Preregistration and Registered Reports are two ways to make research decisions visible before outcomes can influence them. In preregistration, researchers time-stamp hypotheses, methods and analysis plans before examining the relevant results. A Registered Report goes further: a journal reviews the research question and proposed method before data collection or analysis and may grant in-principle acceptance, meaning publication will not depend on whether results are statistically significant or exciting. These practices address selective reporting, publication bias and undisclosed analytical flexibility. They do not guarantee good science. A vague plan can be unhelpful, circumstances may require justified changes, and exploratory analysis remains valuable when labelled honestly. Their main contribution is to separate predictions from explanations developed after seeing the data.
Key takeaways
- Preregistration time-stamps a plan before the relevant results are known.
- Registered Reports add peer review and possible acceptance in principle.
- The methods and importance of the question are judged before the outcome.
- Deviations are allowed when they are disclosed and explained.
- Exploratory analysis remains useful but should be labelled as exploratory.
- Transparency improves interpretation; it does not replace sound design.
The problem of analytical flexibility
A dataset permits many choices: which cases to exclude, which outcomes to emphasise, which model to use and when to stop collecting data. If those choices are made after seeing results, apparently confirmatory evidence can be overstated.
A low p-value is harder to interpret when many unreported analyses were tried. The issue is not that researchers are necessarily dishonest; ordinary reasoning is influenced by knowledge of the result.
What a useful preregistration contains
A plan identifies hypotheses, sampling, inclusion rules, variables, outcomes, statistical models and decision thresholds. It should state when data will be examined and how missing values or multiple tests will be handled.
Specificity is a balance. A plan must constrain important choices without pretending every practical detail is predictable. Versioning and a time stamp allow readers to distinguish planned decisions from later revisions.
When preregistration occurs
For newly collected data, registration normally precedes data collection or at least access to outcomes. For existing data, researchers can register before inspection if they genuinely lack knowledge of the relevant variables.
The timing statement should be explicit. Calling a plan preregistered after results have been examined defeats its main purpose, although publishing a retrospective protocol can still improve documentation.
How Registered Reports work
At Stage 1, reviewers examine the question, theory, method and proposed analysis. They can request stronger controls, adequate sample size or clearer outcomes before the study proceeds.
If the plan is accepted, the journal offers in-principle acceptance subject to following the approved protocol and interpreting results responsibly. Stage 2 review checks execution and conclusions rather than rewarding a preferred outcome.
Reducing publication bias
Conventional publication can favour positive or surprising results, leaving null results unseen. Registered Reports make the publication decision before results exist, reducing this outcome-based filter.
A null result is not automatically decisive. Its meaning depends on precision, design and the range of effects compatible with the data. Confidence intervals help communicate that range.
Changes and deviations
Equipment breaks, recruitment falls short and unexpected distributions appear. Researchers may deviate from the plan when necessary, but they should document what changed, why and when.
A transparent deviation is not a failed registration. The record lets readers judge whether the change was scientifically justified and whether it risks tailoring the analysis to the outcome.
Confirmatory and exploratory analysis
Confirmatory analysis tests decisions specified in advance. Exploratory analysis looks for patterns, builds hypotheses and adapts to the data. Both contribute to science.
Problems arise when exploration is presented as if predicted. A strong paper can report the registered analysis first, then label additional work clearly and recommend independent testing.
What preregistration cannot solve
A registered weak design remains weak. Biased sampling, unreliable measurement and unrealistic assumptions are not repaired by a time stamp. The sampling-bias guide shows why the people or places observed still determine the population to which results apply.
Preregistration also cannot stop fraud on its own or ensure data and code are available. Ethics review, data management, replication and scrutiny remain necessary.
Peer review at the right time
Traditional review often occurs after resources are spent and design flaws cannot be corrected. Stage 1 review can improve the study before collection, which is a practical advantage of Registered Reports.
The general strengths and limits of evaluation are discussed in the scientific peer-review guide. Reviewers can miss problems, disagree or impose unnecessary rigidity, so editorial processes still matter.
Costs and trade-offs
Preparing a detailed plan takes time, and multi-stage review can delay a project. Longitudinal, emergency and highly adaptive research may be difficult to specify fully.
Templates, pilot work and adaptive protocols can help. The relevant question is not whether every study should use an identical format, but which decisions should be recorded before outcomes are known.
How to read a preregistered paper
Follow the registration link and compare dates, hypotheses, sample, outcomes and analyses with the paper. Look for declared deviations and clearly separated exploratory results.
Do not treat a preregistration badge as a substitute for reading methods. Its value lies in the auditable comparison between intention and execution.
Power, sample size and stopping rules
A registration should explain how the sample size was chosen. Power analysis, precision targets, feasibility and expected attrition may all contribute. Stating the assumptions exposes whether an apparently large study can answer its question and prevents the sample from being expanded or stopped simply because a desired result appeared.
Sequential and adaptive designs can still be preregistered. They specify interim looks, decision boundaries and permitted adaptations in advance. Flexibility is compatible with confirmatory inference when the rules governing that flexibility are visible.
Multiple outcomes and researcher degrees of freedom
A study measuring many outcomes creates more opportunities to highlight a favourable result. A plan can identify primary and secondary outcomes, families of tests and any adjustment for multiple comparisons. It can also specify transformations, covariates and exclusion rules that otherwise multiply analytical choices.
Not every decision needs to be frozen. Robustness checks can show whether conclusions survive plausible alternatives. A multiverse analysis is especially informative when several reasonable specifications exist, provided the registered primary test remains distinguishable from later sensitivity work.
Replication and cumulative evidence
Registered Reports make direct and conceptual replication easier to publish because an unexciting result does not threaten acceptance in the same way. Replication then tests whether an effect returns under the same procedure or generalises under a deliberate variation.
One registered study is not the final word. Meta-analysis still needs comparable outcomes, assessments of bias and attention to dependent samples. Registration improves the audit trail from hypothesis to result, helping later researchers understand which findings were predicted and which emerged during exploration.
A worked research example
A team wants to test whether a teaching method improves examination performance. Before recruitment, it identifies the primary outcome, randomisation method, exclusion rules, sample-size calculation and regression model, including the baseline score used as a covariate. It registers secondary outcomes such as confidence and attendance but makes clear that they will not replace the primary result. Reviewers of a Registered Report question whether schools rather than pupils are the true unit of assignment, leading the team to revise its power calculation before data collection.
Afterwards, one school withdraws and residuals violate an assumption. The authors report the registered model, document the deviation and add a robust alternative. Both estimates are shown. An unexpected improvement among lower-attaining pupils becomes an explicitly exploratory interaction that requires replication. The study is informative because readers can see which decisions preceded the outcomes.
Practical limits across disciplines
Laboratory experiments with standard protocols may be easier to specify than ethnography, outbreak research or discovery-driven astronomy. In qualitative work, registration can document sampling aims, researcher standpoint and planned coding without pretending that themes are knowable in advance. In emergency research, a core protocol and decision log may be more realistic than lengthy pre-study review.
The underlying principle is proportionate transparency. Researchers should pre-specify the decisions most vulnerable to outcome knowledge and document unavoidable adaptations. A rigid form that encourages meaningless answers is less useful than a focused record that makes the actual inferential boundary clear.
Conclusion
Preregistration records analytical intentions before results can shape them; Registered Reports add early peer review and an outcome-independent route to publication. Together they can reduce selective reporting, undisclosed flexibility and publication bias. Their success depends on precise plans, visible amendments and fair assessment of both confirmatory and exploratory work. They are best understood as tools for making the research process inspectable, not as certificates that a conclusion must be correct.
Frequently asked questions
Does preregistration ban exploratory research?
No. Exploratory analyses can be reported when they are labelled and distinguished from planned tests.
Can a preregistered plan be changed?
Yes, but changes should be time-stamped, disclosed and justified.
What is in-principle acceptance?
It is a journal commitment to publish a Registered Report if the approved study is conducted and reported appropriately, regardless of result.
Does registration prove a study is unbiased?
No. Sampling, measurement, assumptions and execution can still introduce bias.
Can secondary data be preregistered?
Yes, if prior access and knowledge are described honestly and the plan is registered before the relevant analysis.


