Brown Virtual Lab · Research practice
Give every finding a trail of evidence.
A working bench for sharper questions, inspectable data, thoughtful interpretation, and reproducible scholarship.
Four practical instruments
Work the question. Inspect the reasoning.
Protocol planning instrument
Prepare a structured discussion draft. Each field needs project specific evidence and reviewer agreement before operational use.
A four lesson practice pathway
Learn it. Apply it. Bring evidence to review.
Prerequisite: research orientation and data stewardship guidance. Work through the lessons in order. Allow about two hours plus your applied deliverables. Route feedback through your assigned My Work submission; the support form can help establish an assignment.
Lesson 1From a question to an estimand25 minutes
Learning objective: Define a research question that can be evaluated against a specific analysis.
A broad interest becomes an answerable question when population, exposure or intervention, comparator, outcome, and timing are explicit. Distinguish the unit observed from the unit analyzed: several images from one person create a different dependence structure from one observation per person. An estimand names the quantity the study aims to estimate. An association at a fixed time, a change from baseline, and a risk over follow-up answer different questions. Specify how eligibility, time origin, intercurrent events where relevant, and missing outcomes affect that quantity. Choose the analysis after the target question is stable. Record feasibility assumptions as assumptions until verified.
Applied assignment
Complete all eight planning fields using a fictional or public literature question. Add one alternative interpretation and one condition that would require revising the design.
Review rubric
Award 0–2 points each for population, comparison, outcome/time, estimand, and feasibility/authority. A score of 2 requires a specific, internally consistent statement; 1 indicates partial specification; 0 indicates absent or contradictory content. Revise every zero before mentor review.
Lesson 2A data audit that preserves the evidence30 minutes
Learning objective: Identify quality concerns and create a traceable correction plan.
Start with an unchanged source snapshot and a versioned dictionary. Check record counts, identifiers, expected units, ranges, category codes, timestamps, joins, and missingness. A repeated identifier can indicate an accidental duplicate, a longitudinal measurement, or an intended one-to-many relationship. Resolve its meaning before removing a row. A value outside an expected range is a review flag; it becomes a correction only when the source and applicable rules establish the intended value. Preserve missingness reasons and distinguish unknown, unmeasured, unavailable, and structurally inapplicable values. Compare counts before and after each transformation. A clean format establishes neither validity of measurement nor scientific fitness.
Applied assignment
Run the teaching audit. For every flagged row, write the source question, decision owner, proposed rule, and before/after verification. Keep uncertain records flagged until evidence resolves them.
Review rubric
Award 0–2 points each for finding the five flags, respecting the unit of observation, preserving raw values, recording correction authority, and reconciling counts. Use the audit explanations as the answer key. Full credit requires a reason for every proposed change.
Lesson 3When pooled results change the story25 minutes
Learning objective: Explain a reversal between pooled and stratified comparisons.
An aggregate comparison depends on both the outcome rates within strata and the distribution of observations across strata. The demonstration fixes the low and high baseline risk outcome rates in each group while allowing the proportion in each stratum to vary. Group B has a higher favorable outcome rate within both strata. Group A can still have a higher pooled rate when it contains a much larger share of the low risk stratum. Standardizing both groups to a common stratum distribution makes this weighting visible. This exercise illustrates aggregation and possible confounding. Whether a real variable is a confounder, mediator, or collider depends on the scientific question, timing, and causal structure; automatic adjustment can introduce bias.
Applied assignment
Record the default pooled rates and both stratum comparisons. Set both sliders to 50%. Explain why the ranking changes and identify what information a real observational study would require before causal interpretation.
Review rubric
Award 0–2 points each for within-stratum comparisons, pooled calculation, weighting explanation, equal-mix comparison, and limits of causal inference. Full credit identifies both unequal group composition and the need for a justified adjustment strategy.
Lesson 4Reproduce, reconcile, and hand over35 minutes
Learning objective: Assemble a package that another qualified researcher can inspect and rerun.
Reproducibility requires a connected chain from authorized source data to reported results. Record the extraction logic, dictionary, transformations, analysis specification, code commit, computing environment, and output versions. A random seed helps repeat a stochastic procedure; its value alone does not reproduce an analysis when software or data change. Independent verification should test key counts, denominators, uncertainty estimates, and the correspondence between outputs and manuscript claims. Log discrepancies with their resolution and preserve earlier versions. For AI assisted work, record the permitted inputs, tool/model and date, generated material, human verification, corrections, and required disclosure. Treat synthetic data as teaching or simulation material with an explicit provenance record.
Applied assignment
Review all ten handoff items. Assemble an approved evidence location with a README, analysis plan, dictionary, code/environment details, output index, decision log, and reviewer findings. Submit an approved link through your assigned My Work deliverable.
Review rubric
Award 0–2 points each for lineage, executable instructions, output reconciliation, independent review, and limitations/disclosure. A self-review count documents preparation; the assigned reviewer determines whether the evidence meets the project requirements.
Sources and next steps
The original teaching instruments connect to NIH rigor and reproducibility guidance and EQUATOR reporting guideline resources. Consult the applicable project requirements, current primary methods literature, and qualified reviewers for the actual study.
- NIH: Rigor and reproducibility guidance
- EQUATOR: Search reporting guidelines and extensions
- Continue research fundamentals
- Request assignment or methods support
Content and instrument version: October 22, 2026. Synthetic examples and original educational text: Brown Virtual Lab research practice expansion.