03 · What You Need to Know
When One Primary Outcome Helps and When It Is Not Enough
What Is a Primary Outcome?
A primary outcome is the result designated as central to evaluating the study's main objective.
In a confirmatory intervention study, it commonly provides the principal basis for determining whether the intervention achieved the prespecified effect of interest.
For example, a randomized study of AI-assisted writing instruction might designate argumentative writing performance at the end of the semester as its primary outcome.
Other outcomes, such as writing self-efficacy or satisfaction, may be secondary.
The distinction establishes an inferential hierarchy. It does not mean that secondary outcomes are scientifically unimportant.
Primary outcome designation should follow the research objective rather than whichever result eventually produces the most favorable statistical finding.
Why Do Many Confirmatory Studies Prefer One Primary Outcome?
A single primary outcome can make the study's central claim more precise.
It clarifies which result should drive sample size planning, what effect is most important, and how the principal analysis will be interpreted.
For example, if the main purpose is to determine whether an instructional intervention improves writing proficiency, designating writing performance as primary prevents satisfaction or engagement findings from being substituted for the intended outcome after data collection.
A single outcome can also reduce the multiplicity complications associated with testing several primary hypotheses.
However, simplicity is not sufficient justification for choosing an outcome that inadequately represents the research objective.
The primary outcome must still be appropriate, meaningful, and supported by suitable measurement evidence.
See what makes an outcome measure appropriate for a research question.
One Primary Outcome Does Not Mean One Measured Variable
A primary outcome may be represented by a composite score, a latent construct, a repeated-measures summary, or another prespecified quantity.
For example, argumentative writing performance may be evaluated through several rubric dimensions combined according to a justified scoring procedure.
The study may still designate the resulting overall writing score as one primary outcome.
Similarly, a longitudinal study may define its primary outcome as change in a score between two time points or a prespecified trajectory parameter.
Researchers should distinguish the number of observed indicators from the number of primary inferential outcomes.
Several questionnaire items do not automatically constitute several primary outcomes.
When Might Multiple Primary Outcomes Be Justified?
Some research questions cannot be adequately represented by a single outcome.
For example, a clinical intervention may need to demonstrate benefit while also evaluating a critical safety outcome.
An educational intervention may be designed to improve learning performance without increasing unacceptable student workload.
If both dimensions are indispensable to the main conclusion, a multiple-primary-outcome framework may be appropriate.
However, researchers must define how the outcomes jointly determine the study's conclusion.
Does success require favorable results on all outcomes, or is evidence on any one outcome sufficient?
These decisions have different statistical and interpretive implications.
The existence of several interesting outcomes is not, by itself, a sufficient reason to designate all of them primary.
Multiple Primary Outcomes Require an Explicit Success Rule
When several outcomes are primary, the protocol should specify how they contribute to the main inference.
| Primary outcome strategy |
Interpretation |
Planning implication |
| Single primary outcome |
One outcome provides the principal confirmatory test |
Plan precision or power around that outcome |
| Co-primary outcomes requiring success on all |
All specified outcomes must satisfy the success criteria |
Consider joint power and the intersection of requirements |
| Multiple primary outcomes where any may establish success |
Success may be declared through one or more specified outcomes |
Control relevant multiplicity under the inferential framework |
| Prespecified composite outcome |
Several components are combined into one defined outcome |
Justify components, scoring, and interpretation |
| Multivariate primary analysis |
Several outcomes are evaluated jointly |
Specify the joint estimand, model, and decision rule |
These strategies are not interchangeable.
For example, requiring all co-primary outcomes to demonstrate benefit is different from testing several outcomes and declaring success if any one is significant.
Researchers should establish the success rule before examining the data.
Why Multiple Primary Outcomes Can Create Multiplicity Problems
When researchers conduct several statistical tests, the probability of at least one false-positive result may increase if each test is evaluated independently at the same significance threshold.
For illustration, suppose three independent tests concern true null hypotheses, each evaluated at a two-sided significance level of 0.05.
This calculation applies to the specified independence and true-null assumptions. It is not a universal error rate for every study with three outcomes.
Correlations among outcomes, the decision rule, and the hypothesis-testing framework affect the relevant multiplicity problem.
For example, a co-primary strategy requiring success on every outcome has different type I error implications from a strategy allowing success on any one outcome.
Researchers should use multiplicity procedures appropriate to their inferential objectives rather than automatically applying one correction to every collection of outcomes.
Multiple Primary Outcomes Can Affect Statistical Power
Sample size planning becomes more complex when several primary outcomes determine success.
If success requires demonstrating effects on all co-primary outcomes, the probability of satisfying the joint criterion may be lower than the power for any one outcome considered separately.
Conversely, a strategy allowing success on any one of several outcomes may require multiplicity control that affects the critical thresholds and power.
Outcome correlations also matter.
Researchers should not calculate sample size for one convenient outcome and assume that the resulting sample is adequate for every primary hypothesis.
Simulation or specialized multivariate methods may be useful when analytical calculations are difficult.
The appropriate planning strategy depends on the intended claims and decision rule.
Primary Outcomes Should Be Prespecified in Confirmatory Research
Prespecification helps distinguish hypotheses established before observing the results from analyses developed afterward.
In confirmatory trials, the primary outcome should ordinarily be defined in the protocol, including the measurement method, assessment time point, and analysis.
Relevant trial registration and reporting requirements may also apply.
The SPIRIT guidance provides protocol reporting recommendations, while CONSORT addresses transparent reporting of randomized trials.
Researchers should follow the requirements relevant to their study type and jurisdiction.
Changing the primary outcome after examining results can create selective reporting concerns, particularly when the change favors statistically significant findings.
Legitimate protocol amendments may occur, but their timing, rationale, and consequences should be documented transparently.
Can Exploratory Research Have No Single Primary Outcome?
Yes. Exploratory studies may investigate patterns, relationships, or phenomena without one prespecified confirmatory endpoint.
For example, an exploratory survey may examine several dimensions of students' experiences with generative AI tools.
The objective may be to characterize emerging patterns rather than test one primary treatment effect.
Such a study should still have clear research questions and a defensible measurement plan.
Exploratory status does not justify unlimited analyses or selective presentation of favorable findings.
Researchers should describe the purpose of the investigation and distinguish exploratory results from confirmatory evidence.
Qualitative Research Does Not Necessarily Use Primary Outcome Terminology
Qualitative research often focuses on experiences, meanings, practices, processes, or contextual explanations.
For example, a phenomenological study may investigate how teachers experience the introduction of AI-supported assessment.
The central object of inquiry is not necessarily a numerical endpoint.
Researchers may organize the study around a primary research question and several subsidiary questions rather than a primary outcome.
Applying clinical-trial outcome terminology mechanically to such a design may create confusion.
The appropriate structure should follow the qualitative methodology and its standards of evidentiary adequacy.
Clarity of purpose remains essential even when the concept of a primary outcome is not applicable.
Mixed-Methods Studies May Have Different Outcome Structures
Mixed-methods research combines quantitative and qualitative evidence to address questions that may not be adequately answered by either approach alone.
For example, a study may estimate an intervention's effect on writing performance while also investigating how students experience AI-generated feedback.
The quantitative component may have a prespecified primary outcome.
The qualitative component may instead have a central inquiry concerning experiences or implementation processes.
The integrated research question should clarify how these forms of evidence contribute to the overall interpretation.
Researchers should not assume that every qualitative finding must be subordinate to the quantitative primary outcome.
Nor should they treat distinct forms of evidence as interchangeable measures of the same result.
Primary Outcomes and Primary Research Questions Are Related but Not Identical
A primary research question identifies what the study seeks to understand.
A primary outcome specifies a result used to address that question in designs where outcome-based inference is appropriate.
One primary question may involve several essential outcomes.
Conversely, a study may have multiple research questions organized around a shared outcome.
For example, a study may investigate the overall effect of an intervention on writing performance and whether that effect differs by baseline proficiency.
Both questions involve the same outcome but require different analyses.
Researchers should therefore avoid assuming a one-to-one relationship between the number of research questions and the number of outcomes.
What About Composite Primary Outcomes?
A composite outcome combines several components into a single defined measure.
For example, a composite educational performance score might combine multiple assessment dimensions.
However, a composite should not be created merely to avoid addressing multiple primary outcomes.
Its components should represent a coherent construct or decision-relevant combination, and the aggregation procedure should be justified.
Components may differ in importance, frequency, and responsiveness to intervention.
A favorable composite result can sometimes conceal unfavorable or negligible effects on particular components.
Researchers should therefore explain how the composite was constructed and consider reporting component outcomes when relevant.
Can the Primary Outcome Be Subjective?
Yes. A subjective outcome can be appropriate when it directly represents the phenomenon of interest.
For example, perceived stress, satisfaction, pain, or quality of life may be central outcomes in particular studies.
The important question is whether the measurement supports the intended interpretation.
However, subjective outcomes may be susceptible to response expectations, social desirability, or awareness of treatment assignment under some conditions.
Researchers should consider these risks in the design and analysis.
The distinction between subjective and objective measurement does not determine whether an outcome can be primary.
Should Safety and Effectiveness Both Be Primary?
Some interventions require evaluation of both beneficial and harmful consequences.
For example, an educational intervention might improve examination performance while increasing workload or stress.
Whether both dimensions should be designated primary depends on the research objective and intended decision.
In some studies, effectiveness may be primary while safety or burden outcomes are important secondary outcomes.
In others, demonstrating both benefit and acceptable safety may be essential to the main conclusion.
Researchers should define the decision rule and ensure that the design can support the intended claims.
Calling an outcome secondary does not make it ethically or practically unimportant.
How Should Researchers Choose One Primary Outcome When Several Seem Important?
Researchers should begin by identifying the principal decision or inference the study is intended to support.
They can then evaluate candidate outcomes according to conceptual relevance, stakeholder importance, measurement validity, expected effect timing, and feasibility.
For example, if an instructional intervention is intended primarily to improve writing competence, writing performance may be a stronger primary outcome than satisfaction.
However, if the research question concerns acceptability, satisfaction may be appropriate.
Where several outcomes are genuinely indispensable, a multiple-primary-outcome strategy may be more defensible than arbitrarily selecting one.
The choice should follow the scientific objective rather than a desire to simplify reporting at the expense of relevance.
What Happens When the Primary Outcome Is Missing for Some Participants?
Missing primary outcome data can reduce statistical information and potentially bias estimates.
The consequences depend on the missing-data mechanism, estimand, and analytical assumptions.
Researchers should plan retention and outcome ascertainment procedures to minimize avoidable missingness.
They should also specify appropriate missing-data methods and sensitivity analyses when relevant.
Designating one primary outcome does not eliminate missing-data concerns.
Similarly, having several primary outcomes may create additional challenges when different participants have different patterns of missing observations.
The National Research Council's report on missing data in clinical trials emphasizes preventing missing outcomes and evaluating the assumptions required for analysis.
Do Not Promote a Secondary Outcome After Seeing the Results
Suppose an intervention does not show a clear effect on its prespecified primary outcome but produces a statistically significant difference in a secondary outcome.
The secondary finding may be scientifically interesting and worthy of further investigation.
However, it should not automatically replace the primary outcome as the basis for claiming that the original confirmatory objective was achieved.
Doing so can introduce selective reporting and exaggerate the strength of the evidence.
Researchers should report the prespecified primary analysis and interpret secondary findings according to their inferential status.
Any changes to outcome hierarchy should be documented, including whether they occurred before or after outcome data were examined.
Use an Outcome Hierarchy Matrix
A structured outcome hierarchy can clarify the role of each measurement.
| Outcome |
Possible role |
Reason |
| Argumentative writing performance |
Primary |
Directly addresses the main instructional effectiveness question |
| Writing self-efficacy |
Secondary |
Examines a distinct psychological outcome |
| Student satisfaction |
Secondary |
Assesses acceptability rather than performance |
| Revision behavior |
Exploratory or mechanism-related |
May help investigate how the intervention operates |
| Long-term writing retention |
Secondary or co-primary depending on objectives |
Addresses persistence of learning |
The categories are illustrative rather than universal.
A study focused on long-term retention could legitimately designate that outcome as primary.
The hierarchy should reflect the research question and planned inference, not the perceived prestige of a particular outcome type.
Watch Out
Do not assume that every research design requires exactly one primary outcome, or that designating multiple outcomes as primary automatically strengthens a study. The outcome structure must match the scientific objectives and specify how the resulting evidence will be interpreted.
