A bibliometric approach to international scientific migration Henk F. Moed, Mhamed Aisati, Andrew Plume and Gali Halevi Elsevier (Netherlands, UK, USA)
Contents Introduction: Migration and co-authorship The model Technical aspects Study countries and approaches Results: Global patterns Publication (Scopus) vs. Survey (OECD) data Accuracy / robustness of migration indicators Conclusions
Introduction: Migration vs. co-authorship
Which country has these main collaborators? Main collaborators Argentina USA Portugal France Chile Brazil
Which country has these main collaborators? Main collaborators Thailand India Singapore Iran UK Malaysia
Which country has these main collaborators? Main collaborators France Hungary Germany Italy Bulgaria Romania
Which country has these main collaborators? Main collaborators UK China USA Nigeria Australia Netherlands South Africa
The model
International migration vs. co-authorship RelationshipDefinitionComment International co-authorship Authors from institutions located in different countries jointly publish a paper Country relates to where authors work, NOT to their nationality International migration A scientific author moves from one country to another
The model - 1 Theoretical concept; Interpretation Bibliometric construct / feature ResearcherAuthor id Research active (in a year)Publishing (in a year) Currently activePublishing in 2011 Starting a scientific career during years T1-T2 First publication during T1-T2 Young in 2011First publication year > 2000 Moving from country A to BPublishing authors work country changes consecutively from A to B Transients are deleted
The Model – 2 MasterPhD studentPost docSenior Master degree PhD degree Start senior career T First publi- cation 2nd p3rd p4th p BBBBBAA AABBBAA AAAAA BA 2 1 3
Technical aspects
Technical aspects (in Scopus) 1.Author-affiliation linking 2.Author profiling
Author-affiliation links Is Italian science declining? RESPOL, 2011 Cinzia Daraio (a) and Henk F. Moed (b) (a)Department of Management, Univ Bologna, Via U. Terracini, 28, Bologna, Italy (b)Centre for Science and Technology Studies (CWTS), Leiden University, The Netherlands
Author-affiliation link records AuthorYearAffiliation Daraio, Cinzia2011Univ Bologna,..., Italy Moed, Henk F.2011Univ Leiden,..., Netherlands
Author profiling in Scopus Assigns to each author a unique number Groups each authors documents based on similarity in affiliation, publication history, subject and co-authors Aims at high precision ( lower recall) New algorithm implemented in April 2011 Author feedback system in place
Categorization of authors in year T TYPE T-4T-3T-2T-1 T T+1T+2T+3T+4 Continuant XXX Newcomer XX Mover XX Transient X [Active but not publishing] X X
# Publishing authors per year and type (UK) CONTINUANTS MOVERS TOTAL TRANSIENTS NEWCOMERS Decline in # Newcomers 2009 Cohort can be followed for one year only
Important distinction Use of author ID data to assess an individual vs Statistical analysis of patterns in large datasets X
Study countries and approaches
Countries with the largest increase in publication output during (Scopus) > 10 %: Selected in current study
Study countries (10 fast growers + 7 big countries) D8EUBRICOther EgyptRomaniaBrazilThailand IranPortugalChina MalaysiaIndiaAustralia PakistanGermanyJapan ItalyUSA Netherlands UK
Two complementary approaches FeatureSynchronousAsynchroneous Publication years analyzed Fixed [2011] Variable [ ] Starting years of authors careers Variable [ ] Fixed [ ]
Results: Synchronous approach
Study set: % Young authors currently active in a study country Question: How many of them had stays abroad?
Results: Asynchroneous approach
Study set: authors starting their career in a study country Question: How many moved abroad, how many returned?
Study set: Young authors starting their career in a study country Question: How many move abroad, how many return? Post Docs returning to their home country? Post-Docs not returning to their country? Or PhD students returning after attaining their PhD?
Results: Publication (Scopus) data vs. Survey (OECD) data
Inconsistencies in OECD data on # FTE Res? Country OECD # FTE Res All (2007) OECD # FTE Res Gov+HE (2007) Scopus # Authors (2007) Ratio # authors / # FTE Res All Ratio # authors / # FTE Res Gov+HE UK 254,600159,100154, ITA 93,00056,200113, Differences in ratios between ITA and UK are almost a factor 2!
Inconsistencies in data on # FTE Res? Country # FTE Res All (2007) # FTE Res Gov+HE (2007) # Authors Scopus (2007) Ratio # authors / # FTE Res All Ratio # authors / # FTE Res Gov+HE DEU 290,800116,600150, UK 254,600159,100154, ITA 93,00056,200113, NLD 49,70023,80046, Differences in ratios between {ITA, NLD} and {DEU, UK} almost a factor 2
Results: Migration vs. Co-authorship
Migration and co-authorship patterns are statistically different N=694 Mean=0.75 STD=0.49 Skewness=1.59 Analyzed later
Map of countries with Ratio migration/collaboration > 1.2
Study country = TO COUNTRY FROM Country No. Co- authored papers No. Migrating authors Ratio % Migration / % Co- authorship PAKIND PRTBRA1, INDPAK NLDIRN USAIND12,0133, NLDPAK CHNTWN3,9791, USAIRN3, BRAPRT1, MYSNGA Language similarity drives migration stronger than it drives co-authorship Political tensions affect migration less than they affect co- authorship
Tentative error rates in migration indicators IndicatorSetNMean variat ion STDSkew- ness Rough error rate # or % authors per country All % # or % migrating authors per country pair Top 100 All % >10% Ratio migration/ Collaboration per country pair Top 100 All % >10% Variation = 100*(max-min)/ (2*mean)
Conclusions Migration analysis generates new insights into the global scientific network Relative indicators based on large numbers are insensitive to errors in author profiles.
Thank you for your attention
PART II Robustness / accuracy of migration indicators
Case study: Data sample analysed 100 author-ids in chemistry randomly selected from Scopus Search for other author-ids that relate to the same authors as those represented in the sample Precision of the 100 author-ids themselves was not examined Emphasis on recall rather than precision
Author Countries Asian countries (China, Japan and Korea) are over-represented (n=34) CHN USA JPN DEU KOR
Findings – 1 27% of the researchers did have additional author ID's For the 27 % researchers that did have additional ID's, in total 51 additional author IDs were found Half of the additional author IDs had 1 paper only; and 75 % at most 2
100 sample authors are linked with 151 author-ids # Author-ids with 1 paper = 2 x # authors with 1 paper # Author-ids with 2 papers = 1.6 x # authors with 2 paper # Author-ids with >=20 papers is 1.1 x # authors with >=20 papers
Implications for the migration study HomonymsSynonyms Definition/ Examples One author id relates to different persons (common names) A researcher has more than one author id (split identities) TendenciesThe more common a name, the larger the number of articles Second identity for author ids with few articles only (typically 1-3) In analysisNewcomers as from 2001 Transients are not taken into account
Sensitivity Analysis: publication thresholds 1. ALL 2. P > 2 3. P > 3
Sensitivity analysis: publ per year (Ppyr) thresholds 1. All 2. Ppyr < Ppyr < 5.0
No. Authors moving to a study country 1 2 3
% Authors moving to a study country 1 2 3
Ratio % migration/% co-authorship per country pair for author sets 1 and 2 Rank order hardly changes 1 2
Variation in # or % migrating authors per country-pair RankNo. Authors Mean variationSTD Variation = 100*(max-min)/ (2*mean) +
Tentative error rates in migration indicators IndicatorSetNMean variat ion STDSkew- ness Rough error rate # or % authors per country All % # or % migrating authors per country pair Top 100 All % >10% Ratio migration/ Collaboration per country pair Top 100 All % >10% Variation = 100*(max-min)/ (2*mean)
Conclusions Migration analysis generates new insights into the global scientific network Relative indicators based on large numbers are insensitive to errors in author profiles.