Two analysts working from the same database with two entirely defensible counting rules will produce noticeably different totals.
Which means the rule is part of the number — and publishing the figure without it publishes half a finding.
The decisions that change the total
- Whole or fractional counting for multi-institution work.
- Which affiliation applies — at submission, at publication, or current.
- Which author's affiliation counts — all, first, or corresponding.
- Which output types are in scope.
- Which date field defines the period.
- How versions of one work are handled.
Whole versus fractional
- Whole counting credits each institution fully.
- Fractional divides credit among contributors.
- Whole counting sums to more than the actual number of papers — expected, but it must be stated.
- Both answer different questions and both are legitimate.
- Never mix them within one report.
Affiliation timing
- Using current affiliation reassigns historical work to wherever someone is now — this misrepresents both institutions.
- Publication-time affiliation is usually the defensible choice.
- It requires affiliation history to have been retained.
- Systems that overwrite affiliation cannot compute this at all.
Version handling
- Decide whether preprint and published version count once or twice.
- Double counting is invisible in the output.
- State the rule with any aggregate.
- Never change the rule mid-series in a trend.
Reporting the rule
- State it alongside the figure, not in an appendix.
- State the data source and extraction date.
- State coverage limitations.
- Version the rule so past figures remain reproducible.
- Flag any change and provide overlapping figures for the transition.
When figures disagree
- Compare rules before comparing numbers — this resolves most disputes immediately.
- Check the extraction dates, since databases update retroactively.
- Check coverage differs.
- Two correct figures can differ substantially, and that is not an error in either.
One thing worth remembering
When two figures disagree, compare the counting rules before assuming either is wrong.
Most disputes about research statistics are disputes about definitions that nobody wrote down — and they resolve in minutes once the rules are on the table.
Câu hỏi thường gặp
Which decisions change the total?
Whole or fractional counting, which affiliation applies and when, which author's affiliation counts, which output types are in scope, which date defines the period, and version handling.
What is the difference between whole and fractional counting?
Whole credits each institution fully and sums to more than the actual paper count, fractional divides credit — both legitimate, answering different questions, and never mixed in one report.
Why does affiliation timing matter?
Because using current affiliation reassigns historical work to wherever someone is now, misrepresenting both institutions — publication-time affiliation is usually defensible but requires retained history.
How should the rule be reported?
Alongside the figure rather than in an appendix, with data source and extraction date, coverage limitations, and versioning so past figures remain reproducible.
What should be done when figures disagree?
Compare the counting rules before assuming either is wrong — most disputes are about definitions nobody wrote down.
