What a preregistration contains
A preregistration is a public, time-stamped document that records the essential elements of a study before any outcome data are collected or analyzed. The core elements of a well-formed preregistration are:
The research question and hypotheses. A specific statement of what the study is asking and what outcomes would count as evidence for or against each hypothesis. Vague hypotheses (e.g., "we predict a difference") are insufficient; the hypothesis should be stated specifically enough that a reader could determine whether it was confirmed without consulting the authors.
The experimental design and conditions. Which conditions will be run, how they will differ, and how subjects or specimens will be assigned to conditions (where assignment is relevant). For physical measurement studies, this includes which specimens will be measured, what stimuli will be applied, and in what order conditions will be measured.
The sample size and stopping rule. How many observations will be collected, and what the stopping criterion is. An optional stopping rule — "we'll collect data until we get a significant result" — inflates the false positive rate even with a preregistered analysis, because many significance tests against a growing dataset produce a significant result by chance eventually.
Exclusion and outlier criteria. Which data points will be excluded from analysis, and on what grounds. These criteria must be stated before data collection to prevent outcome-contingent data exclusion, where an inconvenient outlier is removed because it weakens the finding.
The primary analysis plan. Which statistical test will be used, what the primary outcome measure is, and what result would constitute support or non-support for the main hypothesis. Secondary analyses may also be preregistered but should be labeled as such.
The document is filed with a repository that timestamps and locks the submission before data collection begins, creating a verifiable record that is independent of the researchers' later claims about what they planned.
Why it matters
The motivation for preregistration is the problem of analysis flexibility, also called "researcher degrees of freedom." In a typical data analysis, there are many defensible choices: which observations to exclude, whether to transform a variable, which covariates to include, which statistical test to apply. Each of these choices affects the p-value or effect size estimate. When the same dataset is analyzed with multiple combinations of these choices, there is a high probability that at least one combination will produce a nominally significant result even when no true effect exists.
Crucially, this process does not require bad intent. Researchers typically examine their data thoroughly before writing up, and many of the choices described above are made to handle real complications in the data (a skewed distribution, an obvious equipment malfunction, a meaningful confound). The problem is that even well-intentioned choices made after seeing the data are not statistically independent of the result, which inflates the apparent evidence for the hypothesis being tested.
Preregistration addresses this by separating confirmatory analysis — the analysis specified in advance for the primary hypothesis test — from exploratory analysis — the analysis performed in response to patterns observed in the data. Both types are scientifically legitimate. The distinction matters because confirmatory tests support statistical inference at the stated significance level, while exploratory findings are hypothesis-generating and require independent confirmation before they can be treated as established.
Without this distinction, readers cannot determine which analyses in a published paper were planned and which were invented after looking at the data. With preregistration, the analysis plan is on record, and any deviation from it is visible and must be reported and justified.
Registered reports
A registered report is a publication format that takes the logic of preregistration one step further. The manuscript is submitted to a journal in two stages. In the first stage, the introduction, hypotheses, methods, and planned analysis are peer-reviewed and evaluated for scientific soundness and relevance. If the journal finds the study worth conducting — regardless of how the results turn out — it issues an in-principle acceptance: a commitment to publish the final paper. The researcher then conducts the study, writes it up, and submits the second stage manuscript, which reports the results and discusses their interpretation.
The key feature of registered reports is that acceptance is based on the quality of the question and design, not on the direction or magnitude of the result. This eliminates the publication bias mechanism that creates pressure to produce significant results and discourages researchers from reporting null findings. It also makes deviation from the preregistered protocol more visible, because reviewers can compare the second-stage manuscript against the registered plan they already reviewed.
Registered reports are now offered by more than 300 journals across disciplines. Their adoption is one of the most structural responses to the replication crisis, because they change the incentive structure rather than relying on individual researcher virtue.
Preregistration is not a guarantee
Preregistration improves transparency but does not eliminate all problems in research practice. Several limitations are worth understanding:
A preregistration that is poorly specified provides limited protection. If the analysis plan says only "we will use regression," without specifying covariates, functional forms, or how outliers will be handled, substantial flexibility remains available after the data arrive. The quality of a preregistration scales with the specificity of the analysis plan.
Preregistrations can be violated. Researchers may deviate from their registered plan for legitimate reasons (unforeseen data quality problems, a design error discovered mid-study) or for illegitimate ones (the registered analysis produced a null result and a different analysis found significance). Both types of deviations should be reported transparently in the paper, with the registered and actual analyses both presented. Readers should check whether deviations are acknowledged and justified.
Preregistration can, in principle, be filed after the fact — a practice sometimes called "HARKing" (Hypothesizing After Results are Known) when combined with false claims of prospective planning. The timestamp provided by a public repository provides evidence against this, but only if readers verify that the timestamp precedes data collection.
Preregistration addresses the analysis-flexibility problem; it does not address measurement quality, sample representativeness, confounding, or design validity. A preregistered study using poor measurement equipment or an inadequate comparison condition will produce an uninformative result regardless of how carefully the analysis was specified in advance.
How S.P.H.E.R.E. approaches it
The LOC-1 protocol was designed with preregistration as a structural requirement, not an afterthought. The protocol is registered on the Open Science Framework at osf.io/5b6mj, establishing a timestamped public record of the hypothesis, measurement design, and analysis plan before any data collection takes place. The primary analysis is specified in the protocol document; exploratory analyses will be clearly labeled as such in any reporting.
The LOC-1 registration specifies the primary hypothesis (that geographic location produces a systematic, reproducible shift in measured resonance frequencies of controlled copper sphere specimens beyond what can be explained by known confounds), the primary measurement outcome (fractional frequency shift of identified resonance modes), the comparison conditions (same specimens at different geographic locations), the uncertainty budget structure, and the blinding protocol that will be applied during analysis. Deviations from the registered protocol, if any become necessary, will be reported transparently in the final publication. See also blinding in experiments.