Student preparation guide

What your experimental design needs to cover

Everything below is required unless it is marked Optional or Conditional. The protocol checker will normally prevent final dataset generation if required information is missing. Conditional information is required only when your design uses that feature.

1. Study overview

• A short study title.
• The motivation for the study and why it matters.
• The context and setting of the study.

2. Population and sample

• The target population.
• The eligibility criteria, including any exclusions.
• The accessible population and sampling frame: who you can actually reach, and the list, register, or other source participants will be drawn from.
• The recruitment and consent process, including how eligible people will be invited and enrolled, and any incentives or reminders.
• The intended sample size (total number of randomized units).
• A justification for the sample size, such as statistical power, precision, practical constraints, or available resources.

3. Units and dataset structure

• The unit of assignment and intervention: who or what is randomized to a condition and receives the intervention.
• The unit of observation: who or what supplies the measurements. This may differ from the unit of assignment.
• The intended unit of analysis.
• What one row of the final dataset will represent.
• The observation schedule, including when measurements will be taken.
• Any grouping, clustering, nesting, or repeated measurements. This is required if the unit of assignment differs from the unit of analysis. For cluster randomization, give the number, composition, and size of the clusters.

4. Conditions and assignment

• A description of every experimental condition.
• A comparison or control condition where the design requires one.
• The content of each condition, including its intensity or dosage where relevant.
• The timing and duration of each condition.
• Delivery details only when the way the condition is delivered could materially affect participants’ exposure to it.
• The randomization method.
• The proportion of units allocated to each condition.

Provide the following only when applicable:

• Stratification variables.
• Concealment or blinding.
• Factor levels for a factorial experiment.
• Possible spillovers or interference between units.

5. Variable inventory

List every variable or column expected in the final dataset. This may include:

• Identifiers.
• Treatment assignment.
• Outcomes.
• Baseline covariates.
• Manipulation checks.
• Moderators.
• Mediators.
• Time variables.
• Cluster or group identifiers.
• Compliance measures.
• Post-treatment covariates.
• Weights.

For each variable, provide:

• A unique variable name.
• Its role in the experiment.
• A plain-language description.
• Its data type.
• Its units, scale, categories, or allowable values.
• Its measurement instrument or data source.
• The measurement unit and dataset level.
• When it is measured relative to the treatment, and how often.
• Optional: any expected reasons for missing or unusual values, when these differ from the study-wide missing-data process in Section 8.
• For an ordinal variable, list the categories in order from lowest to highest.

If a variable is calculated from other variables, also provide its derivation rule.

Mark a variable as a moderator if you expect it to change the size or direction of the treatment effect.

6. Outcomes and hypotheses

For each outcome, provide:

• The corresponding variable from the variable inventory.
• Whether it is a primary or secondary outcome.
• The measurement window.
• The experimental comparison being made.
• Any aggregation rule.
• Whether it is measured once, repeatedly, or as a time-to-event outcome.

For a count outcome, also provide the exposure period or denominator over which events are counted.

For each primary outcome, provide a hypothesis that states:

• The outcome to which it applies.
• The expected relationship between the conditions.
• The predicted direction: whether the treatment condition is expected to produce a higher value or a lower value, or whether the direction is uncertain because the hypothesis is exploratory.
• A brief plain-language explanation or rationale.

You do not need to propose a numerical effect size.

7. Analysis plan

For each hypothesis, describe the main analysis that will test it, including:

• The conditions being compared.
• What will be estimated, such as a difference in means or a regression coefficient.
• Any covariates to be included.
• How the result will be judged.

Across all analyses, provide:

• Any additional analyses, such as subgroup, robustness, or exploratory analyses. State “None planned” if there are none.
• How missing outcome or baseline values will be handled, and how you will check whether that choice matters.
• How standard errors, confidence intervals, or tests will be calculated, and how they reflect the randomization design and any grouping or clustering.
• Optional: How you will interpret or adjust for multiple outcomes or comparisons.

The analysis plan does not change the simulated data, but it forms part of the approved protocol. Consider it to be a kind of “pre-registration.”

8. Data-generating and realism assumptions

• Expected relationships among measured variables.
• The likely missing-data or attrition process.
• Plausible measurement limitations, boundaries, or unusual observations.

Optional:

• Causal diagram code, using DAGitty-style syntax or a simple edge list. A causal diagram can help the generator preserve the intended dependencies between variables. Every measured variable included in the diagram should also appear in the variable inventory.
• Moderators or expected treatment-effect differences. Include these when you expect the intervention to work differently for particular measured groups or characteristics.
• Non-compliance or differences between assigned and received treatment.

Conditional:

• Time patterns and dependencies for repeated, clustered, longitudinal, crossover, or staged experimental designs.

Before approving the protocol

Check that:

• The units and timing agree across all sections.
• Every outcome appears in the variable inventory.
• Every primary outcome has a linked hypothesis.
• The allocation proportions agree with the listed conditions.
• Variables used in the causal diagram also appear in the variable inventory.
• The protocol contains no names, email addresses, participant records, or other identifiable or sensitive personal information.
• Every hypothesis has a planned analysis, and that analysis reflects the unit of assignment and any grouping.
• The protocol check has been run, every blocking question corrected, and a response recorded for every remaining warning.

Once you approve, the protocol is locked and you will consume your dataset generation key. Your responses to warnings are kept in the approval record.