Does meta-research do parachute research?
Written against the design standard. The author’s own work on Cochrane authorship found it gender-diverse but geographically concentrated (PMID 39947447), and another of the author’s studies examined gender among dental faculty authors (PMID 38935641). Both concern who authors research. This asks a different question: whether the people evaluated are among the evaluators.
1. The question
Parachute research, in which investigators from wealthy countries study populations elsewhere without local collaborators, is criticized in global health. Meta-research has its own version: studies that grade the transparency, reporting or integrity of research from a named country. Whether such studies include anyone from the country whose research they are grading, and whether that depends on the country’s income, has not been asked of the field that asks it of everyone else.
2. Prior work, and the search behind the novelty claim
Log key P26.
corpus regex: (parachute|helicopter) research|(authorship|authors).{0,60}
(low.and middle.income|LMIC).{0,80}(meta-research|systematic review)
PubMed: ("parachute research"[tiab] OR "helicopter research"[tiab]
OR "local authorship"[tiab] OR "local authors"[tiab])
AND ("low- and middle-income"[tiab] OR LMIC*[tiab] OR "global health"[tiab])Twenty PubMed records, none in the corpus. They concern parachute research in global health (PMID 30103996), in microbial genomics (PMID 34704921) and authorship of global oncology research (PMID 35696623). All concern primary research; none turns the question on meta-research.
3. Hypothesis
Meta-research evaluating research from a low- or middle-income country includes at least one author affiliated with that country less often than meta-research evaluating research from a high-income country includes one from that country.
Null. The odds ratio, low- or middle-income against high-income target, is 1.
What would falsify it. An odds ratio at or above 1.
4. Design and unit
Cross-sectional. One article whose sample is defined by a single country is one unit. Multi-country samples are analyzed separately with the outcome defined per country.
5. Sample
Tier A articles whose studied literature is restricted to research from a named country, identified by a classifier on title and abstract gated on the author’s labels.
6. Exposure
World Bank income group of the target country in the article’s publication year: low and middle-income against high-income.
7. Outcome and instrument
At least one author with an affiliation in the target country, from OpenAlex institution country codes, dated to the article. Authors with no country in OpenAlex are counted, and the analysis is repeated treating them as local and as not local, to bound the effect of the gap.
8. Validation gate
The author labels 200 abstracts for whether the sample is country-defined and which country. Kappa on country-defined or not, and exact agreement on the country; below 0.80 or 90 percent the author labels the full eligible set.
9. Analysis
In R.
glmer(local_author ~ lmic_target + year_c + log(n_authors) + (1 | journal),
family = binomial, data = d)Author count enters because larger teams include local authors more often for reasons of size alone.
10. Threats
| threat | handling | removed or disclosed | standard |
|---|---|---|---|
| Missing affiliation countries | Bounded by treating missing as local and as not local | Disclosed | S3 |
| Diaspora authors abroad count as not local | Stated; affiliation measures institutional presence, not nationality | Disclosed | |
| High-income targets are mostly studied from home | That is part of the contrast, not a confounder | Disclosed | |
| Target-country extraction | Gated classifier with full-labeling fallback | Disclosed | S8 |
11. Probes
| probe | computes | goes to | what would change the design | status |
|---|---|---|---|---|
| PA | country-defined articles on an author-labeled pilot of 100 removed from the sample, and their income mix | 5 | fewer than 60 low- or middle-income targets in the full set makes the comparison imprecise | not run |
| PB | OpenAlex affiliation-country completeness for tier A authors, marginally | 7 | over 25 percent missing makes the bounds too wide to use | not run |
| PC | the author’s 200 labels | 8 | below the thresholds the author labels everything | needs the author |
12. What this cannot show
- Whether local authors had real influence on the work.
- Whether the absence of local authors changed any conclusion.