Lead with the question
Write a specific title and one-sentence summary. Use the description for your data, decisions and assumptions. Markdown is supported.
FOR EXAMPLEEstimating treatment effects with a causal forest
StatsOtter / Field guide
Good research.
Made readable.
Find an approach, understand the decisions, and share a process others can follow. Start wherever you are.
A research question↓
Explore→Understand→Contribute
↓A process worth sharingRead here. Run the analysis in your own environment.
01 / ExploreREAD → UNDERSTAND
Read the decisions.
Then read the code.
Every workflow connects a research question to a sequence of analytical choices. Here is a useful way to read one.
- 1
Find the right starting point
Search the library by topic, author or tag. Use Fields to narrow the subject, then compare Popular and Newest.
- 2
Check the question and inputs
Read the overview. Identify the target effect, data format and assumptions. Follow the source when one is provided.
- 3
Inspect one step at a time
Select a step in the workflow to see its explanation, formula and code. Compare the reported output with the original question.
Anatomy of a workflow
A difference-in-differences example
INPUT → STRUCTURE
One row per unit and period.
Check the outcome, unit identifier, time variable and treatment timing. The preparation note should explain exclusions and how the comparison group is defined.
unit_idperiodoutcomefirst_treated
QUESTION → ESTIMATE
Make the target effect explicit.
A group-time ATT describes the average treatment effect for a treatment cohort at a particular time. Read the estimator, the comparison group and the assumptions together.
ATT(g, t) = E[Yₜ(1) − Yₜ(0) | G = g]ESTIMATE → DIAGNOSTICS
Ask how much the answer depends on assumptions.
Look for pre-treatment evidence and sensitivity analysis. A workflow should explain what these checks can reveal and which identification concerns remain.
OUTPUT → INTERPRETATION
Report the uncertainty alongside the effect.
An event-study plot can summarize estimates across relative time. The write-up should identify the reference period, uncertainty intervals and the interpretation of the plotted effects.
An illustrative reading sequence, not an executed analysis or a claim that these checks establish identification.
02 / ContributeEXPLAIN → PUBLISH
Make your process
someone else's starting point.
Choose + Post after signing in. The writing desk keeps the overview, topics and ordered steps together, with a live outline alongside.
01 Overview→02 Topics→03 Steps→04 Review
Add useful topic tags
Type a topic and press Enter, or pick a community suggestion. Combine the research area, method and tool. Use up to 12 tags.
Build a legible sequence
Name each operation, choose its stage and explain why it matters. Add code, a LaTeX formula or a documentation link where useful.
ONE STEP, ONE JOBCheck overlap before estimating the effect.
Read it as a newcomer
Inspect the preview and writing checklist. Publish when the explanation is clear. The form publishes immediately; it does not save a private draft.
AFTER PUBLISHINGUse Edit on your workflow to refine it later.
Stage quick reference
- Prep
- Clean data, define the sample.
- Diagnostic
- Check balance or pre-trends.
- Estimation
- Fit the core estimator.
- Inference
- Compute uncertainty.
- Robustness
- Test sensitivity of conclusions.
- Heterogeneity
- Compare effects across groups.
- Reporting
- Explain estimates and figures.
- Other
- Label a different operation.
Writing about a single estimator without a pipeline? The standalone method form is also available. The main library focuses on workflows.
03 / DocumentsSOURCE → WORKFLOW
Start from the work
you already have.
Use a structured Markdown document for a fuller record, or use AI import to turn an existing tutorial into a draft you can review.
A document you control
Upload Markdown
The upload page includes the supported template. It covers the overview and steps, plus input information, figures, results and AI Notes.
- Copy the template from the upload page.
- Fill it with your explanation and code.
- Upload the file or paste its contents.
A source to build on
Import a tutorial with AI
When AI import is configured, provide a tutorial URL. Review the generated draft, ask for refinements, and publish after checking it.
- Provide a source URL and your direction.
- Inspect the draft and source material.
- Refine the content before publishing.
After a workflow exists
The document stays with the card.
Open Doc on a workflow to read its Markdown source. Owners can edit it; the document view also provides downloads and version history. Review changes before replacing a document.
Workflow card↕Knowledge document↓Version history
Example input tableDownload a sample CSV to help describe your data format.
CSV
04 / CommunitySAVE → DISCUSS → RETURN
Keep good methods close.
Make the discussion useful.
Sign in to build a collection and compare notes with other researchers. Your contributions and saved workflows come together in your profile.
Star what helped
A star is a vote for a workflow or comment. Select it again to remove your vote.
Recognition for useful contributionsSave it for later
The separate Save button adds a workflow to your own saved collection in your profile.
Your reading list, ready to revisitAsk a specific question
Comment in Markdown. Explain the assumption, step or output you are asking about. Replies allow one level of follow-up.
Context makes a question answerableFollow the body of work
Open a contributor's profile to browse their workflows and learn how their methods connect.
People, not just isolated cardsA useful discussion starter
“Which comparison group does this step use, and how would that choice change the interpretation?”
Point to the decision. Share your context. Keep the claim precise.