03 · What You Need to Know
What Preregistration Can Mean for Genuinely Exploratory Research
Exploratory Research Does Not Mean Research Without a Plan
Exploratory research is sometimes caricatured as opening a dataset and looking around until something interesting appears. Some exploration certainly can be highly open-ended, but exploratory studies often begin with considerable prior structure.
You may already know the phenomenon you want to investigate, the population or cases of interest, the dataset you will use, the variables available, the kinds of relationships you want to examine, or the analytical techniques you consider appropriate.
What you may not know is which patterns will emerge and which of those patterns will eventually warrant explanation or follow-up.
Those are different kinds of uncertainty.
Preregistration can document what was known and intended before the exploration while leaving the discoveries themselves genuinely open.
You Do Not Need to Invent Hypotheses
If you do not have a directional hypothesis, do not manufacture one for the sake of preregistration.
A researcher might genuinely begin with a broad question such as: “Which characteristics of students' online learning behavior are associated with subsequent course engagement?” If the purpose is to investigate several plausible relationships rather than test a prespecified prediction, that should be stated accurately.
Preregistration works best when it records the research that researchers actually intend to conduct. Turning an exploratory question into a fictional prediction makes the record less transparent rather than more rigorous.
Exploratory question
Asks what patterns, relationships, explanations, or possibilities emerge from systematic investigation without claiming that a particular result was predicted in advance.
Confirmatory hypothesis
Specifies an expectation before the relevant result is known and uses an advance analytical plan to evaluate that expectation.
Both can be preregistered. They simply record different research intentions.
Preregister the Scope of the Exploration
One useful question is: what exactly are you planning to explore?
You might specify the population, cases, records, texts, variables, time period, outcomes, predictors, or subsets that define the intended search space. If you are using an existing dataset, identify which dataset and which parts of it you intend to examine.
This can matter because exploratory claims can look very different depending on the size of the search space.
Finding an unusual relationship after examining three prespecified variables is not the same research process as finding it after searching hundreds of variables, transformations, subgroups, and model specifications. Neither discovery is automatically invalid, but readers may interpret the evidence differently when the breadth of exploration is visible.
You Can Preregister Analytical Starting Points Without Fixing the Entire Path
An exploratory preregistration might identify the broad analytical approach you intend to begin with.
For example, you might specify that you will examine distributions and missingness before evaluating pairwise relationships among a defined set of variables, followed by clustering or dimensionality-reduction procedures if the data satisfy stated conditions.
You might also identify software, data-processing procedures, initial models, visualization strategies, or decision thresholds where those are already known.
What you do not need to do is pretend that you can predict every analysis that the exploration will inspire. If an unexpected pattern leads to a new model or subgroup analysis, that later step can simply be documented as emerging from the exploration.
Preregister Decision Rules Where They Can Be Known in Advance
Exploratory work often contains decisions that can be specified prospectively even when the eventual findings cannot.
You may know how missing data will initially be handled, which observations are eligible, what constitutes an unusable record, how variables will be constructed, or which criteria will trigger follow-up analysis.
For example, instead of preregistering exactly which relationship will prove interesting, you might state that relationships exceeding a specified effect-size threshold will receive additional investigation, provided that such a threshold genuinely reflects the intended procedure.
Conditional plans can also be useful: if a variable has a severely skewed distribution, use a stated transformation or robust procedure; if an exploratory model reveals a particular structural feature, conduct a prespecified follow-up.
These rules can make the exploratory process more interpretable without eliminating its responsiveness.
Existing Data Require Particular Honesty About Prior Knowledge
Exploratory research frequently uses datasets that already exist. In that situation, the crucial question is not whether the data have already been collected. It is what the researchers have already seen.
If you have inspected the relevant outcomes, correlations, visualizations, or previous analyses before registering the study, say so. A registration created afterward cannot establish that those observations played no role in shaping your research questions.
That does not make the research unusable. It changes what the registration can establish.
The Open Science Framework currently provides a dedicated Secondary Data Preregistration template, reflecting the fact that existing-data research presents distinct issues of prior knowledge and timing.
Watch Out
Do not describe an analysis as prospectively preregistered merely because the registration was submitted before you ran the final statistical command. If earlier inspection of the same data informed the question, variable selection, model, or hypothesis, disclose that prior knowledge.
Preregistered Does Not Mean Confirmatory
This distinction is essential.
“Preregistered” describes something about when research decisions were recorded. “Confirmatory” describes the inferential relationship between an advance prediction and the evidence used to test it.
A preregistration can explicitly state that a study is exploratory. Public OSF registrations, for example, include projects that describe themselves as initial exploratory studies while prospectively registering their methods and analyses.
Conversely, simply labeling an analysis confirmatory does not make it so if the hypothesis or analytical choices were substantially shaped by prior inspection of the same evidence.
This is why researchers should understand the difference between exploratory and confirmatory research independently of whether a registration exists.
New Hypotheses Can Emerge From a Preregistered Exploration
One purpose of exploration is hypothesis generation.
Suppose an exploratory analysis reveals an unexpected relationship between two variables. You may propose a mechanism that could explain it and formulate a new hypothesis.
That hypothesis is a legitimate scientific product of the exploration. What the current dataset generally cannot provide is a fully independent prospective test of a hypothesis that was generated from patterns in those same data.
A subsequent study or independent dataset can provide that stronger test.
The sequence can therefore be productive rather than embarrassing: exploration generates an idea, the idea becomes a hypothesis, and later evidence tests it prospectively.
Exploratory and Confirmatory Analyses Can Coexist in One Study
A study does not need to choose a single identity.
You might preregister two confirmatory hypotheses and their primary analyses while also planning broader exploratory analyses of secondary variables. Alternatively, an exploratory study might generate a hypothesis early enough that an independent holdout sample or later wave can be used for prospective testing.
The important task is to keep the evidential roles clear.
The Center for Open Science's Registered Reports guidance similarly emphasizes that exploratory analyses are not prohibited. Under that publication format, unregistered exploratory analyses can be reported separately from preregistered confirmatory analyses.
The principle applies more broadly: exploration and confirmation can strengthen one another when they are distinguished rather than collapsed into a single narrative.
Preregistration Can Reveal the Breadth of the Search
One less obvious benefit of exploratory preregistration is that it can document how large the initial search space was.
Imagine that researchers ultimately report one striking association. Without additional context, readers may not know whether that relationship was one of five examined or one of five hundred.
An advance record can identify the variables, outcomes, models, or procedures researchers initially intended to explore. Later deviations and additions can then be documented separately.
This does not convert the resulting association into a confirmatory finding. It provides information needed to understand how the finding was discovered.
Preregistration Can Also Improve Exploratory Planning
Exploratory does not mean improvised.
Writing down the intended scope can expose ambiguities before analysis begins. Which dataset version will be used? Which observations are eligible? Which variables are conceptually relevant? How will missingness be addressed? What transformations are permitted? What counts as a follow-up worth pursuing?
Answering these questions can make exploration more systematic while leaving room for unexpected discoveries.
The Open Science Framework supports multiple registration templates rather than prescribing a single form for all studies, including a general-purpose preregistration, an open-ended registration, and specialized forms such as Secondary Data Preregistration. The appropriate template should reflect the actual research design rather than forcing exploratory work into an unsuitable structure.
Sometimes Preregistration Adds Little to an Exploration
Not every exploratory project needs a detailed prospective plan.
If the purpose is genuinely unconstrained familiarization with a new dataset, brainstorming, methodological experimentation, or early-stage discovery in which the analytical path cannot meaningfully be anticipated, an elaborate preregistration may contribute little.
The broader question is whether preregistration adds useful transparency to this particular research process.
When it does, preregister what can honestly be specified. When it does not, do not manufacture constraints merely to acquire a preregistration label.
