<p>Bibliographic metadata underpin large-scale studies of researcher mobility by enabling affiliation timelines to be reconstructed from publication records. Yet the field lacks a systematic benchmark of false-positive affiliations and how they distort mobility indicators, especially at institutional granularity. We evaluate OpenAlex and Scopus using a cohort from the 2020 Neural Information Processing Systems conference. Our two-stage audit proceeds as follows. First, we construct institution- and country-level timelines for a sample of 100 authors using a mode-per-year rule and validate yearly states and reported moves. Second, we manually review 2093 publications for a 25-author sample, verifying each author–publication–affiliation link and classifying errors as author disambiguation or affiliation extraction. Three findings stand out. Institution-level timelines are less accurate than country-level timelines in both platforms: average author-weighted affiliation accuracy is about 0.62 versus 0.75 in OpenAlex and about 0.70 versus 0.86 in Scopus. Platform aggregates favor Scopus, but within-author comparisons on overlapping years show much smaller differences (about <InlineEquation ID="IEq1"> <EquationSource Format="TEX">$-0.03$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mo>-</mo> <mn>0.03</mn> </mrow> </math></EquationSource> </InlineEquation> at institutions and <InlineEquation ID="IEq2"> <EquationSource Format="TEX">$-0.07$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mo>-</mo> <mn>0.07</mn> </mrow> </math></EquationSource> </InlineEquation> at countries, with intervals including zero), indicating composition and coverage effects. Many reported migration events lack two verified endpoints: positive predictive value is about 0.29 in OpenAlex and 0.45 in Scopus at institutions, so raw counts overstate mobility. Error mechanisms differ by platform: in OpenAlex, errors arise from both identity and affiliation extraction; in Scopus, affiliation extraction dominates (about 1.25 errors per publication versus 0.06 in OpenAlex). We provide a cross-platform benchmark and a replicable protocol, and recommend reporting event validity, using simple concordance checks, and prioritizing mechanism-targeted improvements.</p>

Affiliation errors and distorted researcher mobility: evidence from OpenAlex and Scopus

期刊论文
出版模式：

开放获取
发表日期：

2026年6月22日

查看全文

Mitch Mitchell,
Stephanie L. Johnson,
Michael D. Porter

摘要

Bibliographic metadata underpin large-scale studies of researcher mobility by enabling affiliation timelines to be reconstructed from publication records. Yet the field lacks a systematic benchmark of false-positive affiliations and how they distort mobility indicators, especially at institutional granularity. We evaluate OpenAlex and Scopus using a cohort from the 2020 Neural Information Processing Systems conference. Our two-stage audit proceeds as follows. First, we construct institution- and country-level timelines for a sample of 100 authors using a mode-per-year rule and validate yearly states and reported moves. Second, we manually review 2093 publications for a 25-author sample, verifying each author–publication–affiliation link and classifying errors as author disambiguation or affiliation extraction. Three findings stand out. Institution-level timelines are less accurate than country-level timelines in both platforms: average author-weighted affiliation accuracy is about 0.62 versus 0.75 in OpenAlex and about 0.70 versus 0.86 in Scopus. Platform aggregates favor Scopus, but within-author comparisons on overlapping years show much smaller differences (about $-0.03$ $- 0.03$ at institutions and $-0.07$ $- 0.07$ at countries, with intervals including zero), indicating composition and coverage effects. Many reported migration events lack two verified endpoints: positive predictive value is about 0.29 in OpenAlex and 0.45 in Scopus at institutions, so raw counts overstate mobility. Error mechanisms differ by platform: in OpenAlex, errors arise from both identity and affiliation extraction; in Scopus, affiliation extraction dominates (about 1.25 errors per publication versus 0.06 in OpenAlex). We provide a cross-platform benchmark and a replicable protocol, and recommend reporting event validity, using simple concordance checks, and prioritizing mechanism-targeted improvements.