comprise both questions and categories, in programming code the term demog is typically used to refer to these two aspects collectively, demq for demographic questions, cat for categories, and demcat for combinations of questions and categories (since categories are not always unique across the questions).

In addition to the results for all respondents split by demographic questions and categories there are also more detailed results published for five sets of demographics (sex/gender, ethnicity, health status, sexual orientation and socio-economic background). These detailed demographic results publications provide results of cross-tabulation of these five demographics with all other demographic questions, for example results for men and women by grade. The detailed demographic results also include a summary of organisation scores for each of the five demographics.

Typology of data types

From the structure of the People Survey data described above we can devise a simple taxonomy of the different ‘types’ of data published by the People Survey. The attitudinal questions and measures are excluded from the taxonomy, with the taxonomy acting as a guide to the type of units of analysis covered by different sets of attitudinal question and measure results.

  • The benchmark results - the ‘Civil Service benchmark’, i.e. the median scores of participating organisations.
  • The mean scores - the results for all respondents to the People Survey.
  • The organisation results - the overall results for each organisation.
  • The demographic results - the results for all respondents by individual demographic questions and categories.
  • The detailed demographic results - results for multiple combinations of demographic questions and categories.
  • The organisation demographic results - results of the People Survey’s headline measures for each organisation and select demographic categories.

Response categories

The majority of attitudinal questions are asked on a five-point scale ranging from ‘strongly disagree’ to ‘strongly agree’. For these questions the survey results present the ‘percent positive’ which reflects the proportion answering ‘agree’ or ‘strongly agree’ 1.

In addition to the five-point agreement scale the following scales are used for some questions:

  • Yes/no scales: offering respondents either a simple binary ‘yes’ ‘no’ choice or sometimes with a ‘don’t know’ or ‘prefer not to say’ option.
  • Temporal scales: offering respondents options to state how often a situation occurs, typically ranging from ‘always’ to ‘never’ but also including scales that refer to specific time periods (e.g. ‘weekly’ or ‘monthly’).
  • Quality scale: used for ratings of mental and physical health, ranging from ‘excellent’ to ‘very poor’.
  • Numeric extent scale: used for the personal wellbeing questions, respondents are asked to provide a rating from ‘0’ to ‘10’ where a response of ‘0’ represents ‘not at all’ and ‘10’ represents ‘completely’ (e.g. ‘not at all satisfied’ or ‘completely satisfied’).
  • Productivity range scale: used for a question on self-assessed productivity, respondents asked how productive they feel they have been recently and are given a set of percentage ranges to select from.
  • Stay/leave scale: used for the question on future intentions, respondents are asked about the future intentions of working for their organisation with responses ranging from leaving as soon as possible to staying for at least the next three years.
  • Multiple choice scales: offering respondents the ability to select multiple categories in response to a question (e.g. to describe the type of bullying or harassment the individual has experienced).

Indexes and theme scores

The responses to individual questions are used to calculate a small number of summary measures, which are used as headline results for the survey. These summary measures are either called indexes when they make use of the full range of the input questions’ response scales or theme scores when they make use of only part of the range of the input questions’ response scales.

  • The employee engagement index is derived from five questions relating to respondents levels’ of advocacy, attachment and motivation.
  • The PERMA index and Proxy Stress Index are wellbeing indexes that assess the extent to which respondents are ‘flourishing’ (for the PERMA Index) or experiencing factors that potentially increase stress levels (for the Proxy Stress Index).
  • The nine theme scores are derived from the main section of attitudinal questions and measure factors that influence employee engagement. These scores are calculated relating to the of ‘strongly agree’ and ‘agree’ responses to each question within each theme.

Harmonised survey data concepts

The processing documented in this companion aligns the original survey data published by the Cabinet Office into a set of harmonised datasets that allow for easy cross-sectional and time-series analysis of the data.

Unique identifiers

At the core of the processing is the use of regular expressions (or ‘regexes’) which are used to match question or measure text, organisation names, demographic questions and categories to a set of unique identifiers. There are three sets of unique identifiers:

  • Question and measure identifiers:
    • uid_qm_num: A numeric identifier in the format 0.00.000.00
    • uid_qm_txt: A human-readable text identifier in the format thm.question (where thm is a short-code for the survey section and question is short phrase to describe the question or measure)
  • Organisation identifiers:
    • uid_org_txt: Typically a 6-letter code to refer to organisations, in some cases a shorter 3- to 5-letter code is used
  • Demographic identifiers:
    • uid_demq_txt: Typically a 6-letter code to refer to demographic questions, in two cases a 5-letter code is used
    • uid_cat_txt: A 6- to 12-character alphanumeric code to refer to demographic categories
    • uid_demcat_num: A numeric identifier in the format 0.00.00.000 used to refer to unique combinations of demographic questions and categories
    • uid_demcat_txt: A (somewhat) human-readable text identifier used to refer to unique combinations of demographic questions and categories in the format DEMQ_DCAT where DEMQ refers to the demographic question and DCAT to the category.

Output data files

The output data files are a combination of ‘datasets’ and ‘lookups’, datasets contain the processed data from the People Survey aligned to the unique identifiers while lookups provide labels and other metadata relating to the unique identifiers.

The harmonised datasets structure the data in a ‘long format’, that is each row in a dataset relates to an individual value/score from the People Survey, at a minimum each dataset includes the following columns:

  • data_type: a value representing the type of data included in the dataset (see the data typology section for more details)
  • year: the year of the survey’s results the data relates to
  • uid_qm_num: the unique numeric identifier relating to the question or measure
  • uid_qm_txt: the unique text identifier relating to the question or measure
  • response_category: a general identifier relating to the response category that the value represents (see the response cateogries section for more details).
  • value: the actual survey result/score ranging from 0 to 100 (up to three decimal points). Scores typically relate to percentages (e.g. 23 would represent a true value of 23%, or 0.23 in decimal notation).

Datasets may also include identifiers relating to the organisation (uid_org_txt) and/or demographic objects (uid_demcat_num, uid_demcat_txt, uid_demq_txt and/or uid_cat_txt) that the data relates to.

Table 1: Example structure of the benchmark results file 
readr::read_csv(
    here::here("../data/01-benchmarks/csps_benchmarks_2009-2024_5b58c24b.csv"),
    show_col_types = FALSE
) |>
    dplyr::sample_n(10)
# A tibble: 10 × 6
   data_type  year uid_qm_num  uid_qm_txt                response_category value
   <chr>     <dbl> <chr>       <chr>                     <chr>             <dbl>
 1 benchmark  2023 2.07.001.00 rwk.information_needed    agreement         70.9 
 2 mean       2020 3.03.007.00 tpl.manager_trusts        agreement         88.8 
 3 benchmark  2015 3.01.001.00 tpl.trusted_job           agreement         87.8 
 4 mean       2016 2.03.007.00 mgr.confidence_manger     agreement         71.5 
 5 benchmark  2024 3.04.002.00 dvl.devolution_resources  agreement         35.4 
 6 mean       2024 2.09.006.00 lmc.change_managed_well   agreement         31.8 
 7 benchmark  2010 2.09.010.00 lmc.safe_to_challenge     agreement         39.4 
 8 benchmark  2022 2.03.005.00 mgr.manager_open          agreement         86.3 
 9 benchmark  2012 2.11.001.00 act.senior_action         agreement         43.1 
10 mean       2015 6.02.003.98 dhb.bullied_by_someone_e… multi_choice       3.71

  1. Prior to 2020 some organisations published PDF reports of their results that also included a breakdown of the Likert scale for each question. A change in contractor for the People Survey ended the production of these PDF reports and since then most organisations have stopped publishing their own People Survey results independently of the combined results published by the Cabinet Office.↩︎