What Should Be Fixed Before Data Collection Starts?

What needs to stop being provisional before you collect data?

Early research planning can tolerate uncertainty. You might compare several sampling strategies, revise an interview guide, pilot alternative procedures, or reconsider how a variable should be measured. That flexibility can be useful while the study is still taking shape.

Data collection changes the situation.

Once participants have been recruited, measurements taken, interviews recorded, observations documented, specimens collected, or records extracted, some earlier decisions become much harder to reverse. A variable that was never measured cannot simply be recovered during analysis. Participants recruited under poorly defined eligibility criteria cannot always be made comparable afterward. An inadequate consent process cannot be repaired merely by writing a better explanation in the eventual paper.

The question is therefore not whether every detail of the project must become permanently fixed. It is which decisions need to be sufficiently settled before the study begins generating the evidence on which its conclusions will depend.

Finalize the decisions that determine what evidence you will actually have

Data collection is a commitment point, not just another task

Planning before data collection deserves special attention because collecting evidence creates consequences. Resources are spent. Participants may be exposed to procedures or asked to disclose information. Measurements become part of the dataset. Sampling decisions begin shaping who or what is represented. In some studies, investigators may also begin seeing patterns that could consciously or unconsciously influence later choices.

This does not mean that a study becomes completely immutable once the first observation is recorded. Legitimate changes sometimes become necessary. Rather, the burden of justification increases as decisions move from hypothetical plans to choices that affect actual data.

A useful question is:

If I postpone this decision until after data collection begins, could seeing the data, losing the opportunity to collect something, or changing procedures midway make the study harder to interpret?

If the answer is yes, the decision probably deserves resolution before collection begins.

Make sure the research question and study purpose are sufficiently stable

You should know what the study is intended to answer before collecting evidence for it.

This sounds elementary, but a broad topic is not enough. “Artificial intelligence in education,” “employee wellbeing,” or “social media and mental health” may identify an area of interest without specifying what evidence needs to be collected.

By this stage, the question should be stable enough to determine what population, cases, variables, outcomes, experiences, processes, documents, or other evidence the study requires. If you still cannot explain what information would answer the question, the project may not yet be ready for data collection.

If earlier decisions remain unresolved, return to what needs to be decided after settling on the research question before building detailed procedures around an unstable foundation.

Define who or what is eligible to enter the study

Before recruitment, sampling, record extraction, or case selection begins, establish what counts as an eligible unit of evidence.

For human-participant studies, this may mean inclusion and exclusion criteria and the population from which participants will be recruited. For document analysis, it may involve publication type, date range, language, source, or other eligibility rules. For secondary-data research, it may involve which records, observations, variables, or time periods will be included. Laboratory and field studies have equivalent decisions about specimens, sites, observations, or experimental units.

WHO's recommended research protocol format specifically includes the study population or sampling frame, participation criteria, sampling procedures, and study instruments among the elements to specify in advance.

Eligibility criteria matter because changing them after seeing who is available or what the emerging data look like can alter the population represented by the study.

Settle the sampling or case-selection logic

Knowing the population is not the same as knowing how observations will enter the study.

Quantitative studies may require decisions about probability or nonprobability sampling, allocation, stratification, clustering, recruitment targets, or sample-size justification. Qualitative research may instead use purposive, criterion, theoretical, maximum-variation, snowball, or other forms of sampling appropriate to the methodological approach.

The required level of advance specification differs. Some qualitative designs intentionally allow later sampling decisions to respond to developing analysis. A randomized trial typically requires far greater procedural specification before enrollment begins.

What should already be clear is the logic governing selection. Researchers should be able to explain why particular participants, cases, records, documents, sites, or observations are appropriate sources of evidence for the question.

Finalize the information you cannot afford to discover you forgot

Before collecting data, identify the information required to answer the research question and conduct the intended analysis.

For quantitative research, this may include primary and secondary outcomes, exposures, predictors, potential confounders, grouping variables, timing of measurements, and variables needed to describe the sample or address the analysis plan.

For qualitative research, the equivalent task is not necessarily to create a rigid list of variables. You should nevertheless know what domains, experiences, processes, interactions, meanings, or contextual information the study needs to explore.

This is one of the strongest reasons to think about analysis before collection. If a planned comparison requires a variable that you never collect, no statistical sophistication afterward can manufacture it. Research methods occasionally have a rather unforgiving relationship with hindsight.

Finalize or appropriately prepare the data collection instruments

Questionnaires, interview guides, observation forms, assessment tools, laboratory procedures, extraction templates, sensors, software systems, and other instruments determine what becomes data.

Before using them in the main study, they should be sufficiently developed for their intended purpose. Depending on the instrument and design, this may involve checking validity evidence, reliability, wording, scoring, language adaptation, permissions or licensing, technical configuration, usability, and pilot or pretesting.

WHO protocol guidance calls for instruments used to collect information, such as questionnaires, focus-group guides, observation forms, and case-report forms, to be documented as part of the methodology. It also emphasizes detailed specification of procedures and measurements.

“Finalized” does not always mean incapable of any later refinement. A semistructured qualitative interview may legitimately include responsive probing, for example. The relevant standard is whether the instrument is ready to generate evidence consistently and ethically under the chosen methodology.

Specify exactly how data will be collected

A list of variables or an interview guide is not yet a data collection procedure.

Researchers should know who will collect the data, where collection will occur, what participants or data sources will experience, how instructions will be delivered, what sequence will be followed where sequence matters, how long procedures will take, what equipment or software will be used, and how deviations or problems will be handled.

For multisite studies or studies involving several data collectors, procedural consistency becomes especially important. WHO guidance notes that methodology should be clearly defined and, where multiple sites conduct the same protocol, standardized.

Without this preparation, differences between data collectors, sites, sessions, or devices can become unwanted sources of variation.

Train the people who will collect the data

A carefully written protocol does not implement itself.

When multiple people collect data, they should understand the study procedures, eligibility criteria, instrument administration, consent process where applicable, documentation requirements, privacy expectations, escalation procedures, and what to do when unexpected situations arise.

Training needs depend on the research. Interviewers may need practice using probes without leading participants. Observers may need calibration. Laboratory personnel may need standardized specimen procedures. Abstractors extracting information from records may need clear coding rules and exercises to establish consistent interpretation.

If the study depends on procedures being applied consistently, training should occur before the main data collection rather than becoming an informal lesson learned from the first several participants.

Resolve ethics review and participant protections before applicable activities begin

For research involving human participants, ethical requirements are not details to be completed after data collection has started. WHO states that research involving human beings should undergo ethics review to protect participants' dignity, rights, and welfare.

Depending on the study and jurisdiction, researchers may need ethics approval or another formal determination before recruitment, consent, intervention, access to identifiable records, or other covered research activities begin. Researchers should verify the requirements of their own institution and jurisdiction rather than assuming that a project is exempt.

Participant information and informed-consent procedures should also match what will actually happen. WHO's protocol guidance calls for ethical considerations, informed-consent processes, and applicable participant information and consent documents to be incorporated into the research protocol.

Confirm permissions that are separate from ethics approval

Ethics approval does not necessarily grant access to a research site, institutional records, proprietary dataset, school, company, archive, platform, laboratory, or other restricted resource.

You may need permission from a data custodian, site administrator, records office, organization, community authority, copyright holder, database provider, or other gatekeeper. Data-use agreements, confidentiality agreements, licensing conditions, or information-security reviews may also apply.

These dependencies should be resolved before collection where they are prerequisites to lawful or authorized access. If several approvals and access conditions interact, explicitly plan around ethics approval, recruitment, data access, and other dependencies rather than treating them as parallel tasks that will somehow finish at the right time.

Decide how consent will work in practice

Where informed consent is required, having a consent form is not the same as having a consent process.

Researchers need to determine who will approach prospective participants, what information they will receive, how questions will be handled, how voluntariness will be protected, how consent will be documented where documentation is required, and what happens if someone declines or later withdraws.

The process may need adaptation for language, literacy, age, decision-making capacity, cultural context, remote participation, or particular forms of vulnerability. Applicable ethics requirements should guide these arrangements.

The crucial point is that participants should encounter the study that was actually reviewed and described to them, not a materially different procedure improvised after recruitment begins.

Plan what happens to the data immediately after they are created

Data management begins with the first observation, not when analysis starts.

Before collection, determine where data will be stored, how files will be named and organized, what identifiers will be used, how identifiable information will be separated or protected where appropriate, who will have access, how backups will work, how versions will be controlled, and what documentation will accompany the dataset.

WHO protocol guidance explicitly includes data handling, coding, monitoring, verification, and statistical or qualitative analysis planning.

Requirements may also come from funders. Under the NIH Data Management and Sharing Policy, covered researchers must prospectively plan how scientific data will be managed and shared, and NIH currently requires an updated structured DMS Plan format for applicable awards.

Those NIH requirements apply to covered NIH-funded or conducted research, not to every research project. Even when no formal data-management plan is mandated, however, the underlying planning questions remain useful.

Think through the analysis before collecting the evidence

You do not always need every analytical detail frozen before the first datum is collected. You do need enough of an analysis strategy to know that the proposed data can answer the research question.

For quantitative studies, this may require specifying primary outcomes, major comparisons, variable definitions, sample-size assumptions, planned models or statistical procedures, treatment of repeated measurements, and approaches to foreseeable missing data.

For qualitative research, it may mean identifying the analytic tradition or approach, how material will be prepared and coded, how interpretation will develop, and how the chosen process fits the research question.

WHO recommends that protocols describe data-management and statistical methods, including sample-size reasoning and approaches to missing or spurious data, while qualitative projects should specify how data will be analyzed in sufficient detail.

Analysis planning before collection is not merely administrative neatness. It is a diagnostic test for the design. Trying to explain how you will answer the question with the proposed data often reveals variables, comparisons, measurements, or contextual information that would otherwise have been forgotten.

Distinguish confirmatory decisions from exploratory possibilities

The need for advance specification is particularly important when the study intends to make confirmatory claims.

If researchers can examine the results and then choose which outcome to emphasize, which participants to exclude, which subgroup to analyze, or which model to report, the observed data may influence decisions that are later presented as though they were independent of those results.

Exploratory analysis is not a lesser form of research. It answers a different purpose. The problem arises when exploratory decisions made after seeing the data are presented as if they were prespecified confirmatory tests.

Where preregistration, a registered report, clinical trial registration, a statistical analysis plan, or another formal commitment applies, follow the requirements of that system. More generally, distinguish what was planned before observing relevant outcomes from what was developed afterward.

Establish quality-control procedures before quality problems occur

Researchers should decide how they will detect and address data-quality problems while collection is underway.

Depending on the study, this could include range and logic checks, duplicate detection, equipment calibration, inter-rater checks, review of interview recordings, field supervision, audit trails, missing-data monitoring, verification of record abstraction, or routine checks that procedures are being followed.

WHO's recommended protocol structure explicitly includes quality control and quality assurance for study conduct and data management.

The purpose is not to inspect the final dataset and hope problems are repairable. Some errors are much easier to correct while collection is still active.

Decide how deviations and unexpected situations will be handled

No protocol anticipates everything. Participants miss appointments. Equipment fails. A survey platform goes offline. A site changes its access rules. An interviewer accidentally skips a question. Recruitment is slower than expected.

Before collection, identify foreseeable problems and establish who has authority to respond. Clarify which events require simple documentation, which require consultation with the research team, and which may require protocol modification, ethics review, sponsor notification, or another formal process.

This preparation prevents an unexpected event from becoming an invitation to make consequential methodological decisions hurriedly.

Not everything has to become rigid

Fixing what matters does not mean eliminating legitimate methodological flexibility.

The boundary should be determined by the methodology and consequences of the decision. If you are uncertain whether an unresolved choice can safely remain open, revisit what can legitimately remain flexible at the beginning of a research project.

Use the protocol as the operational reference point

By the time data collection begins, important decisions should no longer exist only in scattered notes, email threads, or the researcher's memory.

A research protocol provides a structured record of how the study is intended to be conducted. WHO's recommended format includes the research rationale and objectives, study design, population, methodology, instruments, data management and analysis, ethics, project management, timeline, and other operational information.

The precise formality required varies among studies. A small low-risk project and a multisite clinical trial do not require identical documentation. The underlying principle is that the people conducting the study should have a sufficiently clear and current account of what they are supposed to do.

If you have not yet consolidated these decisions, consider whether it is time to create a research protocol before collection begins.

See what changes when a survey moves from planning to data collection

That final step illustrates why the pre-collection boundary matters. Discovering a missing variable while reviewing a draft survey may cost minutes. Discovering it after several hundred completed responses may change what the study can answer.

Many data problems begin as planning problems

Use reversibility to decide what must be settled now

Before beginning data collection, review every unresolved decision and ask what would happen if you made that decision only after seeing some of the data.

The objective is not to freeze the entire project on the evening before the first participant arrives. It is to ensure that the study crosses the data-collection threshold with its consequential decisions under control.

This is also a useful readiness test. If major decisions about the evidence, participants, instruments, ethics, access, procedures, data management, or analysis remain unresolved, you may need more planning. If the important decisions are settled and only legitimate methodological or operational flexibility remains, the project may be approaching the point where planning has done enough and the research needs to start.

Before collecting the first main-study data, check these items

Common questions before beginning data collection

Do not cross the data-collection threshold with avoidable uncertainty

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.