AI for statisticians and survey researchers
Give Ultramarine Work the protocol, codebook and data files in your study folder, and it drafts the SAP, writes the questionnaire, computes the survey weights and codes the open ends.
Wave-4-weighted-topline.xlsxExample| Margin | Sample | ACS target | Weighted | Check |
|---|---|---|---|---|
| Age 18–34 | 22.1% | 29.4% | 29.4% | On target. |
| Age 65+ | 27.8% | 21.6% | 21.6% | On target. |
| Female | 54.6% | 51.0% | 51.0% | On target. |
| HS or less | 17.9% | 36.8% | 33.9% | Trimming at 5.0 left it 2.9 points short. |
| Design effect | 1.84 | Effective n is 1,093 of 2,012 completes. |
NoteTrimming at 5.0 leaves HS or less 2.9 points under target, so rerun with a 6.0 cap and report the design effect both ways.
Ask it for
“Draft the SAP for protocol VRD-201 from the protocol and last trial’s SAP, with ICH E9(R1) estimands, MMRM primary analysis and empty TFL shells.”
SAP-VRD-201.docx“Write the adult wellness tracker questionnaire with quota screener, 5-point scales, rotated lists and skip logic, plus the variable list for the scripters.”
Tracker-questionnaire.docx“Rake this national web panel to ACS controls for age, sex, race, education and region, then produce the Wave-4 topline with design effect and base sizes.”
Wave-4-weighted-topline.xlsx“Code the open ends to last wave’s code frame, keep an uncodable category in the frame, and report agreement against the 300 verbatims we hand-double-coded.”
Open-end-coding.xlsx“Build the data dictionary and codebook from the de-identified extract, state the grain, keys, units, value codes and null meanings, and mark any mismatch.”
Codebook-draft.xlsx“Review the statistics in the draft manuscript for test choice, model form, missing-data handling and multiplicity, and flag each issue by page number.”
Manuscript-stat-review.pdf
Inside the Statistics & Survey Research mode
Statistical analysis plan
The SAP it drafts sets estimands with all five attributes, specifies primary and sensitivity analyses, and leaves empty TFL shells, numbered and footnoted.
Statistical modeling and review
Writes up the regression or survival analysis with diagnostics and effect sizes, and reviews a colleague’s paper for the test that does not match the design.
Causal and econometric analysis
Builds the DiD or synthetic control study with pre-trend plots, a donor pool, an event-study table and the robustness checks the referee will request.
Design of experiments
Lays out the run sheet for a fractional factorial or response-surface design, with aliasing, center points, randomized run order and the analysis approach.
Survey design and analysis
Drafts questionnaires with quota screeners, rotated lists and skip logic, and turns fielded data into toplines and crosstabs with base sizes and tests.
Survey sampling and weighting
Constructs base weights, nonresponse adjustment and raking to control totals, then shows the design effect, the weighting loss and where the weights moved.
Qualitative research synthesis
Codes verbatims to a fixed frame that keeps the uncodable category, and reports agreement against hand-coded samples before anything is called a theme.
Literature and evidence review
Records the reproducible search and PRISMA flow, appraises each study and grades the certainty of the evidence.
Data documentation and quality
Documents grain, keys, units, value codes and what each null means in the data dictionary, with rule-based checks your users can rerun.
Pra information collection request
Assembles the OMB clearance package with Part A and B supporting statements, the burden-hour table and the 30-day and 60-day Federal Register notices.
Proposals and RFP responses
Puts the response in the solicitation’s own order, with a compliance matrix, past-performance write-ups, staffing tables and the technical narrative.
- Office skills
- Excel, Word, PowerPoint, PDF
- Tools
- Read, edit and write files, Search files, Run code, Search and read the web, View images
- Connectors
- SharePoint, Google Drive, Gmail, and additional custom connectors
- House style
- Your fonts, layouts and rules, on every file
How it works
- 1
Open the study folder
The trial folder holds the protocol, prior SAPs, annotated CRFs and de-identified datasets. The tracker folder holds the questionnaire, wave codebooks, weighting scripts and last quarter’s toplines.
- 2
Ask for the piece of work
You ask for the SAP, the raked weights, the coded open ends or the manuscript review. What comes back cites the source file for every figure and lists what it could not reconcile.
- 3
Review, then sign
Rerun the weights in your own software, hand-code a subsample of verbatims to compare, and check each estimand against the protocol before anything circulates. You sign the SAP, and you decide what the report says.
Who it’s for
Statisticians, biostatisticians, survey researchers and methodologists, clinical data managers and social science research assistants at statistical agencies, survey and research firms, universities and pharma.
Questions
Can it compute the weights and crosstabs, or does it just write text?
Yes, it computes them from the data files in the folder. Ultramarine Work runs code, so the raking, crosstabs and design effects are derived from your data rather than described from general knowledge. That does not settle the accuracy question, because fluent output can still be subtly wrong. Each figure names the file it came from, anything it cannot reconcile is flagged rather than guessed, and you rerun the headline numbers before they go into the deliverable.
Can I use it with Title 13, CIPSEA or DUA data?
No, not unless the data are de-identified under the applicable standard and the agreement allows it. Title 13, CIPSEA and DUA microdata, and participant-level clinical data, stay in the approved environment. The model runs on Paper Instruments’ servers, so the files it works on are sent there, which rules out those datasets. Work from aggregate tables or de-identified extracts. When the agreement is silent, the disclosure review board or DUA administrator decides.
Does the SAP it drafts follow ICH E9(R1)?
Yes, the analysis-plan skill is written around estimands with all five attributes. For each objective it states the treatment, population, variable, intercurrent events and population-level summary, with missing-data handling, multiplicity and sensitivity analyses laid out next to them. It drafts from your protocol and prior SAPs, but you read every choice before the SAP circulates.
Can we have it code open-ended responses for a client?
Yes, and the validation stays with you. Ask it to code against a fixed frame that includes an uncodable category, then double-code a subsample by hand and report the agreement figure, benchmarked against how well your human coders agree with each other. AAPOR’s best practices expect you to name the model, the tasks AI performed and the validation process. The verbatims are participant data, so consent language and the client contract decide where they can go.
Where does our data go when the app works on it?
Ultramarine Work changes files only in the project’s folder. When a request needs it, it can read files elsewhere on your computer, never passwords, keys or app data, and you can turn that off in settings. The model runs on Paper Instruments’ servers, so the content it works on is sent there to be processed. Paper Instruments is SOC 2 Type II, and our security and compliance details are on our compliance page.