A project can have a strong research question, a carefully designed questionnaire and a capable field team—and still produce weak findings if the data are poorly managed. Duplicate records, inconsistent codes, unclear variable names and undocumented changes can quietly affect the results. Data quality is not a final cleaning exercise. It is a set of habits that should be built into the whole project, from planning to archiving.
Start with a clear data plan
Before collecting data, decide what information is necessary, who will collect it, how it will be stored, who may access it and how long it needs to be retained. Collect only information that serves a clear purpose. Sensitive personal information should not be gathered simply because a form has space for it. The plan should also explain how consent, confidentiality, data sharing and secure disposal will be handled in line with applicable ethics approvals and data-protection requirements.

Pilot the instruments and the process
A questionnaire can look clear to the research team but confuse participants. A pilot helps reveal ambiguous wording, response options that do not fit, survey logic errors, questions that take too long and translations that lose the intended meaning. Testing should cover the full workflow, including device settings, skip patterns, interviewer instructions, file naming and how completed records reach the analysis team.
For interviews or focus groups, pilot the topic guide and recording process. For administrative data, inspect the available fields and definitions before assuming they measure the concept you need. It is much cheaper to fix a problem before hundreds of records have been collected.
Build checks into collection
Digital tools can flag impossible values, missing required responses, inconsistent dates and contradictory answers while the participant or interviewer can still correct them. These checks should be sensible rather than overly restrictive: unusual values are not always errors. For paper questionnaires, establish a routine for reviewing completeness and entering data with verification procedures for key fields.
Monitor quality during fieldwork. Review completion rates, interview duration, patterns of missingness, repeated responses and differences between data collectors. A sudden change may indicate a genuine shift in the sample—or a training, translation or technical problem. Follow up respectfully and document what was checked and what action was taken.
Make the dataset understandable
A clean dataset should be understandable to someone who did not collect it. Use consistent variable names and coding rules. Maintain a data dictionary or codebook that explains each variable, its label, units, permitted values, missing-value codes and any derived variables. For example, a code of 99 should not sometimes mean “not applicable,” sometimes “refused” and sometimes a real score.
Keep raw data separate from cleaned and analysis-ready files. Preserve an untouched source copy, restrict access to it and carry out cleaning through documented steps. If a value is changed, there should be a clear reason and a traceable record rather than an unexplained manual edit.
Handle missing data and duplicates thoughtfully
Missing values may occur because a question was skipped, a participant was not eligible, a system failed or a respondent chose not to answer. These reasons are not interchangeable. Use distinct codes or fields where appropriate and document how each type is treated. Similarly, two records that look alike may be a duplicate submission, or they may be two legitimate visits by the same participant. Define unique identifiers and duplicate-checking rules before deleting anything.
Make the analysis reproducible
Whenever possible, use scripts in R, Stata, SPSS syntax, Python or another suitable tool to document data cleaning, scoring, analysis and table production. A script helps another analyst understand how a result was produced and makes it easier to rerun the work after a correction. Save the software version, package information, key settings and a short README explaining the order in which files should be used.
Version control or a structured change log helps the team see what changed and why. Store outputs separately from source data, and avoid naming files “final,” “final2” and “really-final” without a clear versioning system. Where sharing data is appropriate, provide a de-identified dataset, codebook and analysis scripts, subject to consent, ethics approval, data-protection rules and any data-use agreements. Open science does not mean making confidential data public.
Protect participants while protecting the data
Use access controls, strong authentication, secure backups and approved storage locations. Keep direct identifiers separate from research responses when feasible, and limit access to people who need it. Do not send identifiable data through casual messaging channels or leave downloaded files on shared devices. Agree on what happens when a team member leaves and how files will be transferred or destroyed at the end of the project.
A practical quality checklist
Before analysis, confirm that you have: a preserved raw file; documented cleaning rules; a codebook; checks for duplicates, ranges and inconsistencies; a record of missing data; secure storage and access permissions; reproducible analysis scripts; and a clear explanation of exclusions and derived variables. These steps may seem unglamorous, but they help ensure that findings are credible, reviewable and useful beyond the person who ran the analysis.
Good data management is not separate from research quality. It is one of the ways research quality becomes visible and repeatable.
