Why so much of this job already runs through software
Statistical assistants work almost entirely inside files, spreadsheets and statistical packages. Computing means, totals and significance tests is routine work for software. So is tabulating survey responses and turning a clean table into a chart for a report. That is why the honest answer to “will AI replace statistical assistants” is not a flat no, and not a yes either.
On our measure of how much task time AI can handle today, this job scores 55 out of 100. The work that sits outside that number is the judgment around the numbers: deciding which test fits a study design, spotting when source data looks wrong rather than just unusual, and talking with a researcher or client about what the output should actually show. A model can run the test. It cannot take responsibility for the choice.
Pay and headcount matter here too. About 4,710 people held this job in the United States, with median pay of $50,330 a year, and the Bureau of Labor Statistics projects employment falling 1.8% between 2025 and 2035 (BLS, 2025). That is slow erosion in the number of seats, not a cliff. The jobs that remain tend to carry more of the thinking and less of the keying.
What AI does, what it helps with, and what stays with people
AI handles 41% of the task time in this role on its own. The clearest examples are computing descriptive and inferential statistics from a prepared dataset, and generating tables, charts and graphs from those results. Both are well-defined jobs with a checkable answer, which is exactly where current tools are strong. Our coverage method page explains how that share is built from task time.
A further 45% is shared work, where a person leads and the tool speeds things up. Coding open-ended survey answers into categories is one. Checking source data for errors and inconsistencies is another: a model can flag outliers and mismatches fast, but someone who knows how the data was collected decides whether a strange value is a mistake or a real finding.
That leaves 14% of the work with people. Discussing data presentation requirements with researchers, clients or managers sits there, because the task is agreeing what a question means before anyone computes anything. So does selecting the statistical procedure for a given study, and overseeing how a survey or data collection is actually run in the field. The physical part of the job, like handling paper forms and filed records, is small and the robotics section above rates it as fixed automation rather than anything general-purpose.
What the evidence shows, and what it does not
On how well AI performs against a qualified person in this job, our evidence grade is D. That is the grade we use when no study has measured AI output against a trained statistical assistant on this job’s real tasks. So we publish no parity number for this role, and you should treat any confident claim that AI already matches a human here with caution.
What would settle it is specific. A benchmark that hands the same messy, uncleaned dataset and the same research question to both a model and an experienced assistant, then has a statistician score the cleaning decisions, the test chosen and the final table. Until something like that exists, the fair reading is that AI is strong on computation and unproven on the judgment wrapped around it. Our methodology sets out how grades and scores are assigned.
Good to know: cost is not the barrier in this role, since the annual tooling range we track for the AI side sits well below the $19,940 to $45,420 range for the human side, so adoption is driven by trust and data access instead.
When the picture could shift
Most likely between 2034 and 2044 (8 in 10 of our scenarios). For what that window measures and how it is produced, see the replacement-year method.
Two things could pull it earlier. First, statistical software vendors building agents directly into the packages where this work already happens, so the tool sees the live data rather than a pasted sample. Second, continued hiring restraint: when a team shrinks through attrition, the routine tabulating often moves to software rather than to a new hire, which is the pattern we track in our guide to AI and entry-level jobs.
Two things hold it back. Confidential survey and administrative microdata often cannot leave approved systems, which blocks the easiest tool setups in government and research settings. And errors here are expensive and quiet: a wrong recode can survive into a published figure, so organizations keep a named person accountable for checking the output.
How to stay needed as a statistical assistant
Lean into the tasks the work still keeps with people. Get good at the conversation that defines a request, so you can tell a researcher what their question will and will not support. Take ownership of procedure choice, including when a simpler test is the honest one. And know your collection process end to end, because the person who understands how a number was gathered is the one who can defend it.
Two skills compound that. Scripting your own analysis in R or Python turns you from someone who runs steps into someone who builds and audits a repeatable pipeline. Clear written explanation of methods and limits is the second: reports that state what was excluded and why are hard for anyone to hand off blindly.
If you want to look sideways, the closest work in this family is data entry keyers and general office clerks, and the obvious step up is statisticians, where study design carries more of the role. You can also see how this job sits beside the rest of other office and administrative support work and across the administrative support sector. To weigh a move directly, put two roles side by side on our compare page, or scan where data support work lands on the most exposed jobs list.