Which Reporting Guideline Does Your Study Need?

5 min read

The short answer: the guideline you need depends on two things, what kind of study you ran and what stage you're writing up. Most studies need one main design guideline, plus a method-specific one if the analysis is Bayesian or simulation-based. They are not competing standards. They stack.

Here are seven that come up constantly in our work, what each one is for, and how they fit together.

The design guidelines: what kind of study is this?

CONSORT 2025 (randomised trials, the results paper). The update to CONSORT 2010 is a 30-item checklist plus the familiar participant flow diagram. It was published simultaneously in The BMJ, JAMA, The Lancet, Nature Medicine and PLOS Medicine in April 2025. The biggest change is a new open-science section: where the protocol and statistical analysis plan can be found, and where de-identified data and statistical code can be accessed. It also asks you to report changes made after the trial started, including any outcomes or analyses that weren't prespecified.
SPIRIT 2025 (randomised trials, the protocol). CONSORT's companion for the document you write before the trial starts. It has 34 minimum items and a schedule diagram of enrolment, interventions and assessments. New items cover patient and public involvement and trial monitoring. The two were updated in parallel by overlapping author teams, so a SPIRIT protocol now maps cleanly onto the CONSORT report it will eventually become.
APA JARS-Quant (quantitative research in psychology and the social sciences). The American Psychological Association's Journal Article Reporting Standards, updated in 2018. Instead of one design, it gives a core set of items for any quantitative study, plus additional modules for specific designs: experiments, nonexperimental studies, longitudinal designs, replications and more. If you're in psychology, education or the behavioural sciences, this is usually the reviewer's mental checklist whether or not they name it.
TRIPOD+AI (prediction models, 2024). For studies that develop or validate a model that predicts an individual's diagnosis or outcome, whether it's built with logistic regression or a neural network. It has 27 items plus a 13-item checklist for abstracts, and it replaces TRIPOD 2015, which was written mostly with regression in mind. It's the closest thing to a published standard for reporting a clinical machine-learning model.

The analysis guidelines: how did you get the numbers?

SAMPL (basic statistical reporting). Tom Lang and Doug Altman's 2013 guidelines for reporting statistics in biomedical articles. SAMPL doesn't care about your design. It covers how you describe the statistical methods and results themselves: which test or model you used and why, whether its assumptions held, and estimates with confidence intervals rather than bare p-values. Think of it as the floor under every other guideline.
BARG (Bayesian analyses). John Kruschke's 2021 Bayesian Analysis Reporting Guidelines, in Nature Human Behaviour. A Bayesian result can't be checked without things a frequentist paper never needs: the priors and why you chose them, how sensitive the conclusions are to those choices, whether the sampler converged, and how the posterior was summarised. BARG walks through each step.
ADEMP (simulation studies). From Tim Morris, Ian White and Michael Crowther's 2019 tutorial in Statistics in Medicine. Strictly a planning framework rather than a checklist: state your Aims, Data-generating mechanisms, Estimands, Methods and Performance measures. It also asks you to report Monte Carlo standard errors, because a simulation's results are themselves estimates with uncertainty. You need it any time you use simulation to justify a method, including simulation-based power analyses.

What they all have in common

Read side by side, the seven guidelines ask for the same few things:

Say what you planned before you saw the data, and flag anything you decided afterwards. SPIRIT exists for this. CONSORT 2025 and ADEMP build it in.
Report uncertainty, not just point estimates: confidence intervals, posterior intervals, Monte Carlo standard errors.
Give enough detail for someone else to redo it. Every one of these is a reproducibility guideline at heart, and the newer ones say so outright by asking where the data and code live.
Account for everyone and everything. Participant flow in CONSORT, missing data in JARS-Quant and TRIPOD+AI, failed or non-converged runs in a simulation.

How to pick, in practice

Start with the design guideline your study type calls for. Then add an analysis guideline if your methods need one. A Bayesian randomised trial uses SPIRIT for the protocol, then CONSORT plus BARG for the paper. A psychology experiment with a simulation-based power analysis uses JARS-Quant, plus ADEMP for the simulation. A machine-learning risk score uses TRIPOD+AI. SAMPL applies underneath all of them.

Two practical notes. First, look these up before you collect data, not while you're formatting the manuscript. Much of what they ask for can't be reconstructed afterwards. Second, many journals now require a completed checklist at submission. The EQUATOR Network catalogues hundreds more guidelines, including extensions for cluster trials, observational studies and systematic reviews, if none of these fits.

If you're planning a study and want the methods, analysis plan and write-up built around the right reporting standard from day one, that's what our work with researchers looks like.

Sources and further reading

Get new posts by email

One email whenever we publish something new. No spam, unsubscribe anytime.

Check your inbox — click the confirmation link to finish subscribing.