Written and maintained by CASRAI Editorial Board
Published
On this page
A data analysis plan explains how a study’s data will be organised, analysed and presented to answer its research objectives. Write it during study planning, before data collection begins. For a proposal or protocol, the central writing task is to connect each objective to defined data, an analysis and a displayed result.
This guide develops that connection through an annotated example and a source crosswalk. It concentrates on drafting the analysis section, with quantitative examples; the clinical-trial statistical analysis plan template covers the more specialised technical document.
Purpose of the section
The section should let a reader follow what you will do with the data and why that work answers the question. The NIH NCATS Toolkit defines the plan around organisation, analysis and presentation, including cleaning, transformations and analytical assumptions. Its scope therefore extends beyond naming a statistical test.
Begin with the study’s actual purpose. The University of Colorado Anschutz protocol-writing workshop distinguishes describing a phenomenon from testing comparisons. That distinction gives the prose a useful starting point: are you estimating or summarising something, comparing groups, or examining a relationship? Write the objective before listing techniques.
Keep the analysis paragraph consistent with the surrounding proposal. The population, outcome and observation period should be recognisable from the methods section. A reader should not have to decide whether two differently worded outcome names refer to the same measurement. Treat that consistency as an editing task: repeat the agreed label where precision matters.
Use the section as the bridge between the study design and the eventual report. When reviewing a draft, imagine handing it to the analyst with the accompanying methods. Mark every place where that person would have to supply an unstated decision. Those marks identify what the next revision needs to explain.
Structure
The following order is an editorial synthesis of the sources below, rather than a universal institutional template. Follow an application or protocol’s prescribed headings when they exist, and use these steps to organise the material within them.
- State the objective and the analysis population. Name the question this paragraph answers and the observations included. Watson’s observational-study template separates study eligibility from analysis populations and asks writers to keep both explicit. A useful drafting frame is: “For objective [label], we will analyse [observations] from [defined population].” Replace every bracket with a decision already made in the design.
- Define the variables and their derivations. Identify outcomes, predictors and relevant adjustment variables. Cairns and Hinterland’s protocol guide asks for variable roles linked to the objectives. Watson adds measurement timing and calculations for derived variables. If the analysis uses a score, specify which score and how it is constructed. Use a research data dictionary for the detailed variable definitions, while retaining the definitions essential to understanding the paragraph.
- Describe preparation and incomplete observations. NCATS includes cleaning and transformations within the plan. Explain how the proposed analysis connects to that preparation; the separate data cleaning steps guide develops the operational work. APH’s missing-data guidance asks for reasons, handling and missing-data analyses by instrument. “Missing data will be handled appropriately” leaves the main decision unwritten. State the proposed approach and its rationale without substituting a software default for that decision.
- Match the analysis to the objective. Name the summaries, model or other method, then state which question it answers. Include relevant assumptions and the planned response to violations. Watson’s template also covers sensitivity analyses and inference conventions where applicable. Avoid a detached inventory such as “correlation, regression and ANOVA will be used”: attach each method to an outcome and comparison. Remove techniques that have no role in a stated objective.
- Specify the output. NCATS includes how results will be presented; APH recommends designing empty article tables before analysis. Sketch a table or figure shell and work backwards to the quantities it requires. For a percentage, for example, the draft should tell the reader what is counted and what forms its denominator. Write a sentence naming the output, rather than merely promising that results will appear in tables.
- Connect the paragraph to the execution record. APH’s documentation guidance calls for records of questions, input files and executed analyses, with annotated code. Explain how the analytical work will be documented. Keep the proposed plan distinct from the later record of what happened. A filename alone does not explain the analysis; a prose plan alone does not preserve the commands that produced the result.
Keep the sample-size rationale aligned with these decisions. Colorado’s workshop treats methodology and sample-size justification as linked concerns in the protocol. Watson’s template recognises that a fixed existing dataset can have a different size rationale from a prospective power calculation. State the project’s actual basis and cross-reference the proposal’s sample-size justification, rather than inserting a calculation that the study did not use.
The data table below maps writing tasks to specific primary guidance. It is a source crosswalk for checking coverage, rather than a set of mandatory headings.
A short annotated example
Illustrative composite: fictional study and proposed wording. A research team plans a descriptive survey of how laboratory staff record equipment checks. Its objective is to describe record-keeping practices among respondents. The choices below belong to this invented design; they are not findings, universal analysis rules or a template for testing an intervention.
“For objective 1, we will describe the equipment-check recording practices reported by eligible laboratory staff who submit the questionnaire. Each submitted questionnaire will form one record. Eligibility will follow the staff-role criteria in the methods section. The recording-method item will be analysed in its original categories. For each category, we will report the count and percentage of respondents providing a valid answer to that item; the table will identify that denominator and show missing responses separately. We will not impute answers for this descriptive item summary. A blank response will remain missing and will not be recoded as ‘no recording system’. Table 1 will display recording-method categories, counts, percentages and the item denominator. The analysis script will preserve the category definitions and output commands. The report will describe respondent practices without claiming that the responses demonstrate an effect on equipment performance.”
Objective and boundary. The first sentence names the question, population and descriptive purpose. The final sentence keeps the proposed interpretation within that purpose. Read those sentences together: if one promises description and the other promises effectiveness, revise the disagreement before polishing the methods language.
Unit and variable. “Each submitted questionnaire” tells the reader what one record means in this fictional dataset. “Original categories” identifies the proposed treatment of the item. If a later draft combines categories or constructs a score, that choice needs its own explanation; it cannot be inferred from this version.
Denominator and missing answers. The paragraph separates an unanswered item from a substantive response. The no-imputation statement is tied specifically to a descriptive item summary. APH’s missing-data guidance distinguishes presenting observed descriptive data from other analytical situations; this example should not be copied into a comparative model without reconsidering the missing-data strategy.
Output and record. The proposed table has a defined job, while the script has a different one: preserving the implemented work. A collaborator can review the promised table before any values exist. The writer can then check that the analysis paragraph supplies each element that the shell will display.
To adapt the writing pattern, replace the objective, population, unit, variable and output together. Changing only “laboratory staff” to a new population would leave the rest of the analytical commitments untouched. Start a fresh paragraph when a second objective requires different data or a different method.
Common mistakes
Using a software name as the method. “We will analyse the data in R” tells the reader where commands may run. It leaves the objective, variables and analytical procedure unstated. Revise the sentence by naming the work first, then include software information where it helps make that work reproducible.
Letting labels drift. An objective about a baseline measure cannot be matched confidently to an analysis paragraph naming only “the final score”. Watson emphasises agreement among hypotheses, measures and analyses. Read the draft across sections, checking both the measure’s name and its time point, rather than checking each paragraph in isolation.
Hiding incomplete-data decisions. APH distinguishes the protocol’s prevention measures, the plan’s handling strategy, the executed syntax and the decision log. A promise to keep records does not replace the handling strategy. Give each document its appropriate task, with clear cross-references where necessary.
Presenting every later analysis as planned. APH permits adaptation in exploratory work but calls for its exploratory character and deviations to be documented. Write the intended status of the analysis clearly. When the work changes, preserve the reason rather than silently rewriting the original commitment as though it had always been there.
Copying more detail than the design needs. Use the clinical SAP route for its specialist purpose. For this section, every technical sentence should earn its place by explaining a decision in the actual study. An impressive method with no attached objective makes the draft harder to assess.
Checklist
Review the finished section in both directions: from each objective to its proposed result, then from each promised result back to the data and method that produce it. The following is a writing check, not a substitute for methodological review.
- Every objective has a corresponding analysis paragraph.
- Population, variable names and observation times agree with the methods.
- Derived quantities have understandable definitions or precise cross-references.
- Each analytical method has a stated purpose.
- Preparation, missing observations and assumptions have explicit treatment where relevant.
- Planned tables or figures display quantities defined in the prose.
- The sample-size explanation matches the study’s actual rationale.
- The execution record can be connected to the proposed work.
Finish by reading the section without its surrounding notes. Replace unresolved placeholders and ambiguous phrases with the decisions agreed by the team. Cairns and Hinterland recommends statistical consultation; use that discussion to resolve analytical choices, then revise the prose so that the written section accurately records them.
Sources
- NIH NCATS Toolkit: Data analysis plan. Checked 2026-11-15.
- Amsterdam Public Health Quality Handbook: Analysis plan. Checked 2026-11-15.
- Cairns and Hinterland Hospital and Health Service: Guide to writing a Research Protocol, sections 12.1-12.2. Checked 2026-11-15.
- University of Colorado Anschutz: Protocol Development Workshop, Session 5 – Data Analysis Plan. Checked 2026-11-15.
- Hunna J. Watson: A Statistical Analysis Plan Template for Observational Studies: Promoting Quality and Rigor in Research, sections 4.3-4.6. Checked 2026-11-15.
- Amsterdam Public Health Quality Handbook: Data analysis documentation. Checked 2026-11-15.
- Amsterdam Public Health Quality Handbook: Handling missing data. Checked 2026-11-15.
Data: Data analysis plan writing crosswalk
An original editorial crosswalk of primary guidance to the written elements of a proposal or protocol analysis section. Each row names its source and section; the wording is a synthesis, not an official universal template.
| Writing task | What to put in the section | Primary guidance |
|---|---|---|
| Define the plan scope | Data to collect, analysis and assumptions, cleaning, transformations, and presentation. | NCATS Toolkit: Data analysis plan |
| Connect methods to objectives | Identify independent, dependent and relevant confounding variables; describe methods against the objectives. | Cairns and Hinterland: Research Protocol, section 12.2 |
| Define derived variables | Describe raw and calculated variables, calculation methods and assessment timing. | Watson: Observational SAP template, section 4.4 |
| Write the missing-data commitment | Describe anticipated reasons for missingness, handling methods and analyses by questionnaire or instrument. | APH: Handling missing data, Documentation |
| Plan the displayed result | Design empty article tables before analysis to clarify the analyses needed. | APH: Analysis plan, How To |
| Connect the plan to its execution record | Log research questions, input data files and analyses; annotate code and distinguish variable definitions from analyses. | APH: Data analysis documentation, Documentation and How To |
Sources
- NIH NCATS Toolkit: Data analysis plan (accessed 2026-11-15)
- Amsterdam Public Health Quality Handbook: Analysis plan (accessed 2026-11-15)
- Cairns and Hinterland Hospital and Health Service: Guide to writing a Research Protocol, sections 12.1-12.2 (accessed 2026-11-15)
- Hunna J. Watson: A Statistical Analysis Plan Template for Observational Studies: Promoting Quality and Rigor in Research, sections 4.3-4.6 (accessed 2026-11-15)
- Amsterdam Public Health Quality Handbook: Data analysis documentation (accessed 2026-11-15)
- Amsterdam Public Health Quality Handbook: Handling missing data (accessed 2026-11-15)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







