Turing is hiring statisticians and quantitative researchers to build benchmark tasks that test AI models on real statistical work in jamovi or JASP — hypothesis testing, regression, ANOVA, assumption checks, and APA reporting. You construct messy research datasets, run the reference analysis yourself, and write numerical verification criteria precise enough that a grader can mark a model's output right or wrong.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

What the work actually is

You are writing the answer key, not answering questions. A typical unit of work starts with a study-design brief — a psychological assessment battery, a small clinical trial table, an unclean survey export — and a task the model must complete: run the ANCOVA, test the assumptions, report it in APA format. You then execute that analysis end to end in jamovi or JASP, record the execution trace, and specify exactly what a correct response looks like: the F statistic to a stated tolerance, the partial eta-squared, which post-hoc correction was appropriate and why, whether Levene's test result should have redirected the analyst to Welch's correction. The hard part is rarely running the test. It is deciding what counts as correct when a model reaches a defensible answer by a different route, and writing verification criteria that accept the defensible route without accepting the wrong one.

A second stream is review: reading tasks other contributors wrote and checking them for solvability, methodological appropriateness, and completeness. Tasks fail review most often because the dataset does not actually support the test being asked for, because the brief is ambiguous about which variable is the covariate, or because the stated gold answer was computed with a default setting the brief never specified.

What the screen looks for

Tool-specific fluency. Which jamovi or JASP module you use for a given analysis, what its defaults are, and where those defaults differ from SPSS or R. Generic statistics knowledge without tool grounding does not clear this bar.
Assumption reasoning under pressure. Not that you know Shapiro-Wilk exists, but what you do when n=200 and it comes back significant.
Numerical discipline. Can you state a tolerance, justify rounding, and explain why two tools disagree in the third decimal place.
Written precision. Briefs must be unambiguous in English to a reader who cannot ask you a follow-up question.

Logistics

Hourly, remote, fully async, with an initial duration of about five weeks and possible extension. Turing does not publish a band for this listing; rates for comparable domain-expert statistics work on the platform have been observed in the mid-double-digit USD per hour range, but treat that as context and not a quote. Contributors generally set their own hours against weekly throughput expectations; a working install of jamovi or JASP on your own machine is assumed.

Domain Expert - Statistician / Data Analyst – (jamovi / JASP)

What the work actually is

What the screen looks for

Logistics

Interview prep — 10 questions