The state of UK public sector analysis code: 2026

How to use this research

Responding to CARS is voluntary. The results presented here are from a self-selecting sample of government analysts. Because respondents are self-selecting, the results we present reflect the views of the analysts who participated.

For more detail, see the data collection page.

Coding frequency and tools

We asked respondents if they had any coding experience, inside or outside of work. Of 986 respondents, 95.1% had coding experience. We asked all respondents with coding experience “In your current role, how often do you write code to complete your work objectives?”

To see how this compares to other years, see the data collection page

Coding frequency

2026 data

Show chart Show table

In your current role, how often do you write code to complete your work objectives?	Percentage	Count	Total
Never	9.6%	90	939
Rarely	12.7%	119	939
Sometimes	18.4%	173	939
Regularly	39.5%	371	939
Always	19.8%	186	939
Sample size = 939

Coding interest

In 2026, we also asked respondents with coding experience whether they would like to use more coding in their role.

Show chart Show table

Would you prefer to use more coding in your role?	Percentage	Count	Total
Yes	62.9%	591	939
No	16.4%	154	939
Not sure / not applicable	20.7%	194	939
Sample size = 939

Knowledge and use of programming languages

Knowledge of coding tools

Given a list of programming tools, we asked all respondents with coding experience whether they knew how to use each tool, regardless of whether they currently use it in their work. Note this includes respondents who do not code in their current role.

Please note that capability in programming languages is self-reported here and was not objectively defined or tested.

Show chart Show table

Programming tool	Percentage	Count	Total
R	80.3%	792	986
SQL	59.2%	584	986
Python	51.1%	504	986
SPSS	30.1%	297	986
SAS	29.5%	291	986
VBA	14.6%	144	986
Matlab	12.5%	123	986
Stata	11.4%	112	986
DAX	11%	108	986
Spark	10.3%	102	986
Sample size = 986

Use of coding tools

Using the same list of programming tools, we asked respondents which tools they currently use in their role.

Show chart Show table

Programming tool	Percentage	Count	Total
R	64.4%	635	986
SQL	40.9%	403	986
Python	31.1%	307	986
SAS	13.1%	129	986
DAX	7.3%	72	986
Spark	6.7%	66	986
SPSS	5.2%	51	986
VBA	3.5%	35	986
Stata	0.9%	9	986
Matlab	0.2%	2	986
Sample size = 986

Use compared with knowledge of coding tools

The chart below compares the proportion of respondents with coding experience who reported knowing how to use each programming tool with the proportion who reported using it in their current role.

Show chart Show table

Programming tool	Use	Knowledge
R	64.4%	80.3%
SQL	40.9%	59.2%
Python	31.1%	51.1%
SAS	13.1%	29.5%
SPSS	5.2%	30.1%
DAX	7.3%	11%
VBA	3.5%	14.6%
Spark	6.7%	10.3%
Matlab	0.2%	12.5%
Stata	0.9%	11.4%
None	10%	0.5%

Open source capability over time

The proportion of respondents who report having the capability to use R and Python, is shown alongside the proportion who are able to use SAS, SPSS or Stata, for the past four years of the survey.

Show chart Show table

Programming language type	Year	Know how to programme with these tools (percent)	Lower confidence limit (percent)	Upper confidence limit (percent)
2022	Open Source	80.3%	0.7802509	0.8231395
2023	Open Source	72.6%	0.7005951	0.7491138
2024	Open Source	80.9%	0.7869734	0.8296660
2026	Open Source	87%	0.8477586	0.8897332
2022	Proprietary	56.3%	0.5359005	0.5893030
2023	Proprietary	36.1%	0.3351431	0.3873442
2024	Proprietary	40.7%	0.3805305	0.4338641
2026	Proprietary	53.7%	0.5053039	0.5674350

Professions capability in different tools (%)

Differences in preferred languages may lead to silos between analytical professions. Here we show the percentage of respondents reporting capability in different tools, within the different analytical professions.

Please note that respondents might be members of more than one profession, and may report capability in more than one tool.

Profession	R	Python	SQL	SPSS	Stata	SAS	VBA	Matlab	Spark	DAX
Government Operational Research Service	93.7	68.4	70.5	10.5	6.3	33.7	24.2	24.2	7.4	14.7
Data Scientists	90.1	84.7	84.7	23.4	6.3	17.1	13.5	24.3	30.6	10.8
Government Science & Engineering	90	65	55	10	5	20	35	15	10	0
Data Engineers	66.7	88.9	83.3	33.3	11.1	16.7	16.7	16.7	27.8	27.8
Government Statistician Group	86.4	50.3	61.8	34.9	10.7	36.8	13.8	14.4	11.3	7
Government Economic Service	77.1	38.6	34.3	7.1	40	21.4	7.1	4.3	1.4	1.4
Government Digital and Data	75.9	68.5	66.7	14.8	5.6	22.2	11.1	13	16.7	16.7
Government Geography Profession	73.7	73.7	52.6	31.6	5.3	0	15.8	31.6	10.5	5.3
Government Social Research	63.8	33	36.2	61.7	14.9	19.1	5.3	3.2	12.8	6.4

Git

We asked respondents to answer “Yes”, “No” for the following question:

• Do you know how to use Git?

Please note these outputs include people with coding experience who do not code in their current role.

Show chart Show table

Do you know how to use Git?	Percentage	Count	Total
Yes, I use it in my current role	57.2%	537	938
Yes, but I don't use it in my current role	17.1%	160	938
No	25.7%	241	938
Sample size = 938

Coding capability and change

Coding experience

We asked respondents with coding experience how many years of experience they had in a coding role, excluding any years in education.

This data includes any experience in other roles and sectors, but does not define the type of experience.

Show chart Show table

How many years of professional coding experience do you have?	Percentage	Count	Total
None	3.2%	30	939
Less than 1 year	12.7%	119	939
Between 1 and 3 years	21.3%	200	939
Between 3 and 5 years	20%	188	939
Over 5 years	42.8%	402	939
Sample size = 939

Change in coding ability during current role

We asked “Has your coding ability changed during your current role?”

This question was only asked of respondents with coding experience outside of their current role. This means analysts who first learned to code in their current role are not included in the data.

Show chart Show table

How has your coding ability changed during your current role?	Percentage	Count	Total
It has become significantly better	34%	319	938
It has become slightly better	36.4%	341	938
It has stayed the same	16.2%	152	938
It has become slightly worse	9.7%	91	938
It has become significantly worse	3.7%	35	938
Sample size = 938

Managing a coding project

Respondents who do not currently code in their role were asked if they’d feel confident managing a coding project.

Show chart Show table

Would you feel confident managing a coding project?	Percentage	Count	Total
Yes	6.2%	3	48
No	72.9%	35	48
Not sure	20.8%	10	48
Sample size = 48

Motivation and barriers to coding

Motivation

Respondents who do not currently code in their role were asked whether they would like to use coding as part of their work.

Show chart Show table

Would you like to use coding in your current role?	Percentage	Count	Total
Yes	47.9%	23	48
No	20.8%	10	48
Not sure / not applicable	31.2%	15	48
Sample size = 48

Specific barriers

We asked respondents about any barriers they face in learning to code in their current role. The most commonly cited barrier was a lack of opportunity.

Show chart Show table

Barrier	Yes(%)	NA	NA
Lack of opportunity	52.1%	25	48
Availability of resources	27.1%	13	48
No barriers/not applicable	27.1%	13	48
Suitability of resources	25%	12	48
Availability of tools	16.7%	8	48
Lack of peer support	10.4%	5	48
Lack of management support	8.3%	4	48
Sample size = 48

We also asked respondents about any barriers to improving their coding skills in their current role.

Show chart Show table

Barrier	Yes(%)	NA	NA
Lack of opportunity	49.3%	486	986
Availability of tools	22.8%	225	986
Suitability of resources	20.8%	205	986
Availability of resources	19.3%	190	986
No barriers/not applicable	16.8%	166	986
Lack of peer support	14.1%	139	986
Lack of management support	12.9%	127	986
Sample size = 986

AI tools

Use of AI tools

We asked all respondents who use code in their role, it they use AI tools when coding at work.

The term ‘AI’ was not defined in this question.

Show chart Show table

Do you use any AI assistants when coding at work?	Percentage	Count	Total
Yes	70%	594	848
No	30%	254	848
Sample size = 848

AI tool use by profession

Overall, 70% of coders reported using AI tools when coding at work. The chart below breaks this down by analytical profession, showing the proportion of coders who use AI assistants.

Show chart Show table

Profession	Percentage	Count	Total
Government Digital and Data	81.6%	40	49
Government Economic Service	79.6%	43	54
Data Engineers	77.8%	14	18
Government Operational Research Service	76.7%	69	90
Data Scientists	74.8%	83	111
Government Science & Engineering	73.7%	14	19
Government Statistician Group	68.5%	315	460
Government Social Research	60.3%	35	58
Government Geography Profession	60%	9	15

How AI tools are used

We asked all respondents who use AI tools when coding at work, what they use these tools for.

Show chart Show table

AI use	Never	Rarely	Sometimes	Regularly	Always
Debugging code	7.1%	10.1%	29.5%	40.6%	12.8%
Documenting code	41.9%	23.6%	16.8%	12.6%	5.1%
Getting answers to a coding question	1.7%	7.2%	27.9%	50.3%	12.8%
Learning about new concepts	12.5%	18.7%	32.2%	27.6%	9.1%
Testing code	52.4%	21.2%	14.6%	7.7%	4%
Writing code	4%	17%	36.2%	33%	9.8%
Sample size = 594

Cloud tools

Use of cloud tools

Respondents were asked if they use any cloud data platforms in their current role.

Show chart Show table

Data Platform	Percentage	Count	Total
Microsoft (e.g. Azure)	19.5%	192	986
Amazon (e.g. AWS)	18.7%	184	986
Databricks	14.2%	140	986
Cloudera	5.8%	57	986
Google (e.g. GCP)	3.7%	36	986
Sample size = 986

Reproducible analytical pipelines (RAP)

RAP refers to the use of good software engineering practices to make analysis pipelines more reproducible. This approach aims to use automation to improve the quality and efficiency of analytical processes.

The following links contain more resources on RAP:

• you can find minimum RAP standards in the RAP MVP
• you can find guidance on quality assuring code in the Duck Book

Implementing RAP

We asked respondents to what extent they agreed with the statement: “I am currently implementing RAP in my work.”

Show chart Show table

To what extent to you agree with the following statement: 'I am currently implementing RAP in my work'?	Percentage	Count	Total
Strongly Disagree	6.3%	59	938
Disagree	13%	122	938
Neutral	19.4%	182	938
Agree	34.3%	322	938
Strongly Agree	27%	253	938
Sample size = 938

RAP Barriers

We also asked respondents about the main barriers to implementing RAP in their current role. The most commonly reported barrier was a lack of time.

Show chart Show table

Barrier	Percentage	Count	Total
Not enough time	37.7%	372	986
Work not suited to RAP	32.2%	317	986
Current skills	28.6%	282	986
Lack of guidance	20.8%	205	986
Availability of tools	16%	158	986
Lack of peer support	14.2%	140	986
No barriers	13.6%	134	986
Lack of management support	11.1%	109	986
Other	5.7%	56	986
Sample size = 986

Good coding practices

We asked respondents who reported writing code at work about the good practices they apply when writing code at work. These questions cover many of the coding practices recommended in the quality assurance of code for analysis and research guidance, as well as the minimum RAP standards set by the cross-government RAP champions network.

Coding practices have been classified as either ‘Basic’ or ‘Advanced’. Basic practices are those that make up the minimum RAP standards, while Advanced practices help improve reproducibility. The percentage of respondents who reported applying these practices either ‘Regularly’ or ‘All the time’ is shown below.

Open sourcing was defined as ‘making code freely available to be modified and redistributed’

Show chart Show table

RAP component	Type	Percentage	Count	Total
Use open source software	Basic	75.8%	679	896
Work to quality standards	Basic	64.5%	578	896
Version control	Basic	53.5%	479	896
Peer review	Basic	52.8%	473	896
Create project README	Basic	41.3%	370	896
Open source own code	Basic	14.7%	132	896
Manually test code	Advanced	74%	663	896
Write functions	Advanced	56%	502	896
Use configuration files	Advanced	55.5%	497	896
Follow code style guidelines	Advanced	53%	475	896
Use control flow	Advanced	52.8%	473	896
Document dependencies	Advanced	32.3%	289	896
Document functions	Advanced	28.7%	257	896
Write automated tests	Advanced	14.3%	128	896
Sample size = 896

RAP components over time

The table below shows RAP component use over time for 2023, 2024 and 2026. Empty cells indicate that a component was not asked in that year.

RAP component	2023	2024	2026
Use open source software	71.2	79.9	75.8
Manually test code		79.3	74
Work to quality standards		69	64.5
Write functions	55.9	56.2	56
Use configuration files		53.9	55.5
Version control	44.7	56.5	53.5
Follow code style guidelines	49.3	56.9	53
Peer review	53.6	58.7	52.8
Use control flow		58.4	52.8
Create project README	29.4	43.6	41.3
Document dependencies	29.4	34.6	32.3
Document functions	31.6	31.8	28.7
Open source own code	12.5	15.5	14.7
Write automated tests		16.6	14.3

Consistency of good coding practices

We asked respondents who reported writing code at work how frequently they apply good coding practices when writing code at work.

Show chart Show table

Statement	I don't understand this question (%)	Never (%)	Rarely (%)	Sometimes (%)	Regularly (%)	Always (%)
Have code reviewed	2.7%	5.1%	12.6%	23.8%	26.9%	28.9%
Manually test code	3.3%	2.9%	3.9%	11.7%	33%	45.2%
Separate settings from code	8.5%	7.1%	8.4%	17.5%	29.2%	29.4%
Use a standard code style	12.7%	12%	6.1%	13.1%	28.2%	27.8%
Use control flow	7%	9.2%	9%	19.1%	28.8%	27%
Use open source software	2%	4.6%	4.1%	9.2%	21.6%	58.5%
Write automated tests	14.2%	28.5%	23.9%	18.3%	9.9%	5.2%
Write functions	4%	6.2%	10%	20.5%	35.1%	24.1%
Sample size = 848

Code documentation

We asked respondents who reported writing code at work how frequently they write different forms of documentation when programming in their current role.

Embedded documentation is one of the components which make up a RAP minimum viable product. Documentation is important to help others be clear on how to use the product and what the code is intended to do.

Show chart Show table

Statement	I don't understand this question (%)	Never (%)	Rarely (%)	Sometimes (%)	Regularly (%)	Always (%)
Create README files	0%	17.1%	13.2%	22.5%	23.3%	23.8%
Document dependencies	0%	25.5%	17.9%	18.2%	18.3%	20.1%
Document functions	0%	32.8%	15.6%	15.9%	18.1%	17.7%
Document manual QA steps	0%	9.7%	15.1%	25.9%	29.7%	19.6%
Document pipeline design	0%	21.3%	18.8%	26.5%	21.3%	12.2%
Sample size = 785

Working practices

We asked respondents who reported writing code at work how frequently they use good working practices in the coding projects they work on.

Show chart Show table

Statement	I don't understand this question (%)	Never (%)	Rarely (%)	Sometimes (%)	Regularly (%)	Always (%)
Have a succession plan	5.9%	3.4%	4%	15.1%	35.8%	35.7%
Publish code in the open	13.8%	41.7%	14.9%	14%	9.1%	6.5%
Understand project aims	3.7%	0.4%	1.2%	7.8%	38.6%	48.5%
Understand project roles	9.2%	0.8%	2.6%	14.6%	36.6%	36.2%
Use version control	6.2%	16.4%	9.6%	11.3%	23.1%	33.4%
Work to quality standards	9.3%	2.4%	4%	16.2%	38.3%	29.8%
Sample size = 848

Standards

Respondents were asked whether they follow any analysis or coding quality standards, whether that be either internal standards, external guidance, or a combination of both.

Show chart Show table

Do you work to any analysis or coding quality standards?	Percentage	Count	Total
Yes, internal to my organisation	23.8%	223	938
Yes, Civil Service or other external standards	16.6%	156	938
Yes, both internal and external standards	26.8%	251	938
No, I am not aware of any	19.7%	185	938
Not sure	13.1%	123	938
Sample size = 938

Duck Book

Respondents were asked whether they were aware of the Duck Book.

The Duck Book is guidance produced by the ONS Analysis Standards and Pipelines Hub. It outlines software engineering best practices tailored to analysts and researchers who work with data using code.

Show chart Show table

Are you aware of the 'Duck Book' (Quality Assurance of Code for Analysis and Research)?	Percentage	Count	Total
Yes, I use it	29.5%	103	349
Yes, but I don't use it	29.2%	102	349
No	41.3%	144	349
Sample size = 349

Packages

In 2026, we asked respondents whether they or their team use cross-government or department-specific analysis packages in their work.

See the list of packages reported by respondents on the useful pages page.

Show chart Show table

Do you or your team use any cross-government or department-specific analysis packages in your work?	Percentage	Count	Total
Yes	37.2%	349	938
No	39.1%	367	938
Not sure or not applicable	23.7%	222	938
Sample size = 938

Download tables

Download all tables