LO
Get told when a verified record lands on this occupation → · Mark which of these tasks are yours (VOLO Pro, free during the launch) →
Statistician
Designs surveys, samples and trials, analyses data with statistical methods, and checks that published figures are sound. AI has arrived first in the processing around the statistics: the US Census Bureau's language-model autocoder raised the share of occupation and industry answers coded automatically from 28 to 52 percent, and Statistics Canada put machine-learning coding into its Labour Force Survey in 2021. In both, statisticians set the error thresholds and checked the published estimates for breaks. Britain's statistics regulator says AI in official statistics is still mainly research, testing and administrative support rather than routine production, and Europe's medicines agency treats AI used to analyse trial data as part of the statistical analysis, to be frozen in the analysis plan. The US projects statistician jobs to grow 11 percent to 2035, without mentioning AI.
Find out which steps in your team's production now use machine learning or language models, and who checks their output.
This is not a probability of losing your job. It combines how much of the role's task load is exposed to automation with how far adoption has actually gone — useful for comparing occupations on one consistent basis, and for nothing else.
Written for statisticians in national statistics offices, government, health research and clinical trials, and for biostatisticians. Data scientists, data analysts and actuaries have their own pages. The evidence is a US labour projection, statistics offices in the US, Canada and Britain describing machine coding in their surveys, Britain's statistics regulator on how far AI has reached official statistics, and Europe's medicines agency on AI in trial analysis; it establishes where processing around the statistics is automated and what rules govern AI in analysis, not how statisticians' numbers or workloads have changed.
What is actually changing#
The unit of analysis is the task, not the job title. A role is not replaced — its task mix shifts.
Each tile is one task. Its size is how much of the job it is; its colour is where the task is heading. Click a tile to see what the judgement does not establish.
The absence of a record and one regulator's guidance; it does not measure whether AI tools help draft designs or plans.
Two agencies' accounts of coding; they do not show how statisticians' own time on processing changed.
A regulator's observation for one country and one agency's guidance; they do not measure how much analysis AI tools now draft.
Two agencies' accounts; they do not show whether quality-assurance work grew overall.
The absence of a record; it does not measure how statisticians' time on communication is changing.
Is this your job? Say so and this page narrows to your share of it.
A job title is a bundle of tasks bought together, and no two people hold the same bundle. Nothing is sent anywhere — it stays in this browser.
Read all 5 tasks in full — direction, reasoning and limits →
Recent changes#
United States. The statistics bureau projects statistician employment to grow 11 percent from 2025 to 2035, from 31,300 to 34,700, citing more widespread use of statistical analysis to inform business, healthcare and policy decisions and growing amounts of digitally stored data. The page does not mention artificial intelligence, machine learning or automation.
A named person with standing publicly predicted something, on a date, in an attributable statement. It is recorded so that who said what, and when, stays checkable — and it never moves a task's assessment, because a prediction is not an observation. Its value arrives later: the record sits on the same page as the evidence about that occupation, so anyone reading the forecast reads the record of what happened next beside it. That is the reckoning; this site publishes no verdict on whether a forecast came true.
United Kingdom. The regulator says tools such as the statistics office's Survey Assist may help reduce time spent on repetitive processing and let statisticians focus more on quality assurance and meeting user needs, but that its regulatory experience suggests AI use in official statistics remains focused mainly on research and development, testing and administrative support rather than routine statistical production. A regulator's assessment for one country; it gives no counts.
An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
United States. Census Bureau staff report that a language-model autocoder for American Community Survey industry and occupation answers raised the joint coding rate from 28 to 52 percent, sending 32 percent fewer cases to clerical coding each month, or 500,000 fewer a year; the probability threshold was chosen so the error rate is about 6 percent, the limit clerical coders must meet. The slides say the views are the authors', not the Bureau's, and a contractor that built the model is among the authors. The coding was clerical work; statisticians set the threshold.
An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
Great Britain. The transparency record, with phase given as pre-deployment, says the tool asks follow-up questions based on a respondent's input to clarify their job and industry and then determines the SIC and SOC codes; there is close human analysis during research and development, but no direct human intervention in the questions the model asks or the results it provides, with clerical review of accuracy during the research phases. A pilot; no results are reported.
Small-scale trial in a real setting. Tells us the deployment conditions are being tested, not that they hold — so one pilot is never enough on its own; two independent ones are.
European Union. The reflection paper says that when AI or machine-learning models are used for transformation, analysis or interpretation of data within a clinical trial, they are part of the statistical analysis and should follow applicable guidelines on statistical principles; in pivotal trials the data pipeline and models should be pre-specified, frozen and documented in the statistical analysis plan, incremental learning approaches are not accepted, and any change during the trial requires a regulatory interaction to amend the plan. A reflection paper, not legally binding; it binds the sponsor, not a named person.
Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.
An academic benchmark of 411 questions with data sheets from textbooks, online learning materials and academic papers, testing statistical and causal reasoning. The strongest model, GPT-4, achieved 58 percent accuracy; models had difficulty with data analysis and causal reasoning and struggled to use causal knowledge and the provided data together. It tests 2024-era models on textbook-style questions, not statisticians' work.
A demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.
Canada. Agency staff report that machine-learning coding of industry, occupation and class of worker was successfully implemented in the Labour Force Survey production process in October 2021; comparing 13,665 published estimates produced with and without it, they observed no breaks, and they expect machine learning to shift the nature of manual coding work. The agency's own symposium paper.
An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
What this means for you#
If you are starting out, expect routine coding and data cleaning to be done by models, with you checking them. Build what models do not do: study design, judging what data can support, and quality assurance on automated processes.
Expect more of your work to be setting thresholds and checking machine output, and less hand processing. Design, methodology and advice stay with you.
Your options#
Four directions, each with its real constraints and one thing you can test this week. Continuing as you are is a legitimate choice — it just has to be a chosen one.
Take on quality assurance of automated processes
Where statistics offices automated coding, statisticians set the error thresholds and checked published estimates for breaks.
It needs familiarity with how the models work and with the agency's quality standards.
Find out which steps in your team's production now use machine learning or language models, and who checks their output.
Stay in design and methodology
No record shows AI designing samples or trials in production, and trial rules require AI analyses to be fixed in the analysis plan in advance.
Design and methodology roles usually expect years of experience or a postgraduate degree.
List which of your tasks last month were design or methodology and which were routine processing.
Move into clinical trial statistics
Medicines regulators treat AI in trial analysis as part of the statistical analysis, which needs statisticians who understand both.
It needs knowledge of trial design and regulatory guidelines, and often specific training.
Look up trial statistician job ads near you and note what they ask about AI or machine learning.
Common questions#
Not on present evidence. Statistics offices have automated much of the coding of survey answers, but statisticians set the thresholds and check the results, regulators say AI is not yet in routine production of official statistics, and the US projects statistician jobs to grow 11 percent from 2025 to 2035.
We do not answer that with a number of years. Watch whether statistics offices move AI from coding into estimation and analysis, and what trial regulators accept. Those tell you more than any date.
Yes, in production at some statistics offices: the US Census Bureau's language-model autocoder raised the automatically coded share of occupation and industry answers from 28 to 52 percent, with a threshold set to match the 6 percent error limit for clerical coders.
Europe's medicines agency says that when AI models are used to analyse trial data they are part of the statistical analysis, must follow statistical principles, and in pivotal trials must be frozen in the analysis plan; models that keep learning during the trial are not accepted.
What these judgements rest on#
5 of 5 task judgements on this page are backed by a verified event and 0 are platform inference, each labelled where it appears. Behind them sit 1 technology dimensions, a reconstructed trajectory since language models reached the public, and 7 verified events.
See which technologies, how it got here, and the method →
Where it sits in the official classification: skills, knowledge, related jobs →
Other roles in the same function#
A company divides its work into functions before it divides it into jobs. These sit in Technology & data alongside this one — a fact about org charts, not a judgement that they are similar or that they are changing in the same direction.
Junior software developer · Experienced software engineer · Frontend developer · Backend developer · Data engineer · Machine learning engineer · AI researcher · Software tester / QA engineer · Data analyst · Data scientist · Business analyst · Business systems owner · IT support specialist / helpdesk · Security analyst (SOC) · DevOps engineer / platform engineer / SRE · Network engineer · Technical writer / documentation engineer · Grant writer / grants officer