Statistician

$60–120/hrrate as listed

micro1 is contracting statisticians to build and review training data for AI systems — cleaning messy datasets, running descriptive and inferential analyses, and documenting methods in plain language. Work is remote and asynchronous, with observed rates in the $60–120/hr range depending on credentials and project.

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

What the work involves

You receive real-world datasets that are incomplete, inconsistent, or badly structured, and you produce the artifacts a model needs to learn from: a defensible cleaning pipeline, a documented analysis, a visualization that actually answers the question, and a written summary a non-statistician could follow. Some tasks are generative — you write the reference solution the model is trained toward. Others are evaluative — you compare two model outputs and judge which handled the missingness, the test choice, or the interpretation more responsibly, then explain the reasoning in writing.

The recurring theme is statistical judgment under ambiguity. A model will happily run a t-test on non-independent observations, report a p-value without an effect size, or drop rows without noting the mechanism. Your value is catching that and articulating why it matters in a way that generalizes beyond the single example.

What the platform screens for

Depth in your stated tools — expect follow-ups on how you actually handled a specific dirty dataset in R, Python, SAS, or Stata, not a list of packages.
Fluency in core inference: hypothesis testing assumptions, regression diagnostics, missing data mechanisms, multiple comparisons.
Written clarity. Much of the deliverable is writing, and the screen weighs whether you can explain a method precisely without jargon.
Calibration — whether you distinguish a genuine statistical error from a defensible alternative choice.

Logistics

Contractor engagement, fully remote, no fixed hours.
Asynchronous collaboration with project stakeholders via written threads; expect to clarify ambiguous specs rather than guess.
Volume varies by customer project; many contributors work 10–20 hours a week alongside other commitments. Pay bands are as observed and not guaranteed.
No prior AI or ML experience required — the credential and the statistical reasoning are the qualification.

What they look for

Data cleaning and preparation (dirty data)
Basic descriptive and inferential statistics
Programming in Python or R for analysis
Data visualization
Communicating findings to a non-technical audience

Before you apply

Would you pass this screen?

Drawn from what this listing actually requires. Answered in your browser — nothing is sent to us, and we do not store your answers.

Do you hold a Master's degree or PhD in Statistics, Data Science, Mathematics, Biostatistics, or a closely related quantitative field?
Advanced degree in quantitative field · required
Do you have hands-on experience writing and debugging code in Python or R for statistical analysis and data manipulation?
Proficiency in R or Python · required
Have you worked on a real-world project involving the cleaning and preprocessing of large, unstructured, or noisy datasets?
Experience with messy datasets · preferred
Are you able to work asynchronously in a remote environment with written communication as the primary mode of collaboration?
Remote work setup · preferred

0 of 4 answered

Included with this listing

Interview prep — 11 questions

Matched to what this platform's screening actually measures. Answer each out loud in 60–90 seconds: claim → concrete example → implication.

What it tests: Whether the candidate has hands-on experience diagnosing and solving real data quality problems, not just running canned cleaning routines. Can they articulate the reasoning behind preprocessing choices, not just the technical steps.
A strong answer: The answer should name a specific dataset, identify concrete messiness (missing values patterns, outliers, formatting inconsistencies, or domain-specific noise), and explain the statistical or domain reasoning for each decision—not just "I removed nulls." It should show awareness that preprocessing choices affect downstream analysis.
Where people lose it: Candidates say "I cleaned the data" without naming what was actually wrong or why their solution was appropriate. They describe the tool ("I used pandas") instead of the problem they solved. They don't connect preprocessing choices to the analysis goal.