The statistician who ships something colleagues use
$213,610top of the range in Michigan · middle $105,650 / yr
AI is transforming this role
Statisticians in the United States earn a median of $105,650 a year. Pay starts near $64,000. Pay reaches $213,610 at the top of the range in Michigan, the best-paying state for this work among those with at least 500 people in the job.
Source: U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025 (Statisticians, SOC 15-2041). Last checked 9 September 2026.
Entry level
$64,000
Top of the range · Michigan
$213,610
Education
Master's degree in Statistics
Wages — U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025 (Statisticians). Top of the range is the highest state-level figure among states with at least 500 people in the job. AI-impact rating is PayCrunch's editorial assessment. Updated September 2026.
🆕 New & Trending AI Tools for StatisticianReviewed September 2026
We track new AI-tool launches every week and refresh this list — here’s what’s gaining traction for Statistician work right now.
Julius AINEWFree / $20 mo
AI data analyst that runs statistics and charts from plain-language prompts.
How a Statistician uses it: analyze datasets and generate figures without writing code
NotebookLMNEWFree / $7.99 mo
Google tool that answers questions grounded only in the documents you give it — with citations.
How a Statistician uses it: load your own manuals, policies, or PDFs and ask questions that stay accurate to the source
ElicitFree / $12 mo
AI research assistant that finds and summarizes papers.
How a Statistician uses it: run a literature review and extract findings across dozens of papers fast
ConsensusFree / $9 mo
AI search that answers questions from peer-reviewed research.
How a Statistician uses it: get evidence-backed answers with the studies behind them
SciSpaceFree / paid
AI that explains papers and helps with literature review.
How a Statistician uses it: decode dense papers and trace citations quickly
SciteFree / $20 mo
Shows whether other studies support or contradict a paper's claims (Smart Citations).
How a Statistician uses it: check if a finding is actually backed by the wider literature before you cite it
ChatGPTFree / $20 mo
The most-used AI assistant — writing, analysis, research, and images from a plain-language chat.
How a Statistician uses it: draft emails and documents, summarize long files, and get instant answers to on-the-job questions
ClaudeFree / $20 mo
AI assistant known for careful writing, long-document analysis, and coding.
How a Statistician uses it: analyze big reports or spreadsheets and turn messy notes into clean, finished writing
Google GeminiFree / $20 mo
Google's AI assistant, built into Gmail, Docs, and Search.
How a Statistician uses it: draft and reply inside Google Workspace and research without leaving the page
When the decision depends on whether the number holds
A statistician is the person an organization calls when a decision depends on whether a number will still look true tomorrow. A public-health team wants to know if a program changed anything. A manufacturer wants to know if a change in a process is real or noise. A product group wants to know if an experiment on a website is large enough to trust. The statistician does not hand over a bigger adjective. The statistician hands over a result, a plain account of how it was reached, and a clear statement of what would make it fall apart.
The raw material is other people's measurements. Surveys, clinical records, sensor logs, administrative files, and experiments all arrive with quirks: missing fields, changed definitions, and samples that do not match the population someone hopes to describe. Your first loyalty is to those limits. A beautiful model that ignores how the data were collected is a story, and organizations already have plenty of stories. They hire you because you will disappoint them accurately when the evidence is thin, and because you will say so early enough that they can collect better evidence.
The rooms vary. Federal and state statistical programs, university labs, hospital research groups, insurers, manufacturers, technology firms, and consulting practices all employ statisticians. In some rooms you design the study before any data exist. In others you inherit a file and must say what it can and cannot support. The public rarely sees the work. A policy memo, a drug development meeting, or a pricing decision quietly depends on it. That distance from the spotlight is fine. The craft is the reasoning, not the applause.
A methods week, from messy file to plain answer
A typical week starts with a conversation that is still fuzzy. A collaborator knows the decision and does not yet know the measurement. You slow that conversation down. What would count as a change worth acting on? Who is in the group being described? What was collected, and what was only hoped for? You write those answers down, because later arguments often come from two people using one word for two ideas. A morning spent on definitions saves a month of rework.
Then you touch the file. You look for duplicated records, impossible values, and dates that drift. You compare the sample you have with the population someone wants to talk about. You choose an approach that matches the design, not the approach that is fashionable. You fit it, you check whether the result depends on one fragile choice, and you try the alternative that a skeptical colleague will ask about. Software matters, and so does the notebook that records what you ran. A result nobody can reproduce is not a result you should defend in a meeting.
The last part of the week is translation. You sit with a scientist, a regulator, a product manager, or an executive and explain uncertainty in words they can repeat without distorting them. You distinguish a hint from a finding. You say what a larger or better-designed collection of data might change. You decline to produce a false precision. Strong statisticians are invited back because leaders learn that your caution protects them. Weak ones are remembered for a chart that collapsed the first time someone asked how the sample was drawn.
Graduate study, and a society rather than a licence
A graduate degree is common in this field. Employers posting a statistician role often want a master's or a doctorate in statistics, biostatistics, or a closely related quantitative subject. A bachelor's degree can open adjacent analyst jobs, and some people move from those jobs into a statistician title after further study. The degree is not a formality. Coursework in probability, inference, and study design is what lets you recognize a flawed plan before the data arrive. Programming is expected alongside the theory. The hiring manager wants both, and will notice if one of them is only a claim.
There is no licence that authorizes you to work as a statistician. The American Statistical Association is a professional society. Joining it, attending its meetings, or reading what it publishes can be part of a serious career. Membership is a society affiliation, separate from a licence, and it does not grant you a legal scope of practice. Nobody at a hiring panel can demand an association card the way a state board demands a professional licence. They can demand a graduate record, a clear analysis, and references who have seen you tell the truth about a weak result.
Society, not a permit
Treat the American Statistical Association as a community of practice. It can sharpen your judgment and widen your network. It is a different thing from a state licence, and this occupation does not run on a state licence.
Preparation, then, is school plus proof. Keep coursework projects that you can explain without exposing private data. If you work with a professor or an employer, ask to own a piece of the analysis end to end: the design conversation, the cleaning, the method, and the paragraph a non-specialist can read. Internships in research groups, public agencies, and companies do that faster than another certificate with a vague name. When you apply, lead with the decision your analysis informed and the limit you refused to hide. That story is the credential the degree is supposed to make possible.
How junior roles grow into senior ones
Early roles are supervised. You clean files, run analyses a senior statistician designed, and draft the first version of a methods note. The learning is in the red ink. Why was this comparison unfair? Why did this missing-data choice change the conclusion? People who want only the final chart get impatient here. People who want the judgment collect every correction. Promotion out of this stage comes when a senior colleague trusts you to design a modest study and to know when you are past your depth.
The middle of the career is independent work with real collaborators. You are the statistician on a project, not the assistant to one. You negotiate the design, you defend the method in review, and you write so the finding can travel without you in the room. You may specialize: clinical trials, survey work, industrial experiments, reliability, or the messy observational files that governments and companies actually have. Specialization raises your value when it is deep. It traps you when you cannot explain it to the next field over. Keep one foot in plain language.
Later titles are senior statistician, principal, lead, or manager of a methods group. Some people stay individual experts because their taste for hard problems is the asset. Others lead, which means hiring, reviewing, and protecting junior staff when a powerful client wants a softer conclusion than the data allow. Universities add a research and teaching path with its own pressures. Consulting adds variety and a client clock. None of these paths requires a licence. All of them require a record of analyses that survived contact with a skeptic. Pay follows that record, the sector, and the market, which is what the next section is for.
May 2025 wages, with Michigan and the District apart
The figures here are Occupational Employment and Wage Statistics for May 2025 for Statisticians. Someone early in the published distribution is at $64,000. The national median, the middle of that distribution, is $105,650. The step between those two is $41,650. Use the median as the center of a full-year talk, and use $64,000 as the early marker, often a supervised role or a market where the middle itself is lower.
The top figure is $213,610. That number is the high end of the published range in Michigan. The highest state median is in a different place: the District of Columbia, at $140,670. A range top and a median are different statistics, and Michigan and the District of Columbia are different places. Do not let a recruiter slide from one to the other. From the national median up to the Michigan range top, the spread is $107,960. From the national median up to the District of Columbia median, the difference is $35,020. The first spread describes how high the published range goes in Michigan. The second describes how far one city's middle sits above the national middle.
State medians worth setting beside the national figure, in this order, are the District of Columbia at $140,670, New York at $136,020, California at $135,960, Maryland at $132,620, and New Jersey at $118,280. The District of Columbia holds the highest median. Missouri holds the lowest, at $66,330. The gap between that highest median and that lowest median is $74,340. Missouri's median sits close to the national entry figure of $64,000, which is a useful warning: a median in one place can resemble an entry wage on the national chart. New York, California, Maryland, and New Jersey all show medians above the national $105,650, and none of those medians is the Michigan range top.
Negotiating with a long upper range
If an offer is near $64,000, you are at the entry figure. The $41,650 distance to the national median is the climb you can discuss. Ask what, in this organization, moves a statistician across it: designing studies rather than only running them, being the named author of the methods, reviewing other people's work, or carrying a specialty the group lacks. If the offer is in Missouri, remember that the state median itself is $66,330, so a figure near $64,000 may be close to that place's middle even while it looks like national entry. Say that distinction out loud. It changes whether you are underpaid for the market or priced at the local middle.
If the offer is near $105,650, you are at the national middle. Further movement is either a higher-median place or a senior scope. The District of Columbia median of $140,670 sits $35,020 above the national median. New York at $136,020, California at $135,960, Maryland at $132,620, and New Jersey at $118,280 are the other middles to use if those are the places you might actually work. Michigan's $213,610 belongs in a separate sentence. It is the high end of the published range there. It is a poor description of a typical offer, in Michigan or anywhere else, and it is not a stand-in for the District of Columbia's median.
The $74,340 gap between the District of Columbia and Missouri is your answer when someone says the title pays roughly the same in every market. It is a gap between medians. It does not measure the distance from Missouri up to Michigan's range top, and you should not pretend you have that subtraction. Compare an offer with the entry figure, the national median, and the median of the place you will live. Then ask whether bonus or consulting revenue is being waved in place of base. A good year that sometimes approaches the upper range is a different promise from a base near $105,650 or near $140,670.
Bring a one-page example of an analysis you can discuss, with private data removed, and bring these anchors: $64,000, $105,650, $213,610 in Michigan as a range top, $140,670 as the District of Columbia median, $66,330 as Missouri's median. Ask the employer which anchor the role matches, and why. A junior supervised post matches the early figure. An independent statistician in a high-median city can argue from that city's median. A principal role with a long record can talk about the upper part of the range without quoting $213,610 as if it were the offer. The society you may join will not set this pay. Your degree will not set it either. The work you can defend, and the market you are standing in, will.
The top of Statistician pay — and how to get there with AI
$213,610what Statistician pay reaches in Michigan
Highest state-level top-of-range annual wage for Statisticians, among states with at least 500 people in the job. U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025.
And the role it leads to — Mathematicians — reaches $195,190 nationally.
$64,000entry$105,650middle$213,610top end
Plenty of people in this field write correct analyses; the ones at the top of the range have written something the group installs, cites in its protocols and cannot easily do without.
Providing consultation to clients and colleagues, reviewing research protocols and recommending appropriate analyses, and preparing tables and graphs for reports are demand-driven tasks with no top end on the demand. Every statistician handles them by answering one request at a time. A few turn the most repeated request into working software with tests, documentation and a version number, and those few become structural rather than helpful. Code assistance has made packaging and documenting far cheaper than it was, which means the barrier is now deciding what deserves to exist, not finding the hours to build it.
Your playbook, by where you are now
Just startingTurn one repeated consultation into a script
Log every consultation request for three months with the question behind it and the minutes it consumed.
Take the one you answer most, usually a design and sample size question, and write it as a single tested function in Python.
Put everything in Git from the first commit, throwaway analysis included, so your working history is inspectable.
Write the analysis plan before the code even for a small request, and store the two together.
Rebuild one legacy IBM SPSS Statistics workflow in Python and reconcile the output line by line before you trust either.
What proves it: A version-controlled function with a test that reproduces a published worked example.
Realistic span: the first two years
A few years inPackage it so somebody else can run it
Turn the scripts into an installed package with documentation, a version number and a changelog, so colleagues stop copying files around.
Build a small front end, in Qlik Tech QlikView or in a notebook investigators can open, so a feasibility question is answered before a meeting is booked.
Ask Claude to draft the documentation and a tutorial from your source, then check every argument and default against the code itself.
Push the resampling and simulation work onto Apache Spark so a method question is not limited by one laptop.
Hold a standing clinic hour and use the questions that arrive to decide what the package should do next.
Write your analysis descriptions so the tool's output can be pasted straight into a protocol without editing.
What proves it: An installed package with documentation, tests and named users outside your own team.
Realistic span: years three through seven
ExperiencedMake the tool part of the process
Require its output in the analysis section of research proposals and grant applications, so the standard is written into how work begins.
Hand maintenance to a second person and review their changes, because a tool with one owner is a liability rather than an asset.
Write the methodological support line into grant budgets, since funded support is what makes the work defensible at review.
Teach a graduate or continuing education module built on the tool, which forces the assumptions inside it into the open.
Weigh location, with Michigan heading the state table, and note that the mathematics roles nearby reward the same method-building instinct.
What proves it: A tool cited in your organisation's protocols and maintained by more than one person.
Realistic span: year eight onward
The next 90 days
Spend the next ninety days logging consultations before you build anything. Question asked, who asked, how long it took, and what you actually had to look up. Thirty entries is enough. One question will dominate, and for most statisticians it is some version of how many observations are needed for a proposed design. Write that answer once, properly, as a tested function with the assumptions named as arguments rather than buried in comments. Put it in Git, document the three decisions a reviewer would challenge, then sit with two colleagues and watch them use it. Their confusion tells you exactly what your documentation lacks. Fix that, then send it to the group with a short note about what it replaces.
Wage figures: BLS OEWS, May 2025. The playbook is PayCrunch editorial guidance, not a guarantee of pay or placement.
Every figure is the national median from the U.S. Bureau of Labor Statistics (OEWS) shown on that role’s own page.
Never used AI before? Start here (2 minutes).
Start by wiring an AI pair-programmer into your editor. Put GitHub Copilot or Claude next to your R or Python (in Positron/RStudio or VS Code) and use it to write, refactor and debug analysis code as you go. Prove out every suggestion on simulated data where you know the right answer.
For method questions and explanation, use Claude or ChatGPT to walk through a technique, derive a likelihood, or translate a model between R and Python - then verify against a textbook or the documentation. These are free or low-cost; keep any confidential or regulated data out of them.
The one rule, forever: Statistics is where AI's confident wrongness does the most damage. LLMs suggest inappropriate tests, mishandle assumptions and multiple comparisons, and generate plausible but invalid code - never report an estimate you haven't validated on simulated data with known truth, never let AutoML's accuracy substitute for causal and assumption checking, and keep confidential or regulated data (PHI, PII) out of consumer tools. You own the inference.
The plays — exact steps, exact prompts
Do these in order. Each one is copy-paste ready. You do not need to know anything about AI going in.
1
10x your R and Python with an AI pair
Why this pays: A statistician's throughput is gated by code - cleaning data, fitting models, making figures. An AI pair-programmer collapses that time, so you deliver more analyses and take on the harder, higher-value questions that separate a $106k analyst from a $214k lead.
GitHub CopilotClaudePositron / RStudio
1
Keep Copilot/Claude in Positron or VS Code to write tidyverse/pandas pipelines, build modeling functions and generate ggplot/matplotlib figures - reviewing every line.
2
Generate a full, assumption-checked modeling script.
Copy-paste this prompt
Write clean, commented [R tidyverse] code to: load [dataset], handle missing data via [multiple imputation with mice], fit a [mixed-effects logistic regression] with [random intercepts by site], check model diagnostics and assumptions, and produce a coefficient plot with confidence intervals. Explain each modeling choice.
AI-written stats code compiles but can be subtly wrong - validate on simulated data with a known answer before trusting it on real data.
What you'll haveFar more analyses delivered, freeing you for the high-value modeling that earns the lead-statistician salary.
2
Own causal inference and experimentation
Why this pays: Correlation is commoditized; causal inference is the scarce, top-paid skill in tech and pharma. The statistician who can design experiments and defend causal estimates - using AI to build the models and simulations fast - commands the best-paid roles in A/B testing, econometrics and clinical inference.
DoWhy / EconMLStan / PyMCClaude
1
Build causal analyses with DoWhy/EconML (DAGs, identification, refutation) and Stan/PyMC for Bayesian models; use Claude to scaffold the code and pressure-test your identification strategy.
2
Set up a defensible observational causal analysis.
Copy-paste this prompt
I want to estimate the causal effect of [feature X] on [retention] from [observational logs] with confounders [list]. Walk me through a DoWhy workflow: draw the DAG, state the identification assumptions, choose an estimator (matching, IPW, or double ML with EconML), and specify the refutation tests to try to break the estimate. Explain what would invalidate each assumption.
AI can code the estimator but can't vouch for your assumptions - the DAG and the untestable assumptions are your professional judgment, and the estimate is only as good as they are.
What you'll haveDefensible causal estimates and clean experiments - the scarce skill behind top-of-range tech and pharma statistics roles.
3
Use AutoML as a baseline, then beat it
Why this pays: AutoML is coming for the button-pushers, so out-flank it: use it to set a fast baseline, then add the domain structure, calibration and uncertainty it can't. The statistician who delivers a well-calibrated, interpretable model that beats AutoML keeps the high-value seat.
H2O.ai / DataRobotscikit-learnClaude
1
Run H2O AutoML to establish a benchmark quickly, then build a principled model (sensible priors, calibration, monotonic constraints, uncertainty) that improves on it where it matters.
2
Design an interpretable model that beats the black box.
Copy-paste this prompt
H2O AutoML gives me [AUC 0.82] on [a default-risk model] but it's a black box. Help me build a competitive, interpretable alternative: recommend a [monotonic gradient boosting or GAM] specification, how to calibrate the probabilities, how to quantify predictive uncertainty, and the fairness and stability checks a regulator would expect. Compare on the metrics that matter for [credit decisions].
A higher AUC is not a better decision - validate calibration, subgroup performance and stability, and make sure the model is defensible to regulators and stakeholders.
What you'll haveCalibrated, interpretable models that beat AutoML on the metrics that matter - the value that keeps you above the automation line.
4
Automate reporting, simulation and study design
Why this pays: Reproducible reports, power simulations and design documents eat a statistician's week. Automating them with AI-generated Quarto and simulation code frees you for interpretation and lets you support more studies - the productivity that scales your impact and pay, especially in trial-heavy pharma.
Quarto + ClaudeR simstudy / Python simulationSAS Viya
1
Generate parameterized Quarto reports and simulation-based power/sample-size code with Claude; standardize trial design docs (in SAS Viya environments where required).
2
Build a simulation-based sample-size analysis.
Copy-paste this prompt
Write [R] code using a simulation approach to determine the sample size for a [cluster-randomized trial] with [describe outcome, ICC, effect size, alpha, power], including how to handle [unequal cluster sizes]. Produce a power curve across effect sizes and a short methods paragraph I can paste into a protocol.
Simulation assumptions drive the answer - vary them and sanity-check the power curve; a mis-specified simulation gives a confidently wrong sample size.
What you'll haveReports, power analyses and design docs produced in a fraction of the time - capacity to support more studies and raise your value.
5
Translate statistics into decisions
Why this pays: The statistician who only produces p-values gets ignored; the one who turns analysis into a clear decision recommendation gets promoted. AI helps you build dashboards, explain uncertainty to executives and communicate risk - the influence that converts technical skill into a leadership, role at the top of the range.
ClaudeShiny / StreamlitJulius AI
1
Use Claude to translate results into plain-language, decision-focused narratives, and build interactive Shiny/Streamlit tools so stakeholders can explore scenarios.
2
Rewrite a technical result for decision-makers.
Copy-paste this prompt
Rewrite this statistical result for [a non-technical executive audience]: state the decision it informs, the recommendation, the uncertainty in plain language (avoid 'p-value' and 'statistically significant'), the key assumptions, and the risk of being wrong. Then give me 3 things they might misinterpret and how to preempt them. Result: [paste].
Simplify without distorting - never drop the uncertainty to make a cleaner story; misleading a decision-maker is the cardinal statistical sin.
What you'll haveAnalyses that drive decisions and get you in the room - the influence that turns statistical skill into top-of-range leadership pay.
Your 12-month sequence to the top of the range
How the plays above stack into a path from median pay toward the $213,610 tier.
Month 1
Wire Copilot/Claude into your R/Python workflow; validate every suggestion on simulated data.
Months 2-3
Build one end-to-end causal analysis with DoWhy/EconML and a Bayesian model in Stan/PyMC.
Months 3-6
Benchmark a project against AutoML, then ship a calibrated, interpretable model that beats it.
Months 6-9
Automate your reporting and power-simulation workflow with Quarto.
Months 9-12
Turn your analyses into decision tools and executive narratives - build the influence premium.
Gear for this job
As an Amazon Associate, PayCrunch earns from qualifying purchases. Links to books and tools are for the job on this page; we only recommend what we’d use in the work.
Same live O’Reilly 3rd already on data-scientist / python-developer. This page’s first play is 10x your R and Python with an AI pair and step 1 writes tidyverse/pandas pipelines. Not CompTIA Data+ and not leftover Ross Exam P (that is actuary).
Next steps for a Statistician
Some links below are affiliate or partner links. PayCrunch may earn a commission if you enroll or subscribe through them, at no extra cost to you. Wage figures on this page still come from the Bureau of Labor Statistics, not from these programs.
Statistician work is specific enough that a stamped 'check out these courses' block would be noise. BLS files this work as Statisticians (SOC 15-2041). O*NET Job Zone 5 is typical: graduate or professional school, so the honest next credential is a graduate-level or professional certificate — not a random catalog dump.
The occupation's listed knowledge area is Medicine and Dentistry, which is what the course searches below actually query.
Statisticians in this dataset list Amazon Redshift among the tools in use, so a program that names that stack is a better fit than a survey course.
Coursera search for medicine and dentistry — a graduate-level or professional certificate that lines up with computing, not a generic professional-development aisle.
FlexJobs screens remote, hybrid, freelance, and flexible listings so you are not wading through unverified ads. This is a job-board search for Statistician work, not a claim that they list a counted SOC 15-2041 inventory.
Write a Statistician resume, or one aimed at Mathematicians, instead of a blank template. Resume Now is a resume builder; we are not claiming a counted template set for this SOC.
A Statistician resume that names the actual tasks on this page, or the step-up title Mathematicians, beats a blank template when you apply.
What Statisticians earn by state
These are the Bureau of Labor Statistics’ own figures for Statisticians, state by state — not a cost-of-living adjustment applied to the national number. Only states employing at least 500 people in the occupation are shown, because a state median drawn from a handful of workers is noise rather than a signal.
District of Columbia
$140,670
highest of them · +33% vs the national median
Missouri
$66,330
lowest of the 19 states and D.C. that qualify · -37% vs the national median
The same job pays $74,340 more a year at the median in District of Columbia than in Missouri — 112% higher. That gap is what the Bureau measured, before any question of what it costs to live in either place. The top-of-range figure quoted at the head of this page, $213,610, is a different statistic in a different place: it is the 90th-percentile wage in Michigan. The state that pays the typical worker most and the state where the best-paid go highest are not always the same one.
Source: U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025, SOC 15-2041. 19 states and D.C. clear the 500-employee reporting floor for this occupation; those below it are left out rather than shown with a wide error band.
Free data. Use any of it.
PayCrunch publishes verified, BLS-sourced salary + AI-playbook data on 1,000+ professions — free, no signup.
It replaces button-pushing, not statisticians. Study design, causal reasoning, assumption-checking, uncertainty quantification and defending an inference stay human. Statisticians who move up the value chain (causal, Bayesian, communication) thrive; those doing routine model-fitting are the most exposed.
Can I trust AI-generated statistical code?
Only after validating on simulated data with known truth. LLMs pick the wrong test, mishandle assumptions and multiple comparisons, and write plausible-but-invalid code. You own correctness - reproduce every reported number independently.
Is it safe to paste my data into ChatGPT?
Not regulated or confidential data - no PHI, PII, or material non-public information in consumer tools. Use approved or enterprise environments with data agreements, and reserve consumer AI for code and public or simulated data.
What's the highest-value AI-adjacent skill for a statistician?
Causal inference and experimentation, plus clear communication of uncertainty - the things AutoML can't do and that pharma, tech and finance pay at the top of the range for. Being the person who defends the causal claim is where the money is.
How does this raise my pay?
By multiplying throughput and letting you specialize in scarce, high-value work (causal, Bayesian, biostatistics) and communicate it to decision-makers - the route to lead, principal and director-level statistics roles.
Methodology & sources
Salary (median, 10th, top of the range) — U.S. Bureau of Labor Statistics, OEWS.
By state — the Bureau of Labor Statistics’ own state medians, limited to states employing at least 500 people in the occupation. No cost-of-living arithmetic is applied to a wage anywhere on this page.
The plays — PayCrunch's own step-by-step guidance using publicly available AI tools. Tool names/URLs are real and current as of August 2026; prompts written to work as-is. Verify any professional output before relying on it.