Does It Matter Where We Publish? Adam Eyre-Walker University of Sussex
Reasons Prestige peers employers Impact scientific community society
Measuring impact General difficult to measure Citations Easy to measure Velvet Underground effect “only sold 10,000 copies, but everyone who bought it formed a band” (Brian Eno)
Question Do papers of similar quality garner more citations in better quality journals?
Ideal experiment Papers made anonymous authors journal structure Assessed for quality by experts Assessments confidential Citations tracked
Research Assessment Exercise (RAE) Each department submits 4 publications per academic (75%) 4 indicators of Esteem (5%) Environment PhD student numbers Grant income Research structure Facilities Superceded by REF
RAE dataset 1170 papers published Scored by AEW Scored 1 to 4 (or unclassified) Subjects : evolutionary biology, genomics, bioinformatics, ecology, animal behaviour and organismal biology Ranked without reference to journal citations
RAE scores RAE scoreNo. of papers
Journal quality Impact factor Number times papers published in the two previous years are cited in target year Two components Quality of papers Effect of the journal on citations
IF categories IF categoryIFNo of journals No. of papers 1IF < < IF < < IF < < IF < < IF < < IF < < IF < IF >
Citations No. of citations from ISI Web of Science October 2008
Data For each paper Year of publication RAE score (RS) No. of citations Impact factor of journal (IF category) Otherwise anonymous
How good is the assessor? Correlation between AEW and co- scorers > 0.7 Correlation between RS and no. of citations within each IF category and within each year
How good is the assessor? Correlation between AEW and co- scorers > 0.7 Correlation between RS and no. of citations within each IF category and within each year Average r = 0.27 (p<0.001) Average r max = 0.73
RS v citation correlation r = p = 0.005
Central question Do papers of similar quality accumulate more citations in some journals than others?
RAE score 2 in 2003
RAE score 2 p < 0.001
RAE score 3 p < 0.001
ANCOVA RS gradep-value30 v (1.7) (0.3) (0.4)
ANCOVA RS grade p-valueAll 30 v 5 IF<10 30 v (1.7)1.2 (0.4) (0.3)3.0 (0.3) (0.4)5.4 (1.4)
Alternative explanation Systematically underestimate quality of papers in high ranking journals Test Assume no effect of journal Consider correlation between RS and journal (IF) controlling for citations
Partial correlations YearCorrelation *** *** *** *** *** *** ***
ANCOVA RS gradep-value30 v (1.7) (0.3) (0.4)
Question Impact factors – relative contribution of Paper quality Journal effect
Summary Assessor scores correlated to citations strength correlated to age Assessor scores influenced by journal Papers published in higher quality journals accumulate more citations independent of quality F1000 data
Question Where will we submit?
Thanks Nina Stoletzki
Assessor and journal Assume citations measure quality no journal effect