Statistician
| Median wage (USD) | 105650 |
|---|---|
| Projected employment growth (%) | 10 |
| AI exposure (%) | 71 |
| Maximum exposure (%) | 95 |
| Source | eloundou |
| What AI changes in this occupation | Cleaning and weighting raw data, computing estimates and confidence intervals, producing charts and tables, writing the standard report and coding the model now take minutes with an assistant: the statistical production chain — from raw data to final table — is the share being delegated first Requesters now arrive with an analysis already produced by AI, which shifts the work towards checking the method, the assumptions and the validity of the conclusions; routine analyses — dashboards, A/B tests, descriptive statistics — are self-served inside teams; and the statistician is increasingly brought in upstream, on the question and the study design, rather than downstream on the computation Deciding whether the method is valid for the question asked, judging whether the data can support the conclusion one wants to draw, designing the study and the sampling, saying "this result does not prove that" when the analysis is ambiguous, and carrying responsibility for the published figure before peer review, a regulator or a clinical trial remain human — it is the commitment to validity that makes the statistician |
| Key skills | Probability theory and statistical inference · study design and experimental design · sampling, adjustment and weighting · statistical modelling (regression, time series, Bayesian models, mixed models) · R, Python and SQL · assumption checking and data quality control · communicating results to non-specialists · technical writing and reproducibility of the analysis · generative AI tools for cleaning, code and chart production |
| Courses to get ahead | Statistical Inference · Coursera (Johns Hopkins University) · About 18 hours in total · 92 · The course that carries the share AI does not take: estimating a parameter, building a confidence interval, setting up and testing hypotheses, and above all knowing when the inference is valid. This is the gesture the statistician must be able to defend before a client, a regulator or a review — a language model produces the interval, it does not understand why the assumption behind it is violated. Bayesian Statistics: From Concept to Data Analysis · Coursera (University of California, Santa Cruz) · 4 weeks, 2 to 5 hours a week · 88 · The method the statistician is increasingly called on to arbitrate: choosing a prior, justifying a model, updating a belief from data and explaining what the result means. The course comes from a university that teaches it to its own students and covers the side of the job where the method is decided — the side the ζ of 95% makes critical. AI For Everyone · Coursera (DeepLearning.AI) · About 7 hours in total · 84 · For deciding what to delegate and what to keep: the course explains what generative AI can do, where it fails and how to frame a project that uses it. A statistician whose cleaning, code and charts now take minutes needs to know what these tools actually produce — and what they cannot validate. Introduction to Statistics · Coursera (Stanford Online) · About 6 to 8 hours in total · 82 R Programming · Coursera (Johns Hopkins University) · About 25 hours in total · 80 SQL for Data Science · Coursera (University of California, Davis) · 4 weeks, 3 to 5 hours a week · 78 |
| Occupations to pivot toward | Data Scientist · 120230 · 35 · 59 Operations Research Analyst · 88940 · 12 · 63 Actuary · 130000 · 9 · 54 |
| Skills to acquire for the pivot | Eloundou measures the statistician's exposure at 71% (human β) and up to 95% (human ζ): preparing and weighting raw data, processing large volumes, writing modelling code, producing charts and tables and drafting the standard report are largely delegable, but choosing the method that is valid for the question asked, evaluating the procedures used, interpreting an ambiguous result and designing the study remain human. Within the repository, 71 puts the statistician third of the ninety-three occupations, behind the writer / author (81) and the translator / interpreter (84): ninety of the ninety-two occupations already published sit lower, and the nearest neighbour is the customer service rep (70), just ahead of the mathematician (69), the tax examiner, the computer programmer and the web / UI designer (68). Its ζ of 95% places it in the leading group, alongside the loan officer, the personal financial advisor and the data engineer (95), behind the five occupations at 100 (writer / author, mathematician, web / UI designer, proofreader, information security analyst). The explanation is direct: the job produces a deliverable — a number, a table, a chart — that the machine now produces too, but it commits to a validity the machine cannot commit to. The GPT-4 self-rated series (β 79, ζ 100) is kept in expositionGpt4 and is not displayed, in line with the decision of 21 August 2026. BLS projects 10% growth in mathematicians and statisticians employment from 2025 to 2035, faster than the average for all occupations, with about 2,000 openings a year over the decade and 33,500 jobs in 2025; the figure is for the combined group, because BLS does not publish a separate projection for statisticians — only the wages of the two occupations are distinguished on the page. O*NET, for its part, puts employment at 32,200 people in 2024, with 2,000 projected openings over 2024-2034 and growth rated "much faster than average (7% or higher)", and flags the profile as a Bright Outlook occupation. The driver is demand for analysis: companies, government bodies, laboratories and insurers handle data volumes that require defensible estimates, and BLS explicitly ties the occupation to the scientific and technical services sector and to government. The ζ of 95% says almost the whole job is at least partly transformable; what protects employment is therefore not technical shelter, but the fact that the hiring decision rests on responsibility for the conclusion — a published figure has to be defensible, and a model does not defend it. The occupation's median salary is $105,650 a year, the May 2025 figure from BLS published on the Occupational Outlook Handbook page "Mathematicians and Statisticians" — $50.79 an hour, with 10% of workers below $64,000 and 10% above $174,050. The same page gives the median for the combined group ($105,720, or $50.83 an hour) and that of mathematicians ($126,710): BLS distinguishes the wages of the two occupations but not their employment projection. For context, in the same May 2025 snapshot: data scientists at $120,230 (+35%), operations research analysts at $88,940 (+12%) and actuaries at $130,000 (+9%) — the three pivots on this card, each taken from its own BLS page. The job usually requires a master's degree in statistics or biostatistics; O*NET places it in "Job Zone Five: Extensive Preparation Needed", and its 15-2041.00 profile confirms the same median of $105,650 ($50.79 an hour) along with 32,200 workers in 2024. The pay ceiling comes from biostatistics in clinical trials, risk modelling or leading an analysis team. Every value on this card comes from a named source; only the task shares, derived from O*NET importance scores, and the course scores are editorial estimates, documented in _ponderation. |
| GPT-4 exposure — beta / zeta (%) | 79 · 100 |