Preregister an observational analysis by documenting what you already know

1. Decide what the registration needs to make visible

OSF describes preregistration as a time-stamped plan recorded before data collection or analysis and provides templates that include secondary-data research. The practical value for an observational project is a public or appropriately embargoed record of the intended questions and decisions. The existence of a template does not establish that your completed plan is sufficiently specific.

State whether data already exist, whether any team member has seen individual outcomes or aggregate results, and whether previous analyses informed the hypothesis. A registration after those observations can still document a plan from that point forward, but it cannot turn a decision already influenced by results into a prospectively specified one. Explain the actual sequence rather than selecting a flattering label.

Reference: OSF: registrations and preregistrations

2. Separate data existence from investigator knowledge

Consider a wholly fictional secondary-data study of appointment booking. Records already exist. On 10 July 2026 the analysis team has the variable dictionary but has not received individual data or seen outcome summaries; its planning notes disclose that knowledge state. On 12 July a completed registration records the intended analysis, with a receipt preserved in the exercise.

On 15 July the team receives the permitted dataset and examines outcome summaries. On 18 July it proposes a new hypothesis after seeing those summaries. The new hypothesis is exploratory relative to that knowledge, even if recorded before its own model is run. No real registry entry, ethics approval, access permission or study result is provided by this invented timeline.

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

3. Write decisions that another analyst can interpret

Identify the population, eligibility rules, source and version of the data, exposure, outcome, time window and estimand or descriptive target. Specify the intended model and justified covariates, handling of missing values, exclusions and any planned sensitivity analyses. Explain what would trigger a decision that cannot yet be fixed. Do not fill an unknown field with a generic default simply to complete the form.

For the appointment example, “analyse waiting time” is insufficient. Does time begin when a request is created or when a clinic receives it, and does it end at booking or the first visit? Identify the available fields and unresolved definitions. A plan cannot authorise treating different intervals as interchangeable. If a necessary timestamp is unavailable, the team must revise the question or state what cannot be estimated.

4. Distinguish the draft from the registered record

Prepare the plan with the actual contributors and required institutional checks, then inspect the selected registry’s current submission, administrator and visibility process. On OSF, a draft and a completed registration are separate states. Preserve the actual receipt and record URL after completion. Do not add an invented identifier to a manuscript because a form has been saved.

Choose public fields and attachments deliberately. Registration does not require placing patient-level or otherwise restricted material in an open document. In the fictional exercise, only the dated plan and its supplied receipt are asserted; no public dataset is supplied. The institution’s permissions and any necessary ethics decisions remain separate from what the registration service records.

Reference: OSF: registrations and preregistrations

5. Preserve later changes without rewriting the past

If measurement feasibility or a justified methodological reason changes the plan, retain the original and record the change, reason, timing and knowledge available. Use the registry’s supported update process where appropriate, and explain the deviation in the manuscript. An updated registration is useful context; it is not proof that the revised decision was made before the results were known.

For the hypothesis proposed on 18 July, an honest note says it followed inspection of outcome summaries and that its analysis has not yet been run. Do not recast it as a confirmed preregistered hypothesis or report a result. Later exploratory work may be informative when labelled and interpreted within its design; concealing its origin is the avoidable problem.

6. Check the plan against the report and the service’s current state

Before submission, compare the dated plan with the methods, results and analysis history. Identify planned analyses not performed, unplanned analyses performed and changes in definitions. Ask a colleague to check whether the report preserves those distinctions. A registration badge cannot substitute for this comparison or establish that an observational association is causal.

As of 15 September 2026, OSF has announced future changes to Projects: no new projects or components from 16 November 2026, and read-only Projects from 19 February 2027. Its notice states that Registrations continue. Check the current service information when arranging the active workspace; do not confuse a future Projects transition with closure of the registration record.

Reference: OSF: announced Projects transition

Questions for a practical final check

• Existing data and prior knowledge disclosed.
• Variables, timing and analysis decisions specified.
• Draft distinguished from completed registration.
• Sensitive material excluded from public fields.
• Later exploration and deviations dated.
• Manuscript checked against the original plan.