What Was the Primary Analysis in the Research Study?

Which analysis was supposed to provide the study's main answer?

A results section may contain unadjusted comparisons, adjusted models, subgroup analyses, alternative outcome definitions, sensitivity analyses, interactions, several follow-up times, and enough supplementary tables to test the patience of even enthusiastic reviewers. Which one is the primary analysis?

The primary analysis is not simply the analysis with the smallest P value, the largest effect, or the most prominent graph. It should be tied to the study's primary question and, in confirmatory research, ideally specified before the results are known. Identifying it helps you determine which result carries the main inferential burden and which analyses play supporting or exploratory roles.

Find the analysis intended to answer the primary question

How to identify the analysis carrying the main inferential weight

Start with the primary question and primary outcome

The primary analysis should make sense in relation to what the study was actually trying to answer. If a randomized trial's primary objective concerns whether an intervention changes depressive symptoms at 12 weeks, the primary analysis should ordinarily correspond to that question rather than to an interesting subgroup result at week 4.

Similarly, identifying the actual outcome and its primary time point helps narrow the search. A study can collect many outcomes without treating them all as primary.

The primary analysis is more than the name of a statistical test

Saying that researchers used “linear regression,” “ANOVA,” “logistic regression,” or a “t test” does not fully specify the primary analysis.

A useful reconstruction usually includes several elements:

In modern clinical-trial methodology, another useful concept is the estimand: a precise description of the treatment effect the trial seeks to estimate. ICH E9(R1) emphasizes alignment among the trial objective, estimand, design, data collection, and statistical analysis. It also distinguishes the main estimator from sensitivity and supplementary analyses.

Look in the statistical methods before looking at the results

If you identify the primary analysis only after seeing which result appears most impressive, you risk allowing the results to define the question retrospectively.

Start with the statistical-analysis section. Look for wording such as “primary analysis,” “primary outcome was analyzed using,” “main analysis,” or a description explicitly connected to the primary objective.

Then compare that description with what actually appears in the results. The methods tell you what the authors say they planned to do; the results tell you what they report having done.

Prespecification strengthens the distinction between primary and additional analyses

For confirmatory studies, the analytical hierarchy is most informative when it is established before researchers know the relevant results. A protocol, trial registration, or statistical analysis plan may therefore be important for determining whether the reported primary analysis was genuinely planned as primary.

CONSORT 2025 asks trial reports to describe statistical methods for primary and secondary outcomes, identify additional analyses such as subgroup and sensitivity analyses, and distinguish prespecified analyses from post hoc ones.

If the published paper and earlier documentation differ, examine what was prespecified and what appears to have been decided later.

The primary analysis should identify the analysis population

A result cannot be understood without knowing who contributed to it.

In a randomized trial, the primary analysis might preserve randomized assignment while using a specified approach to missing outcomes. Another study might define a different analysis population according to its research question. In observational research, a regression model may use only participants with the necessary exposure, outcome, and covariate information.

Do not rely on the study's headline sample size. Determine whether all recruited participants were included in the analysis and what inclusion rule applied to the primary estimate.

Adjustment can be part of the primary analysis

Researchers sometimes report both crude and adjusted estimates. The adjusted estimate may be designated as primary, particularly in observational studies where the analysis attempts to account for specified confounders, or in trials where prespecified covariate adjustment is used.

STROBE asks observational-study reports to provide unadjusted estimates and, when applicable, confounder-adjusted estimates with their precision, and to make clear which confounders were adjusted for and why.

Do not assume that “adjusted” automatically means more valid. The appropriateness of adjustment depends on the variables selected, their measurement, the causal structure, the model, and the study question. Your immediate task here is narrower: determine which model was intended to carry the primary interpretation.

The main table is not necessarily the primary analysis

Formatting can mislead. The first results table may contain descriptive characteristics. A visually striking figure may show a secondary analysis. The abstract may emphasize a result that the protocol did not designate as primary.

Primary status comes from the analytical hierarchy and research plan, not graphic prominence.

A sensitivity analysis is not simply another primary analysis

A sensitivity analysis examines whether the conclusion from the main analysis remains credible when particular assumptions or analytical choices are varied. ICH E9(R1) describes sensitivity analysis as a way to assess robustness of conclusions from the main statistical analysis while remaining aligned with the same estimand.

If the sensitivity analyses agree with the primary analysis, confidence in the robustness of the conclusion may increase. If they differ materially, the disagreement deserves investigation rather than choosing whichever result you prefer.

Supplementary analyses may answer related but different questions

ICH E9(R1) distinguishes supplementary analyses from sensitivity analyses. Supplementary analyses can provide additional insight and may address a different estimand from the main analysis. They generally play a lesser role in interpreting the principal treatment effect when the main analysis is reliable.

This distinction is useful beyond regulated clinical trials. An additional analysis can be scientifically informative without becoming the study's main answer.

Several “primary” outcomes can complicate the hierarchy

Some studies designate more than one primary outcome. Others have several co-primary endpoints, primary outcomes at multiple time points, or multiple primary hypotheses.

In such cases, there may be more than one corresponding primary analysis. Determine how the researchers handled multiplicity and what combination of results was required to support the intended conclusion.

Do not simplify a genuinely multidimensional primary analysis into one preferred result simply because it is easier to summarize.

The primary analysis is not necessarily the analysis you would have chosen

You may disagree with the model, covariate set, missing-data assumptions, outcome definition, or analysis population. That is a separate appraisal question.

First identify the primary analysis accurately. Then evaluate whether it was appropriate. Otherwise, your preferred analysis can quietly replace the one the study actually used.

Observational studies may have a less explicit hierarchy

Not every observational paper formally labels one model as “primary.” You may instead find a sequence of crude and adjusted models, several outcomes, or multiple related hypotheses.

In such cases, look for the analysis most directly tied to the stated objective and identify whether the authors designate a principal model or estimate. If the hierarchy remains unclear, say so. Do not manufacture a primary analysis merely because one model looks methodologically sophisticated.

Finding the primary analysis among several plausible results

Common mistakes when identifying the primary analysis

Write the primary analysis as a complete analytical sentence

Instead of writing “they used regression,” reconstruct the primary analysis in enough detail that another researcher could understand what the estimate represents.

A useful sentence might read: “The primary analysis compared 12-week outcome scores between randomized groups according to assigned intervention using a prespecified regression model adjusted for baseline outcome score, with missing outcomes handled using the specified method.”

Once you have identified the primary analysis, the surrounding results become easier to organize. Instead of seeing a flat collection of P values and models, you can distinguish the analysis intended to answer the main question from analyses intended to probe robustness, extend the inquiry, or generate new hypotheses.

Before accepting the paper's main result, identify its primary analysis

Questions about identifying the primary analysis

The main analysis should be identified before the most attractive result

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.