Research

Plan every analysis decision before you see the data

Lay out each stage of your analysis, give every method choice its own branch, and plan what you will do when the data surprises you.

Start from a template

An analysis-plan template you can clone. Fill in your study, then work through each stage and its decision points before you open the data.

Clone this template

How it works

Describe your study and the model outlines an analysis pipeline. Each stage gets its own branch, and each decision under it keeps your study design in context.

  1. Describe the study. Enter your research question, your design and the variables you collect. Ask for the main stages of an analysis that fits.

  2. Branch on each method choice. Under each stage, ask about the decisions it holds, such as multiple imputation or listwise deletion for missing data, or mixed-effects models or GEE for repeated measures. Each option gets a branch with its assumptions and tradeoffs.

  3. Plan for surprises. Add a branch for each thing that could go wrong, such as residuals that are not normal or dropout that is not random. Write the response you will use under each one.

  4. Share the plan. Publish the map and anyone with the link can read it without an account, so you can send it to coauthors or link it from your pre-registration. It shows the path you chose and the alternatives you weighed.

Why a map instead of a written plan

A written plan reads top to bottom: do this, then that. Real analysis branches, because the right test at step three depends on what you found at step two. A map holds each of those conditions as its own path.

Mapping the choices before you see the data also shows which ones could change your conclusions, so you know which to pre-register.

Example

Say you are planning the analysis for a randomized trial of a mindfulness program for chronic pain. The plan branches into the primary outcome, secondary outcomes and subgroups. Under the primary outcome you compare intention-to-treat with per-protocol analysis. Under intention-to-treat, a branch on participants who never started the program shows you need to pre-specify a complier average causal effect analysis.

Now try it yourself

Plan a data analysis pipeline for a longitudinal study of cognitive decline in aging adults
The pipeline has four stages: data cleaning and missingness assessment, descriptive trajectory characterization, mixed-effects modeling of decline rates, and sensitivity analyses for dropout and practice effects.
What approaches handle informative dropout in this design?
Pattern-mixture models stratify by dropout timing. Shared-parameter models jointly model the outcome and dropout process. Multiple imputation under missing-not-at-random assumptions is a third option, though it requires explicit sensitivity parameters.

Ready to try data analysis planning?