Data collection


Please note, these are the initial summary statistics for CARS 2023 and further analysis will follow. We advise linking directly to this document when distributing to ensure the most up to date information.

How we collect data

The Coding in Analysis and Research Survey (CARS) data collection takes place for approximately one month, every autumn. The survey is self-selecting and participation is voluntary. Launch dates vary slightly by year to maximise response rate, for example by avoiding clashes with other internal surveys. In 2023, data collection took place from 16 October to 4 December.

We invite analysts to participate in the survey using a variety of online channels, mailing lists, networks and newsletters. For the past four years, the most common source of data has been through departmental Reproducible Analytical Pipeline (RAP) champions, who promote the survey in their organisations. We rely on various champion networks, Heads of Profession (HoPs) for analysis and Departmental Directors of Analysis (DDans) to promote the survey and encourage their analytical communities to participate. This means the response rate and any selection bias will vary across organisations.

Our promotional materials make it clear that we are interested in responses from all analysts, whether or not they use coding in their work. The survey may however attract a disproportionate number of respondents who have an interest in coding and RAP. We advise against making strong inferences about differences between professions and departments or attempting to estimate real frequencies from the data because of these potential limitations.

Lastly, while the survey is open to all public sector analysts, the vast majority of responses come from the UK and devolved Civil Service (93.2% in 2023). As such, follow-up questions on grade and profession applied only to civil servants.

Where our data comes from

Link tracking allows us to see where responses are coming from. Links promoted by RAP champions were the most commonly used for the past three waves, and accounted for over half of responses in 2023.

Tracking link 2020 2021 2022 2023
RAP champions 33.9% 47.6% 40.8% 50.9%
HoP/DDan mailing list 4.7% 12.8% 15.7% 20.2%
Other 12% 2.6% 6.8% 11.9%
Profession newsletters/mailing lists 7.5% 10.7% 15.6% 9.7%
Slack 12.2% 3.7% 11.8% 7.3%
Quality champions 0% 14.7% 5.2% 0%
Other champions 12.3% 0.8% 0% 0%
ONS RAS mailing list 17.5% 7% 4.2% 0%

Sample size by year

Year Sample
2020 1060
2021 912
2022 1322
2023 1297

Respondent characteristics

Coding frequency

Every year, we ask respondents how often they code to achieve work objectives. While our communication strategy has changed over time, particularly to encourage more non-coders to respond, the findings remain consistent, with a gradual increase in the number of coders over time. Although we seek responses from all analysts the data probably over-represents people with current or prior coding experience.

Show chart Show table
In your current role, how often do you write code to complete your work objectives? 2020 2021 2022 2023
Never 15.1% 12% 12.3% 13.3%
Rarely 12.9% 13% 10.6% 11.7%
Sometimes 20.4% 18.4% 18.2% 19.7%
Regularly 29.7% 30.9% 29% 27.4%
All the time 21.9% 25.7% 29.9% 27.9%


Across all years, over 80% of Civil Service respondents reported that they are at H, S or Grade 7 grades. While this will be representative of the grade distribution of analysts in some government organisations, it may not be the case for all organisations.

Show chart Show table
Grade 2020 2021 2022 2023
Administrative officer or Executive officer 8.1% 9% 8.1% 6.9%
Higher Executive Officer 28.2% 27.7% 27.1% 28.6%
Senior Executive Officer 32.9% 29.6% 32.1% 32.1%
Grade 7 24.2% 26.8% 23.8% 27.5%
Grade 6 or above 6.7% 6.9% 8.9% 5%


Below is a breakdown of the proportion of respondents in different Civil Service professions. These cover the Analysis Function professions and do not apply outside of the civil service. The exception to these are data scientists and data engineers who do not have an official government profession. They are included separately here to avoid skewing the data for other professions. Note that respondents can be members of more than one analytical profession. Profession data is difficult to compare across years as these questions have changed in line with changes to the Analysis Function.

The CARS sample has high representation from statisticians compared with other professions. This again may be representative of some organisations but not all.

Profession Percent
Statisticians 31.7%
Social researchers 17.9%
Civil servant - no profession membership 11.3%
Data scientists 11.1%
Operational researchers 10.9%
Economists 9.6%
Digital, data and technology profession 6.2%
Civil servant - other profession 3.7%
Data engineers 2.3%
Geographers 1.5%
Actuaries 0%