Statisticians
15-2041.00Median wage $105,650/yr29,030 employed (US)Rank #99 of 923 scored · top 11% by substitution
Develop or apply mathematical or statistical theory and methods to collect, organize, interpret, and summarize numerical data to provide usable information. May specialize in fields such as biostatistics, agricultural statistics, business statistics, or economic statistics. Includes mathematical and survey statisticians.
Sub-scores
0–100 · band = confidence interval from rater disagreement
Substitution47
Exposure42
Augmentation82
Substitution — the headline: capability discounted by cost, barriers and adoption.
Exposure — technical capability alone, regardless of whether anyone deploys it.
Augmentation — how much AI assists without replacing. High here + moderate substitution = a changing job, not a disappearing one.
Tasks on the substitution scale
19 rated tasks, binned by substitution score.
Position among all scored occupations
Distribution of 923 occupation scores; the marker is this occupation.
Tasks with substitution ≥ 70
11%
Run 1.0.0-draft.1 · computed 2026-08-05 · rater panel: claude-sonnet-5, claude-haiku-4-5-20251001 · intervals span rater disagreement.
Why this score
The five weighted dimensions of the composite, averaged across this occupation's tasks (importance-weighted, panel mean). Exact weights and formulas: /api/v1/methodology.
Task automatabilityw 35%43
panel mean rating 2.7/5 → substitution pressure 43/100
Technical feasibility todayw 20%42
panel mean rating 2.7/5 → substitution pressure 42/100
Cost vs. human wagew 15%45
panel mean rating 2.8/5 → substitution pressure 45/100
Adoption barriersw 20%inverted — strong barriers lower the score60
panel mean rating 2.6/5 (barrier strength) → substitution pressure 60/100
Sector adoption velocityw 10%48
panel mean rating 2.9/5 → substitution pressure 48/100
Task breakdown (19 tasks)
Substitution pressure per task, weighted by O*NET importance in the composite. Expand a task for the full rater audit trail — every rating, every model, every rationale.
Report results of statistical analyses, including information in the form of graphs, charts, and tables.
84CI 75–92 · exposure 87 · augmentation 100 · importance 4.5/5 · click for rater detail
Report results of statistical analyses, including information in the form of graphs, charts, and tables.
84| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Data-driven sectors (finance, tech, analytics firms) have rapidly adopted automated reporting and BI tools that generate graphs and tables; widespread production use is evident in dashboards and self-service analytics platforms. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | placeholder |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI substantially augments statisticians by automating routine report generation, formatting, and visualization, freeing them to focus on interpretation, quality assurance, and communicating insights while remaining fully in the loop. |
| Augmentation potential | claude-sonnet-5 | 5/5 | placeholder |
| Task automatability | claude-haiku-4-5-20251001 | 5/5 | Generating reports with statistical analyses, graphs, charts, and tables is highly automatable with current AI systems. Existing tools can extract data, run standard statistical analyses, and produce formatted visualizations with minimal human input, easily meeting the 50% time-saving threshold. |
| Task automatability | claude-sonnet-5 | 4/5 | Generating narrative summaries, charts, and tables from statistical outputs is highly automatable with current LLMs and code-generation tools (e.g., via Python/R integration), meeting the time-saving threshold for most routine reporting.”, |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | While report generation itself faces minimal legal barriers, the interpretation and validation of statistical results often requires human expertise and sign-off, creating some organizational friction but not a hard legal requirement for human performance. |
| Adoption barriers | claude-sonnet-5 | 2/5 | placeholder |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | The cost of AI-generated reports (inference on standard models plus visualization libraries) is orders of magnitude cheaper than paying a statistician's loaded wage to manually create the same output. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | placeholder |
| Technical feasibility today | claude-haiku-4-5-20251001 | 5/5 | Mature products like Python/R libraries (matplotlib, ggplot2), BI platforms (Tableau, Power BI), and AI-assisted code generation tools (GitHub Copilot, ChatGPT) reliably produce statistical reports with visualizations in production environments at scale. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | placeholder |
Process large amounts of data for statistical modeling and graphic analysis, using computers.
78CI 75–81 · exposure 75 · augmentation 100 · importance 4.0/5 · click for rater detail
Process large amounts of data for statistical modeling and graphic analysis, using computers.
78| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | Financial services, tech, healthcare, and research sectors are rapidly adopting AI-assisted statistical pipelines and code generation, with widespread production deployment of automated data processing and model fitting in information-intensive organizations. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Statistics-heavy fields (finance, tech, research, analytics) show fast adoption of AI-assisted coding and automated data pipelines, though full end-to-end automation is less common. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI tools substantially augment statistician productivity by automating boilerplate data cleaning, generating candidate models, and producing visualizations, while the human retains interpretative, validation, and decision-making authority. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI coding assistants, AutoML, and automated visualization tools dramatically speed up data processing, model iteration, and graphic generation while statisticians retain interpretive control. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Large portions of data processing, statistical modeling, and graphic analysis can be automated with current AI tools (pandas, scikit-learn, GPT-4 code generation) to achieve >50% time savings. However, the task requires domain judgment on model selection, interpretation, and validation that typically still benefits from human oversight, preventing a full 5. |
| Task automatability | claude-sonnet-5 | 4/5 | Data cleaning, transformation, and standard statistical modeling/visualization workflows can largely be automated via scripts, AI coding assistants, and automated ML pipelines, saving substantial time for routine cases.rd cases and novel model design still require human judgment.rd cases and novel model design still require human judgment. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few legal or regulatory barriers prevent automation of data processing and standard statistical modeling; institutional preference for human expertise and model validation/interpretation requirements provide modest friction, but nothing categorical blocks deployment. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement for this technical task, though organizations may require statistician sign-off on model validity and interpretation for high-stakes decisions. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Cloud-based AI statistical tools and LLM-assisted code generation cost orders of magnitude less than statistician labor (typically $80–120/hr loaded) per unit of processed data and standard modeling output. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Cloud compute and AI coding assistants are far cheaper per unit of data processed than statistician hours for routine data wrangling and modeling tasks. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature, deployed products (statistical software, cloud ML platforms, AI coding assistants) reliably perform data processing and graphic generation at scale in production environments. Minor limitations exist around novel model architectures or highly specialized statistical requirements, but general-case performance is robust. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Products like AutoML platforms, Python/R AI-assisted notebooks (Copilot, code interpreters), and BI tools reliably process data and generate models/graphics in production settings today. |
Prepare data for processing by organizing information, checking for inaccuracies, and adjusting and weighting the raw data.
65CI 55–75 · exposure 62 · augmentation 100 · importance 4.5/5 · click for rater detail
Prepare data for processing by organizing information, checking for inaccuracies, and adjusting and weighting the raw data.
65| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Data science and analytics teams across finance, tech, healthcare, and professional services are rapidly adopting automated data pipelines and ETL tools. Adoption is swift and measurable in production environments, especially in large digitized organizations. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Statisticians work across sectors like government, academia, and finance with varying digitization; some fields show fast tool adoption for data prep while others remain more traditional, yielding middling overall velocity. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI dramatically augments statisticians' productivity in data preparation by automating routine checks, flagging anomalies, and suggesting weightings, while the statistician retains oversight and validates quality. This is a textbook case of human-AI collaboration. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI tools (e.g., code generation, automated anomaly detection, statistical software integrations) substantially speed up organizing, cleaning, and preprocessing data while the statistician retains control over final weighting decisions. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems excel at data cleaning, validation, and transformation workflows. Tools like automated data profiling, anomaly detection, outlier handling, and weighting algorithms can handle most of this task end-to-end with significant time savings, though domain-specific adjustments and validation may still require human oversight. |
| Task automatability | claude-sonnet-5 | 3/5 | AI tools can automate much of data cleaning, validation, and standard weighting procedures via code generation and libraries, but complex domain-specific judgment calls about outliers, sampling design, and weighting schemes still require statistician oversight. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few regulatory or licensing barriers prevent automation of data preparation itself. The main friction comes from organizational review requirements and need for statistician sign-off on data quality, but these are soft barriers rather than legal mandates. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement mandates a human perform this task, though organizational quality-control norms and downstream liability for flawed weighting create some friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | Cloud-based data preparation services and open-source solutions are inexpensive to operate per unit of data processed compared to loaded statistician wages. Integration and oversight costs are modest, making the total cost per task-equivalent significantly lower than hiring human labor. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI-assisted data cleaning and scripting reduces time substantially, but oversight, validation, and correction of automated outputs by a trained statistician still adds meaningful cost, keeping the ratio moderate rather than dramatically cheaper. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature products (pandas automation, Trifacta, DataRobot, cloud data pipelines) perform data preparation reliably in production at scale. However, some edge cases around complex weighting schemes and domain-specific accuracy thresholds still often require human judgment, preventing a full 5. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Products like AI coding assistants and automated data-cleaning platforms (e.g., data wrangling tools with ML features) are deployed and used in production, but they still have material error rates and require human review for edge cases and domain-specific adjustments. |
Apply sampling techniques, or use complete enumeration bases to determine and define groups to be surveyed.
64CI 35–92 · exposure 62 · augmentation 88 · importance 3.8/5 · click for rater detail
Apply sampling techniques, or use complete enumeration bases to determine and define groups to be surveyed.
64| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Data-driven and technology sectors (finance, tech, analytics firms, government statistics agencies) adopt automated sampling and statistical tools rapidly and at scale. This is mainstream practice in information and analytical services, with minimal organizational friction. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Statistical and survey research fields have been slower to adopt AI agents for core methodological design work compared to text-heavy professional services. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI powerfully augments statisticians by auto-generating design recommendations, validating sampling assumptions, and flagging edge cases, allowing humans to focus on problem framing and methodological judgment rather than mechanical calculations and routine decisions. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist by suggesting sampling frameworks, calculating sample sizes, identifying potential biases, and automating parts of enumeration base construction, boosting statistician productivity. |
| Task automatability | claude-haiku-4-5-20251001 | 5/5 | AI systems can fully determine sampling methodologies, select appropriate techniques (stratified, cluster, systematic, etc.), define survey populations, and calculate sample sizes with documented efficiency gains. Modern statistical software and AI tools automate these decisions end-to-end, easily achieving >50% time savings at equal or superior quality. |
| Task automatability | claude-sonnet-5 | 2/5 | Designing appropriate sampling frames and defining survey populations requires domain judgment about bias, coverage error, and study goals that current AI cannot reliably determine end-to-end without heavy human specification.assist |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Sampling design is a technical, mathematical task with no licensing requirement for the automation itself, though organizations may impose internal oversight or validation protocols. No regulatory or legal mandate requires a licensed statistician to perform the sampling design—only to vouch for it in regulated contexts. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement, but organizational reliance on statistician expertise for methodological validity creates moderate friction against pure automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Automated sampling and enumeration via software or AI inference costs pennies per task compared to the fully loaded cost of a statistician's labor (typically $60–100+ per hour). AI-driven tools are orders of magnitude cheaper, especially at volume. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply compute sample sizes or execute sampling algorithms once parameters are set, but the costly expert judgment of defining groups and frames still requires a statistician, keeping overall cost comparable to human-led work. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 5/5 | Mature, production-grade statistical software (R, Python, SAS, SPSS) and AI-assisted tools demonstrably perform sampling design and enumeration reliably at scale across organizations. These are standard, deployed capabilities with minimal error rates when properly configured. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | No deployed product autonomously designs sampling strategies for surveys; existing tools (e.g., statistical software with sampling functions) require expert configuration and interpretation. |
Prepare and structure data warehouses for storing data.
62CI 38–87 · exposure 58 · augmentation 75 · importance 3.3/5 · click for rater detail
Prepare and structure data warehouses for storing data.
62| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Cloud-native data platforms and AI-assisted ETL tools are seeing rapid adoption in information and finance sectors, with widespread pilots and production deployments. Enterprise adoption of automated data warehouse management is accelerating. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Data engineering and analytics teams in tech-forward sectors are adopting AI-assisted coding and pipeline tools at a moderate pace, though full automation of warehouse design remains uncommon in production. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI significantly augments human statisticians by automating boilerplate schema design and data profiling, allowing them to focus on validation, optimization, and complex modeling decisions. This transforms productivity while the human remains in quality oversight. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI coding assistants meaningfully speed up writing schema definitions, ETL scripts, and documentation, giving statisticians substantial productivity gains while they retain design authority. |
| Task automatability | claude-haiku-4-5-20251001 | 5/5 | Data warehouse preparation and structuring involves well-defined ETL processes, schema design, and data integration tasks that current AI systems can largely automate end-to-end. Tools like code-generating AI agents can write SQL, design schemas, and orchestrate data pipelines with minimal human input, achieving well over 50% time savings. |
| Task automatability | claude-sonnet-5 | 2/5 | Data warehouse design requires architectural judgment about business needs, schema design tradeoffs, and integration with existing systems that current AI cannot fully own end-to-end, though it can generate boilerplate schemas and ETL scripts. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Minimal regulatory or licensing barriers exist for data warehouse automation itself, though data governance and compliance oversight remain human responsibilities. The low barriers facilitate rapid substitution of routine preparation work. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement exists, but organizational risk aversion around data architecture decisions (given downstream impact on all analytics) creates moderate friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | AI inference for code generation and data structuring is extremely cheap per task compared to the loaded salary of a statistician or data engineer, especially at volume. Cloud automation tools further reduce per-unit costs by orders of magnitude. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can cut some scripting time, the overall effort of designing, validating, and maintaining a data warehouse still requires substantial skilled human oversight, keeping costs comparable to human-led work with AI assistance. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Multiple production systems (Dataedo, dbt AI, and cloud-native data warehouse tools) reliably handle schema generation, data profiling, and ETL automation at scale. While some complex custom logic may still require human judgment, core warehouse preparation tasks are demonstrably deployed and reliable in enterprise environments. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI coding assistants and data tools can help write SQL DDL, suggest schemas, and automate parts of ETL pipelines, but production data warehouse architecture is still largely human-driven with AI as a helper rather than an autonomous performer. |
Identify relationships and trends in data, as well as any factors that could affect the results of research.
60CI 59–61 · exposure 50 · augmentation 100 · importance 4.3/5 · click for rater detail
Identify relationships and trends in data, as well as any factors that could affect the results of research.
60| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Finance, tech, e-commerce, and pharma sectors have rapidly deployed automated statistical analysis and trend detection in production analytics systems. Wide availability of AutoML and self-service analytics platforms indicates fast, deep adoption in digitized organizations. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Data science and research-oriented sectors have moderate AI adoption for analytics, with pilots and tools in production but full trust in AI-driven trend/factor identification still developing. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI significantly augments statistician productivity by automating exploratory searches, generating hypothesis candidates, and flagging anomalies and confounders, freeing experts to focus on interpretation, validation, and causal reasoning. This is a textbook case of high-value human-AI collaboration. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially augments statisticians by rapidly surfacing patterns, correlations, and candidate confounders, letting humans focus on validating causal interpretation and study design. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | Current AI tools can identify statistical relationships, correlations, and basic trend patterns in structured data and execute exploratory analyses with moderate automation. However, identifying causal factors and interpreting contextual risks requires domain expertise and judgment, limiting full end-to-end automation to roughly half the task. |
| Task automatability | claude-sonnet-5 | 3/5 | AI/statistical tools can automate exploratory data analysis, correlation detection, and trend identification for well-structured datasets, but interpreting causal factors and confounders in novel research contexts still requires human statistical judgment.4) |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few legal or regulatory barriers prevent AI-assisted statistical analysis; however, organizational norms, the need for domain expertise to validate findings, and liability for incorrect conclusions create moderate friction against full automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement mandates human sign-off for exploratory analysis, though organizations often want statistician validation for research validity and interpretation, creating moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | AI-powered statistical analysis tools are significantly cheaper per inference than hiring senior statisticians, especially at scale. Integration and minimal oversight costs remain modest, yielding roughly 5–10× cost advantage for routine exploratory analysis. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Automated analytics and AI-assisted exploratory analysis can process large datasets far more cheaply than a statistician manually reviewing data, though oversight for correctness adds some cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Deployed products (statistical software with ML pipelines, BI tools, and data exploration platforms) can reliably detect relationships and trends, but human review is typically required to validate findings and assess confounding factors. Material error rates persist in automated causal inference without expert oversight. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Deployed tools (AutoML, BI dashboards, LLM-based data analysis agents) reliably surface correlations and trends, but confidently identifying confounding factors and validating relationships still shows meaningful error rates and requires domain expertise. |
Analyze and interpret statistical data to identify significant differences in relationships among sources of information.
56CI 55–57 · exposure 50 · augmentation 100 · importance 4.7/5 · click for rater detail
Analyze and interpret statistical data to identify significant differences in relationships among sources of information.
56| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Information, finance, tech, and research organizations are actively deploying automated statistical platforms and AI-assisted analytics; adoption is measurable and accelerating in digitized sectors. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Data science and analytics functions in finance, tech, and research are adopting AI-assisted statistical tools at a moderate pace, with pilots common but full production reliance on AI for critical interpretation still limited. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI excels at augmenting statisticians by automating data exploration, generating candidate tests and models, and surfacing patterns at scale, allowing humans to focus on validation, interpretation, and domain-specific decision-making. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially boosts statisticians' productivity by automating routine computations, generating exploratory analyses, and drafting interpretations, letting humans focus on validation and higher-level judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | Current AI can perform exploratory data analysis, hypothesis testing, and relationship identification with libraries and statistical packages, but selecting appropriate methods and interpreting findings in domain context typically requires human expertise—roughly half the task can be automated with significant setup. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can run statistical tests, generate interpretations, and flag significant relationships in structured datasets, but nuanced judgment about causality, confounds, and domain context still requires human oversight for reliable results.tabular data or novel study designs. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Light regulatory barriers in most contexts; liability and client trust issues exist but do not legally mandate human sign-off in most domains, and organizational adoption is growing steadily. |
| Adoption barriers | claude-sonnet-5 | 2/5 | There's no formal licensing requirement for statistical interpretation in most contexts, though high-stakes fields (clinical trials, regulatory submissions) impose some review requirements and liability concerns that create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 3/5 | AI-driven statistical analysis tools cost substantially less per run than hiring a statistician, but domain expertise and validation oversight still require human involvement, keeping total cost roughly comparable for critical applications. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI tools can cut analysis time substantially for straightforward statistical tasks, but oversight, validation, and interpretation by trained statisticians still add significant cost, keeping the ratio closer to parity than dramatic savings. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Statistical software and ML platforms with automated EDA and hypothesis testing exist in production, but material gaps remain in method selection, validity checking, and contextual interpretation where human statisticians add critical value. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Products like AI-assisted analytics tools (e.g., ChatGPT with code interpreter, AutoML platforms) can perform statistical analysis and interpretation, but production reliability varies and complex or ambiguous datasets often need expert review. |
Present statistical and nonstatistical results, using charts, bullets, and graphs, in meetings or conferences to audiences such as clients, peers, and students.
55CI 55–55 · exposure 50 · augmentation 88 · importance 4.3/5 · click for rater detail
Present statistical and nonstatistical results, using charts, bullets, and graphs, in meetings or conferences to audiences such as clients, peers, and students.
55| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | BI and analytics firms are adopting automated reporting and visualization tools, but live presentation by statisticians in high-stakes settings remains human-dominated; adoption of AI-generated presentation slides is growing but not yet the norm in production environments. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Data analytics and professional services sectors are adopting AI-generated visualizations and slide tools moderately fast, though live presentation delivery by AI agents remains uncommon in production. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI tools meaningfully assist statisticians by rapidly generating draft charts, identifying key insights, and formatting results, allowing the human to focus on narrative and delivery; this augmentation is already widely integrated into analytics workflows. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially speeds up creation of charts, bullet summaries, and even speaker notes, letting statisticians prepare presentations far faster while they still deliver and field questions personally. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can generate charts, graphs, and bullet-point summaries from data and statistical results with modern tools, but meaningful presentation requires judgment about framing, narrative flow, and audience-specific contextualization that typically demands human oversight and revision. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can generate charts, slides, and narrative summaries from data quickly, but live presentation, audience Q&A, and adapting delivery to an audience's reactions still require human presence and judgment. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | While clients and peers may still prefer a human presenter for credibility and live Q&A, there are few hard regulatory or licensing barriers preventing AI from generating the visual and textual content; human oversight is common but not legally mandated. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement blocks AI-assisted presentation creation, but professional and academic norms still expect a human presenter to represent and defend the work, especially to clients or in academic settings. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 3/5 | The cost of AI-generated visualizations and summaries is comparable to or slightly cheaper than paying a statistician to manually prepare presentation materials, depending on the complexity and customization required. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI can cheaply automate the chart/slide-creation portion, but the live presentation and audience engagement portion still requires paying the statistician's time, keeping overall costs roughly comparable to full human execution. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Products exist (e.g., automated report generators, BI platforms with AI-assisted visualization) that can create presentable charts and graphs from data, but they often require significant human curation and lack the nuance needed for live or high-stakes conference presentations. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Tools like Copilot, ChatGPT with code interpreter, and BI dashboard generators reliably produce charts and draft slide decks today, but actually delivering/presenting to a live audience is not something deployed AI does autonomously in professional settings. |
Develop software applications or programming for statistical modeling and graphic analysis.
52CI 46–57 · exposure 50 · augmentation 88 · importance 3.9/5 · click for rater detail
Develop software applications or programming for statistical modeling and graphic analysis.
52| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Data science and tech organizations are actively piloting AI code assistants, but adoption remains in the tooling/augmentation phase rather than replacement. Statistically rigorous sectors (pharma, finance) show slower, more cautious adoption due to validation burdens. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development and data science are among the fastest-adopting domains for AI coding tools, with widespread production use of AI-assisted coding in analytics teams. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI coding assistants demonstrably improve productivity for statistical programmers by accelerating drafting, debugging, and documentation. Tools like Copilot and ChatGPT are widely used to scaffold statistical workflows, allowing human statisticians to focus on design and validation decisions. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI coding assistants substantially speed up writing statistical scripts, generating boilerplate, debugging, and creating visualizations, while the statistician retains control over model design and interpretation. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can assist with code generation and boilerplate for statistical modeling and visualization (e.g., via GitHub Copilot, ChatGPT), but developers typically need to architect the pipeline, validate algorithms, and integrate domain-specific logic. This falls short of 50% time savings at equal quality without substantial human oversight. |
| Task automatability | claude-sonnet-5 | 3/5 | AI coding assistants can generate substantial portions of statistical modeling and visualization code, but end-to-end development of correct, validated applications for novel modeling needs still requires significant human design and debugging.};, |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Statistical software development often requires regulatory compliance (e.g., FDA, pharma), organizational sign-off on model correctness, and liability concerns around faulty algorithms. However, no explicit licensing barrier prevents AI-assisted coding; human statisticians still typically review output. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement mandates a human write this code, though organizations often require expert review of statistical logic for correctness and reproducibility, creating moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI coding tools are cheap per inference, but the overhead of validation, debugging, and domain expertise required to ensure correctness means total cost remains comparable to hiring experienced statisticians for novel modeling work, though lower for routine scripting. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI coding assistance reduces time on boilerplate and common routines, but integration, validation, and debugging by a skilled statistician still command comparable costs to fully human-driven development for non-trivial applications. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Deployed products (code assistants, LLM-based IDE plugins) exist and perform basic statistical coding tasks, but they produce material errors in algorithm selection, edge-case handling, and statistical correctness. Production use remains mixed and typically requires expert review. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Deployed products like GitHub Copilot, ChatGPT, and code-completion tools reliably assist with common statistical scripting (R, Python), but building robust standalone applications still shows material error rates and requires human review. |
Report results of statistical analyses in peer-reviewed papers and technical manuals.
41CI 29–54 · exposure 38 · augmentation 88 · importance 3.9/5 · click for rater detail
Report results of statistical analyses in peer-reviewed papers and technical manuals.
41| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Academic and technical publishing sectors have been slow to adopt AI for autonomous paper generation; most use remains supplementary (drafting aids), and many journals still prohibit or heavily restrict AI authorship. Institutional resistance and professional norms limit velocity. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Academic and research sectors are adopting AI writing tools at a moderate pace, with many pilots and increasing use of AI-assisted drafting, but institutional caution, plagiarism/authorship policies, and disclosure norms slow deeper deployment. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI provides substantial productivity gains as an assistive tool—generating first drafts, formatting tables, suggesting phrasing for methods sections, and accelerating the writing cycle—while statisticians retain full control over interpretation, validation, and final submission. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially boosts productivity in drafting text, formatting tables, summarizing results, and improving clarity, while the statistician retains responsibility for accuracy and interpretation. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can generate statistical summaries and draft text descriptions of results, producing publication-ready peer-reviewed papers requires originality in framing, critical interpretation of findings, and careful navigation of discipline-specific conventions that current AI systems handle inconsistently. The task involves far more than just reporting numbers—it demands novel insight and accountability that AI cannot reliably provide end-to-end. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can draft results sections, tables, and narrative descriptions of statistical output, but ensuring correct interpretation, framing for peer review, and scientific accuracy still requires substantial human effort, so only partial time savings are realized end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Peer-reviewed publication and technical documentation carry high liability and reputational stakes; journals and organizations require human authorship accountability, and journals often have explicit policies against AI-generated text. Professional credibility and legal responsibility create substantial barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | There's no licensing requirement to write a paper, but journals require named human authors accountable for results, peer review scrutiny, and scientific integrity norms create meaningful friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | A capable statistician's loaded labor cost (salary, benefits, overhead) typically far exceeds the marginal cost of running an LLM for draft generation and formatting, though human review remains mandatory, limiting the net savings. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI drafting tools are cheap per use, but the need for expert verification, correction, and liability review for peer-reviewed accuracy keeps total cost roughly comparable to a human-driven process with AI assistance rather than an order-of-magnitude cheaper full replacement. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed system reliably produces complete peer-reviewed papers or technical manuals; AI tools can assist with drafting sections and formatting but consistently fail on accuracy, originality verification, and compliance with journal-specific standards. Researchers use AI for assistance, not autonomous paper generation. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Products like ChatGPT, Claude, and specialized writing assistants are used today to help draft manuscript sections and technical reports, but they are not reliably deployed to autonomously produce publication-ready statistical reporting without heavy human review. |
Evaluate sources of information to determine any limitations, in terms of reliability or usability.
35CI 32–38 · exposure 25 · augmentation 63 · importance 4.1/5 · click for rater detail
Evaluate sources of information to determine any limitations, in terms of reliability or usability.
35| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Data quality and governance tools have moderate adoption in large organizations and tech-forward sectors, but the critical evaluation of source limitations remains heavily manual; adoption of AI-assisted evaluation is in pilot phase rather than production-standard across the statistical workforce. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Statisticians work across finance, research, government, and tech—sectors with mixed but growing AI adoption for data quality tasks, though full evaluative judgment remains largely manual. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by automatically surfacing data quality flags, suggesting outliers, and summarizing metadata, helping statisticians focus their expertise on interpreting limitations; however, the assistance is partial and does not transform the core judgment-intensive aspects of evaluation. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can significantly speed up initial data profiling, anomaly detection, and literature/source cross-referencing, meaningfully augmenting a statistician's evaluation process while the human retains final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can flag some data quality issues (missing values, outliers, schema mismatches) automatically, evaluating reliability and usability requires domain expertise, understanding of data provenance, and judgment about whether limitations matter for a specific analytical purpose—tasks that currently demand human oversight and cannot achieve 50% time savings end-to-end. |
| Task automatability | claude-sonnet-5 | 2/5 | This requires nuanced judgment about data provenance, methodology quality, and domain-specific validity that current AI can assist with but not reliably perform end-to-end without significant human oversight., especially for novel or complex data sources. |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | While there is no formal licensing requirement for this specific evaluation task, organizational processes and statistical best practices create moderate friction; liability for downstream decisions based on faulty source assessment creates pressure to retain human accountability and review. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement mandates a human perform this specific evaluative task, though professional norms in research and regulated industries (e.g., clinical trials) create some expectation of expert judgment. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Automated data quality checks are inexpensive, but they address only a fraction of the evaluation task; the human expert still dominates the cost for the complete judgment, making the all-in AI solution not substantially cheaper than employing a statistician. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can cheaply run automated checks, the residual need for expert human review to catch subtle validity and reliability issues means overall cost savings are modest, not order-of-magnitude. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Data validation and quality assessment tools exist, but they are typically narrow (detecting technical anomalies) and require significant human configuration and interpretation; no deployed product reliably evaluates the full spectrum of limitations (bias, representativeness, source credibility) that statisticians must assess. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools can flag some data quality issues (missing values, inconsistencies) but no deployed product reliably evaluates source reliability and usability at the level a trained statistician does across varied domains. |
Adapt statistical methods to solve specific problems in many fields, such as economics, biology, and engineering.
34CI 30–38 · exposure 25 · augmentation 75 · importance 4.3/5 · click for rater detail
Adapt statistical methods to solve specific problems in many fields, such as economics, biology, and engineering.
34| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption of AI assistants for statistics is emerging (code generation, visualization) but remains slow in production settings where methodological rigor is paramount. Most organizations still employ statisticians to make adaptation decisions rather than relying on AI recommendations. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Sectors employing statisticians (finance, biotech, tech) are relatively fast adopters of AI tools for coding and analysis, but true methodological adaptation remains a pilot-stage use case. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI systems can substantially augment statisticians by suggesting applicable methods, generating exploratory analyses, translating between domain terminology, and rapidly prototyping alternative approaches, allowing the human statistician to focus on problem diagnosis and validation rather than manual implementation. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI substantially helps statisticians by suggesting methods, generating code, drafting model formulations, and explaining techniques from adjacent fields, meaningfully speeding up the adaptation process. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Adapting statistical methods to novel field-specific problems requires deep domain knowledge, creative problem-solving, and judgment about which methods fit a context. Current AI can suggest standard approaches or generate code for known methods, but cannot reliably diagnose the underlying problem structure or justify novel methodological choices across diverse domains at equal quality. |
| Task automatability | claude-sonnet-5 | 2/5 | Adapting statistical methods to novel domain-specific problems requires judgment about assumptions, causal structure, and domain context that current AI cannot reliably automate end-to-end, though it can assist with parts of the workflow. |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Organizations value expertise and peer review in statistical methodology choices, especially in high-stakes domains like clinical trials or regulatory submissions. Liability concerns and professional standards create moderate friction against full automation, though no hard legal bar exists. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement generally restricts this work, but organizational trust and the cost of statistical errors in fields like biology or engineering create moderate friction against pure automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI can reduce coding time for implementation, but the methodological consultation and domain-specific problem diagnosis that constitute the core value remain primarily human work. Integration overhead and quality assurance mean the total cost remains comparable to or exceeds the cost of a statistician's time on this task. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can cheaply generate code or draft analyses, the need for expert validation and correction of methodological choices keeps effective cost closer to human-comparable for nontrivial adaptation tasks. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI tools can generate statistical code and suggest methods given explicit problem descriptions, no deployed product reliably performs end-to-end adaptation of statistical methods to novel cross-domain problems in production. Existing tools require significant human direction and validation of the adaptation choice itself. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI coding/analysis assistants can suggest and implement standard statistical methods, but no deployed product reliably performs creative adaptation of methodology to new problem domains without expert oversight. |
Determine whether statistical methods are appropriate, based on user needs or research questions of interest.
33CI 25–41 · exposure 25 · augmentation 75 · importance 4.5/5 · click for rater detail
Determine whether statistical methods are appropriate, based on user needs or research questions of interest.
33| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption of AI for statistical method selection remains limited; most organizations still rely on human statisticians or in-house review processes. While some tech and data science teams experiment with AI-assisted suggestions, production-scale displacement is rare and slow in regulated and research-intensive sectors. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Statisticians work in research, pharma, and business analytics sectors where AI coding/analysis assistants are being piloted and increasingly used, but full trust in method selection remains limited. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI excels at augmenting statisticians by rapidly generating candidate methods, surfacing papers and precedents, and checking assumptions against datasets—tasks that accelerate decision-making while the human retains judgment authority. LLMs and statistical engines already provide meaningful productivity gains in this advisory capacity. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can effectively suggest candidate methods, flag assumptions, and cite relevant literature, meaningfully speeding up a statistician's evaluation process while the human retains final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can recognize common statistical methods and suggest approaches for standard problems, determining appropriateness requires deep understanding of research context, assumptions, and domain constraints that typically need human judgment. Current systems lack the nuanced contextual reasoning to independently assess fitness-for-purpose at equal quality without significant human oversight. |
| Task automatability | claude-sonnet-5 | 2/5 | This requires judgment about research design, data quality, and domain context that current AI can assist with but not reliably determine independently, especially for novel or ambiguous research questions.5 |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Statistical method selection carries material liability for downstream analysis validity; organizations typically require a credentialed statistician to sign off on methodological choices, especially in regulated fields (clinical, regulatory). Professional norms and liability asymmetry create strong organizational friction against pure automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement mandates a human statistician for this specific judgment, though downstream consequences (publication, regulatory submissions) create moderate incentive for human validation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Integration of AI advisory systems requires setup, validation, and human review, making the all-in cost comparable to or exceeding a statistician's time for this core task. The cost of errors in method selection can be high, requiring expensive human verification afterward. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI can quickly generate suggestions cheaply, but the need for expert review to catch errors in method selection narrows the effective cost advantage. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI tools can recommend statistical methods and flag potential issues, but no deployed product reliably performs the full task of determining appropriateness end-to-end. Products exist (statistical advisory systems, LLM-based helpers) but operate with material gaps in handling edge cases, unusual research questions, or domain-specific requirements. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools can suggest statistical methods given a described problem, but no deployed product reliably validates method appropriateness against nuanced user needs without significant human oversight. |
Evaluate the statistical methods and procedures used to obtain data to ensure validity, applicability, efficiency, and accuracy.
31CI 30–32 · exposure 25 · augmentation 75 · importance 4.7/5 · click for rater detail
Evaluate the statistical methods and procedures used to obtain data to ensure validity, applicability, efficiency, and accuracy.
31| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Statistical organizations and pharma/biotech firms are exploring AI-assisted code review and anomaly detection, but deep adoption of AI for *evaluating* methodology validity remains limited. Most adoption remains at the pilot or assistant stage rather than replacement. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Data science and analytics teams increasingly use AI-assisted coding and QA tools, but adoption for core methodological evaluation remains at pilot stage rather than widespread production use. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI excels at augmenting this task: it can quickly flag missing-data patterns, suggest alternative methods, surface potential confounds, and generate diagnostic plots, raising a statistician's speed and coverage while they retain decision authority. LLMs can also help document and justify methodological choices, creating a strong assistive loop. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist by checking code, flagging statistical anomalies, suggesting alternative methods, and summarizing literature, significantly speeding up a statistician's evaluation process while they retain final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can assist with basic validation checks (e.g., identifying missing data, outliers, summary statistics) but evaluating the *validity* and *applicability* of statistical methods requires domain expertise, understanding of study context, and judgment about appropriateness that current systems struggle with reliably. The task's emphasis on ensuring validity and applicability—not just mechanically flagging issues—places it below the 50% time-saving threshold. |
| Task automatability | claude-sonnet-5 | 2/5 | Evaluating statistical methodology requires domain judgment about study design, confounders, and applicability that current AI cannot reliably replicate end-to-end, though it can assist with parts of the review.dea |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Professional responsibility and liability create moderate friction—a statistician's name often goes on methodological validation, and organizations are cautious about automating judgment that affects downstream conclusions. However, no explicit legal requirement mandates human sign-off, so barriers are not absolute. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No formal licensing requirement in most contexts, but organizational reliance on statistician sign-off for high-stakes decisions (e.g., research validity, regulatory submissions) creates moderate friction against pure AI substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI inference is cheap, but meaningful evaluation requires substantial human oversight and correction, adding labor costs that approach or exceed a junior statistician's hourly rate for tasks requiring real validation work. Integration and quality-assurance burden is non-trivial. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Human statistician expertise is still needed for nuanced judgment; AI assistance reduces some time but doesn't eliminate the need for costly expert oversight, so savings are modest relative to expert wages. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While LLMs and code-checking tools can surface syntax errors and suggest method alternatives, no deployed product reliably performs end-to-end evaluation of statistical methodology with the nuance required. Existing tools lack deep understanding of domain context and can miss subtle methodological flaws that a trained statistician would catch. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools (e.g., code review assistants, statistical QA checkers) can flag some errors or common pitfalls, but no deployed product reliably performs full methodological evaluation of data collection procedures in production. |
Plan data collection methods for specific projects, and determine the types and sizes of sample groups to be used.
30CI 25–35 · exposure 30 · augmentation 75 · importance 3.8/5 · click for rater detail
Plan data collection methods for specific projects, and determine the types and sizes of sample groups to be used.
30| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While statistical software has high penetration, true AI-driven automation of sampling design is nascent. Most organizations still rely on experienced statisticians for design decisions; adoption of pure AI automation remains limited and concentrated in low-stakes or routine tasks. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Statistics-heavy fields (research, pharma, social science) are only beginning to pilot AI-assisted study design; production use for full method planning is rare. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI tools demonstrably assist statisticians by automating sample size calculations, suggesting design alternatives, and flagging potential issues. This augmentation significantly speeds up the planning process while the human statistician retains critical judgment and oversight. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools substantially help with literature review, power calculations, and drafting methodology sections, meaningfully speeding up a statistician's planning process while they retain control. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can suggest sampling methods and calculate sample sizes given clear parameters, designing data collection requires domain expertise, understanding of project context, and iterative refinement that typically involves human judgment. Current AI tools offer limited end-to-end automation without substantial human oversight and rework. |
| Task automatability | claude-sonnet-5 | 2/5 | This requires domain judgment about study objectives, confounders, feasibility constraints, and statistical power tradeoffs specific to a project; AI can support calculations but cannot autonomously design a valid sampling and collection plan end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | In regulated domains (clinical trials, official surveys), regulatory frameworks require qualified statisticians to design and sign off on data collection methods. Professional licensing, liability for design errors, and organizational requirement for human statistical expertise create strong adoption barriers. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement generally, but methodological errors have high downstream costs (invalid studies, wasted funding), creating organizational insistence on expert sign-off. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI tools reduce the computational overhead but do not eliminate the need for expert statistician review, validation, and refinement. The integrated cost of AI suggestions plus required human expertise remains comparable to or higher than direct expert work, particularly for novel or complex designs. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply run sample-size formulas, but the bulk of the task—context-specific design choices and validation—still requires expensive expert time, keeping overall cost comparable to human-led work. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Products like statistical software with built-in sampling calculators and some AI-assisted analytics platforms exist and can perform parts of this task, but they require significant human input to specify study design, constraints, and objectives. Fully automated design without material error or scope limitations is not standard in production. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Tools exist for power analysis and sample size calculation, but no deployed product reliably designs full data collection methodologies and sampling strategies without expert oversight. |
Design research projects that apply valid scientific techniques, and use information obtained from baselines or historical data to structure uncompromised and efficient analyses.
29CI 25–32 · exposure 25 · augmentation 75 · importance 4.3/5 · click for rater detail
Design research projects that apply valid scientific techniques, and use information obtained from baselines or historical data to structure uncompromised and efficient analyses.
29| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Academic and regulated research sectors have been slow to adopt AI for core design decisions; adoption remains at the pilot/assistive stage rather than production replacement of statistician roles. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Research-intensive sectors (pharma, academia, tech) are adopting AI assistance for literature review and design brainstorming at a moderate pace, but full design authority remains human-led in most organizations. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can meaningfully assist by generating candidate designs, identifying methodological pitfalls in drafts, and suggesting alternatives for statistical technique selection, substantially raising a statistician's productivity while they retain final judgment. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist by summarizing historical data, suggesting design frameworks, flagging potential biases, and speeding up literature synthesis, substantially aiding the statistician's productivity while they retain final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with data manipulation and suggest statistical techniques, designing uncompromised research projects requires deep domain expertise, causal reasoning, and judgment about scientific validity that current systems cannot reliably perform end-to-end. The task involves synthesizing baselines, historical context, and methodological constraints in ways that exceed routine analysis. |
| Task automatability | claude-sonnet-5 | 2/5 | Designing a valid, uncompromised research study requires judgment about confounders, domain context, and study goals that current AI can support but not reliably originate end-to-end without expert oversight.: |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Research integrity, IRB approval, and regulatory requirements in clinical/social science contexts typically require a qualified human statistician's professional judgment and sign-off, creating legal and accountability barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement universally mandates a human statistician, but institutional review boards, funding bodies, and scientific norms create strong expectations of qualified human oversight for methodological validity. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Integration, quality oversight, and expert review of AI-generated designs would still require significant human investment, making the all-in cost comparable to or exceeding direct human design work for high-stakes research. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can cheaply generate draft designs, the cost of expert review and correction to avoid flawed studies keeps the effective cost comparable to or only modestly cheaper than a human statistician doing it directly. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed products perform full research design reliably in production. Current AI can generate boilerplate analysis plans or suggest techniques, but real statisticians report needing substantial manual refinement and expert oversight to ensure scientific rigor and validity. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools can suggest study designs or statistical approaches, but no deployed product independently designs rigorous research projects in production without a statistician validating assumptions and methodology. |
Develop and test experimental designs, sampling techniques, and analytical methods.
28CI 25–30 · exposure 25 · augmentation 75 · importance 4.4/5 · click for rater detail
Develop and test experimental designs, sampling techniques, and analytical methods.
28| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption of AI for core methodological design remains limited in academic and research settings, which prioritize human expert judgment and professional accountability. Most statistics teams use AI for routine calculations and literature search, not for autonomous design innovation. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Statistics-heavy fields like biostatistics, market research, and academia are adopting AI coding and drafting tools, but deep production-level automation of experimental design and method development remains uncommon and largely pilot-stage. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can substantially augment statisticians by rapidly generating candidate designs, comparing approaches against published standards, simulating power analyses, and flagging potential methodological pitfalls. This leaves final validation and novel innovation to human experts while accelerating the exploratory phase. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools significantly speed up code generation, literature review, boilerplate design templates, and exploratory analysis, meaningfully augmenting a statistician's workflow while the human retains responsibility for design validity. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Developing novel experimental designs and analytical methods requires substantial human creativity and domain judgment. While AI can assist in literature review and suggest standard techniques, creating new designs that account for specific research constraints and generating defensible methodological innovations remains primarily a human task. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can assist with drafting designs and running standard power calculations, but developing novel experimental designs and validating analytical methods for specific research contexts requires domain judgment and iterative refinement that current tools cannot fully replace end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong professional and regulatory barriers exist: research institutions require credentialed statisticians to sign off on experimental designs, institutional review boards must approve human-subject research, and liability for flawed methodology rests with responsible human statisticians. Legal and reputational risk prevents full delegation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No formal licensing requirement exists for statisticians in most contexts, but organizational reliance on correct methodology (especially in regulated fields like clinical trials or econometrics) creates strong incentives for human sign-off and accountability. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI assistance in design generation and literature synthesis is inexpensive, but the core task of methodological development still requires expert statisticians. The labor cost remains high relative to AI tool expenses, and oversight of AI suggestions is substantial. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can cheaply generate boilerplate code or suggest standard designs, the cost of expert review, validation, and correction for non-standard or high-stakes designs keeps overall cost comparable to or only modestly cheaper than a skilled statistician. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Current AI systems can generate and evaluate standard experimental designs through pattern matching on published work, but they lack the ability to reliably validate complex sampling techniques or novel analytical methods against real-world constraints. Production systems do not independently develop experimental designs in research settings. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some statistical software and AI coding assistants can suggest sampling frameworks or generate code for standard designs, but no deployed product reliably develops and validates novel experimental designs and analytical methods without expert oversight. |
Examine theories, such as those of probability and inference, to discover mathematical bases for new or improved methods of obtaining and evaluating numerical data.
28CI 20–35 · exposure 20 · augmentation 63 · importance 3.4/5 · click for rater detail
Examine theories, such as those of probability and inference, to discover mathematical bases for new or improved methods of obtaining and evaluating numerical data.
28| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Even in highly digitized academic and professional settings, theoretical discovery remains driven by human researchers; adoption of AI for autonomous method innovation is minimal and confined to narrow exploratory tasks rather than end-to-end discovery workflows. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Academic and research statistics adopts AI tools slowly for foundational theoretical work, though there is growing use of AI as a brainstorming or literature aid. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can usefully augment this work by automating literature search, symbolic computation, generating exploratory visualizations of mathematical relationships, and flagging candidate approaches—meaningful but partial support that leaves the core theoretical insight to the human statistician. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist by surfacing related literature, checking derivations, suggesting proof strategies, and simulating scenarios, boosting researcher productivity substantially. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist in exploring mathematical theories and generating candidate statistical methods through pattern recognition, the creative discovery of genuinely novel theoretical foundations requires deep insight into mathematical structure and validation against established principles—tasks requiring sustained human reasoning that current AI cannot reliably perform end-to-end. |
| Task automatability | claude-sonnet-5 | 2/5 | This is open-ended theoretical research requiring original mathematical insight and novel proof construction, which current AI can assist but not reliably generate independently at expert quality.ed |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Peer review, publication standards, and institutional credibility create strong barriers: novel statistical methods must be vetted by the research community and attributed to qualified researchers, and mathematical innovations require human authorship and accountability for correctness. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement, but peer review, academic credibility, and publication norms create moderate friction against AI-authored theoretical claims being accepted. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI computational costs for sustained mathematical exploration, combined with human oversight required to validate theoretical soundness, remain comparable to or exceed the cost of a human statistician exploring the space directly. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply generate exploratory text or code but cannot substitute for the expert labor of theoretical development, so cost comparison favors humans for real output quality. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably discovers new statistical methods or mathematical bases for data evaluation. AI tools can support symbolic math and literature review, but autonomous discovery of validated, peer-acceptable theoretical advances remains research-stage with unproven production reliability. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously develops new statistical theory or inference methods; this remains research-stage even for advanced LLMs and math-AI systems like AlphaProof-type tools. |
Supervise and provide instructions for workers collecting and tabulating data.
16CI 5–26 · exposure 13 · augmentation 50 · importance 3.0/5 · click for rater detail
Supervise and provide instructions for workers collecting and tabulating data.
16| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Supervision and worker management remain deeply relational and context-dependent; sectors employing statisticians (government, research, professional services) have not adopted AI systems to replace supervisors, and cultural/legal norms strongly favor human oversight of workers. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Statistics and research organizations are adopting AI for data processing but human supervisory management structures remain largely unchanged and slow to shift. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist a human supervisor by auto-generating status reports, flagging data quality anomalies, or suggesting protocol refinements, moderately reducing administrative burden. However, augmentation is limited because interpersonal instruction and judgment remain central to the task. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help by tracking task progress, flagging data quality issues, and drafting instructions, but the core supervisory judgment and interpersonal management remain human-driven. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | This task involves human supervision, instruction-giving, and worker management—functions requiring contextual judgment, interpersonal adaptation, and real-time responsiveness. While AI can assist with creating data collection protocols or automating status reports, it cannot reliably supervise humans or provide nuanced, adaptive instructions across varied field conditions. |
| Task automatability | claude-sonnet-5 | 1/5 | Supervising human workers and giving them instructions is inherently a management/interpersonal function requiring real-time judgment, accountability, and adaptive communication that current AI cannot perform end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Legal and organizational liability for worker safety, compliance, and performance rests on identified human supervisors in most jurisdictions. Replacing human supervision creates ambiguous accountability and likely violates labor management standards, creating strong regulatory and institutional barriers. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement specifically, but organizational structures, accountability for worker oversight, and HR/legal responsibilities create real friction against full automation of supervision. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The cost of deploying AI for supervision (custom training, integration with HR/project systems, continuous monitoring, human oversight of AI decisions) exceeds the marginal benefit over a human supervisor, particularly given the need for accountability and relationship-based instruction. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no AI substitute performing this supervisory function, so no meaningful cost comparison favors AI; a human manager is still required. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed AI system reliably supervises and instructs human workers in production environments. Chatbots can generate templated instructions, but they cannot manage real-world team dynamics, handle exceptions, provide responsive feedback, or maintain accountability—core elements of supervision. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product supervises human data-collection teams autonomously; management software assists but does not replace the supervisory role. |
Related occupations — Computer & Mathematical
How to read this
A high substitution score does not mean this job disappears — it means a large share of its current tasks face replacement pressure, so the mix of tasks is likely to change. High augmentation alongside substitution typically means the occupation reorganizes around the protected tasks. Wide confidence intervals mean the rater panel disagreed: treat those scores as open questions, not verdicts.
What would change this score
New model capabilities (automatability, feasibility), falling inference costs (cost ratio), regulation and licensing shifts (barriers), and measured sector adoption (velocity) all re-enter at every index release. Each release is recomputed, versioned and kept queryable — scores are claims with a date on them, not permanent labels.