Report synthetic data generation in your manuscript by clearly labeling it as synthetic, documenting the exact generation process, and including a dedicated section that allows reviewers to assess its validity and reproducibility. Transparency is the core principle — reviewers must be able to distinguish synthetic from real data and evaluate whether the synthetic data is fit for your stated purpose.
Start with a clear disclosure up front
In the Abstract and Methods, explicitly state that the study uses synthetic data. Do not bury this detail. Use unambiguous phrasing like "we generated synthetic data using [method]" rather than euphemisms such as "simulated" or "augmented" unless those terms are technically accurate. This prevents accidental misrepresentation and helps readers interpret your findings correctly.
Document the generation method in detail
In the Methods section, include:
- The exact algorithm or model used (e.g., GANs, VAEs, agent-based simulation, differential privacy mechanisms, or rule-based generators), including version numbers and key hyperparameters.
- The training or input data — what real data, if any, informed the generator, and how it was preprocessed.
- The sample size and how it was determined.
- Any assumptions or constraints baked into the generation process (e.g., independence assumptions, distribution choices).
- The software environment — language, libraries, and random seed(s) for reproducibility.
If you used an existing framework or package, cite it. If you wrote custom code, consider depositing it in a public repository and referencing it in the manuscript.
Justify why synthetic data was appropriate
Explain the rationale — for example, privacy constraints, rare-event coverage, cost, or the need to test a methodology under controlled conditions. Reviewers will want to know that synthetic data was not a shortcut but a deliberate, justified choice.
Assess and report fidelity and validity
Include a subsection on validation:
- Compare synthetic data to real data (if available) using relevant statistical metrics (e.g., distributional overlap, correlation structure, utility metrics).
- Report any known limitations, such as mode collapse or over-smoothing.
- State clearly what the synthetic data is not suitable for, if applicable.
Use a dedicated "Data Availability" statement
Under Data Availability, say: "The dataset used in this study is synthetic and was generated as described in [Methods section]. The generation code is available at [repository link]." This makes your compliance with journal policies explicit and helps readers locate the resources.
Follow journal-specific guidelines
Some journals have explicit policies on synthetic data reporting — check author instructions before submission. If none exist, err on the side of more detail. When organizing your references and ensuring every methodological claim is properly cited, TrueCite can automatically verify that your citations match the sources you actually used, reducing the risk of citation errors in a methods-heavy manuscript.
