.jpg){kind=link}
- Published
- Updated
Many rows in an analysis dataset do not establish an equal number of participants. Visits, organs or recurrent events can create several records for one person. Define the analytical unit and denominator explicitly before interpreting record counts as sample size.
Define the question and assumptions
State whether each table counts people, observations, events or another unit. Connect uniqueness keys to the scientific data structure and distinguish genuine repeats from duplicates. Joins can multiply rows unexpectedly. Review participant-level and observation-level counts separately rather than relying on total row count as the only reconciliation between analytical data and source information.
Hypothetical study example
In a hypothetical trial, visit records are joined with several laboratory observations. The same person-visit combination appears repeatedly and the table seems to contain a larger sample. The team corrects the join and uses source evidence to distinguish real repeated measurements from multiplication introduced by the processing operation.
Design review checks
- Specify the analysis unit and uniqueness key for every reported summary.
- Compare person and record counts before and after each major join.
- Define selection or modelling rules for genuine repeated observations explicitly.
Record the decision and its limits
Accepted outputs label participant and record counts consistently. Repeated contributions from one person are not presented as additional independent sample size. Duplicate handling should not become a crude filter deleting valid observations. Regenerate affected derivations and tables, documenting the defect scope and its numerical consequences in a traceable manner rather than correcting only the visible final count.
Frequently asked questions
- Is row count the sample size?
- Not always. It depends on the data unit and repeat structure.
- Does a repeated participant identifier indicate an error?
- Not necessarily. Different visits or events for one person may be valid.