Communicating methods and statistics clearly
Scientific communication covers far more than press releases. For most researchers it starts with explaining, in a manuscript or protocol, how data were analysed and why those choices were made. The articles collected here focus on that problem: how to describe a statistical approach so that reviewers can judge it, readers can trust it and other teams can reproduce it. Topics range from analysis planning for a hypothetical prospective cohort, to the choice between fixed-effect and random-effects models in meta-analysis, to biostatistical monitoring of clinical risk, to meta-analytic structural equation modelling with the R package metaSEM, and a quick-reference table for writing the introduction of a systematic review.
Questions readers bring to this section
- Which model should my meta-analysis use? A fixed-effect model assumes one true effect shared by all studies; a random-effects model assumes the true effects vary and estimates their average. The choice should rest on the clinical and methodological diversity of the included studies, decided in the protocol, not on the result of a heterogeneity test.
- What analyses suit a prospective cohort? Typical choices include incidence rates, risk ratios, Cox proportional hazards models for time-to-event outcomes and methods for handling loss to follow-up, all reported in line with STROBE.
- How can statistics support safety monitoring? Approaches include predefined stopping boundaries reviewed by a data monitoring committee, signal detection in adverse event data and risk-based monitoring as described in ICH E6.
- When is structural equation modelling useful in a meta-analysis? When the question concerns relationships among several variables, such as mediation, rather than a single pooled effect.
Writing statistics for different audiences
The same analysis is described differently depending on the reader. In the methods section, give enough detail for replication: software and version, model specification, handling of missing data, adjustment variables and the threshold for significance. In the results, report effect sizes with confidence intervals rather than P values alone, following the SAMPL guidelines for basic statistical reporting. In the abstract and in lay summaries, translate relative effects into absolute terms where possible, since a 50% relative reduction can mean very different things depending on baseline risk.
Readers at the planning stage will find the cohort and meta-analysis articles most useful; those writing up will benefit from the systematic review reference table; and teams responding to a statistical reviewer can use the model-choice article to justify their approach.
Related resources
For a step-by-step treatment of pooled analysis, read how to do a meta-analysis for a manuscript, and for what journals check at review, what a statistical review involves. Support with analysis plans and interpretation is available through our biostatistics and statistical programming services, and wider audiences can be reached through simplified abstracts.


