Baseline-adjusted domain contrasts

Explore the executed results

Selection switches exact precomputed results. It changes the figure, denominators, table, summary and current-selection downloads together. No analysis runs in this browser.

Python · All participants · 5 computed rows. Review population counts and exact values in the table below.

Image unavailable. Use the numerical table and downloads below.

CSV · all three strata (overlapping; do not sum)

Values behind this selection

The table scrolls within this panel. All computed rows are included in the current-selection CSV and JSON.

Display rounded to three decimals. Downloaded CSV/JSON retains analysis precision. “Unavailable” is not zero.

Why this figure, and what it estimates

An explicit treatment contrast answers a different question from two separate arm means. Aligned pointwise intervals show model-based precision without relying on overlap of within-arm intervals.

Analysis contract

For each domain and All/F/M stratum fit Week12 = intercept + Investigational indicator + centered baseline using complete pairs. The treatment coefficient is Investigational minus Reference. Classical OLS standard errors and pointwise 95% t intervals use residual df n minus 3. Require at least two complete cases per arm and a full-rank design with condition number at most 1e6.

Assigned fictional arms; baseline = scheduled Week 0. No visit windows or treatment switching. Subjects are independent; domain scores are invented 0–100 points. Missing outcomes stay missing; missing scheduled rows are rejected.

Full input, units, missingness, interval, rounding and tolerance specification

Do not overinterpret

Complete-case selection is not a missing-data correction. Common baseline slope, linearity, independent observations and homoscedastic approximately normal errors underlie the reported intervals. Five domains and overlapping strata have no multiplicity adjustment. These synthetic nonrandomized teaching patterns do not support a causal efficacy claim.

Alternative views

Prespecify the estimand and missing-data strategy for a real trial. Consider justified robust covariance, treatment-by-baseline interaction, longitudinal models or multiple-imputation sensitivity analyses, each with separate assumptions and QC.

Read, reproduce, then adapt

Four-step review guide · Compare complementary figures

Compare the All-participants Fatigue coefficient with the two within-arm mean changes. Explain why subtracting separate interval endpoints is not the ANCOVA interval. Inspect complete cases by arm and identify which randomized-population questions remain unanswered.

1. Open the selected CSV and reconcile counts before comparing outcomes.
2. Run the minimal example locally, then inspect the native Figure or ggplot object.
3. For your own authorized data, explicitly map subject, arm, scheduled time and units. Keep expected missing visits as rows, document exclusions, and revise the contract for your trial design.
4. Run invalid-input and known-answer checks before comparing R/Python outputs.

Cross-domain adaptation notes and clinical coverage matrix

Executed source and separate QC layers

Original author: Jaime Yan. Personal noncommercial use only. License · Required citation · Upstream attribution

Python reusable implementation · R reusable implementation · Complete reproducible source bundle

Extract the complete bundle and run these commands from its root. Recorded environments: Python · R.

python examples/adjusted_domains.py All
Rscript --vanilla examples/adjusted_domains.R All
python scripts/render_adjusted_domains.py --rscript Rscript

Numerical validation

15 independently fitted domain/stratum contrasts; max absolute difference 6.821e-13. Tolerance 1e-9 absolute/relative; counts exact.

Data QC

Keys, schedules, units, finite ranges, empty and sparse populations tested.

Export QC

CSV/JSON selection, parsed SVG, PNG dimensions, readable PDF text and Python text bounds checked.

Package-risk assessment

Not run. Numerical agreement is not package-risk or regulatory validation.

Browser evidence

Recorded separately in the local delivery report; no accessibility certification claimed.

Machine-readable numerical QC · Source, input and artifact hashes

Input: the library's original synthetic generator, seed 20260921, 36 fictional participants. Separate from CDISC Pilot. The deterministic missingness mechanism is described in the contract. No patient or employer records.