Last but not Least: Additional Positional Effects on Citation and Readership in arXiv

1 Introduction

• Earlier work found that articles near announcement beginnings received higher median citations through visibility and self-promotion effects.Self-promotion involved submissions timed to appear near the beginning; visibility described benefits for articles placed there without conscious intent.Source: §1 · pp. 1–2
• End positions may attract readers because the /recent page reverses article order within each day and shows titles and authors, potentially reducing the fatigue of scanning announcements.Submitters aware of that readership pattern could have an incentive to aim for the bottom of the list.Source: §1 · p. 2
• More than 80% of hep-th and hep-ph web readers in the earlier study used the canonical forward-order listing, so reverse-visibility was not pursued then.The vast majority also received announcements by legacy email.Source: §1 · p. 2
• The paper returns to reverse-visibility, noting that list-recall studies find primacy and recency advantages, though announcement scanning is a different task.Readers may scan less attentively through the middle and refocus near the end; submitter behavior also motivates the terms “willful primacy” and “willful recency.”Source: §1 · pp. 2–3
• A deadline can prompt a final burst of focused work, and submission counts show a smaller burst before the larger post-deadline spike.The paper calls this pattern the “procrastination effect,” a flip-side of self-promotion.Source: §1 · pp. 3–4

• Positions are indexed from both ends of each mailing around L/2 to compare variable-length lists; mailings shorter than 11 items for astro-ph or 7 for hep-th and hep-ph are excluded.The retained samples include roughly 770 astro-ph, 720 hep-th, and 750 hep-ph mailings.Source: §2 · p. 4, Figure 2 · p. 5
• End-position articles in hep-th and hep-ph show higher readership during the initial active period and higher median citations years later, with the citation effect most prominent in hep-th.Source: §2 · p. 4
• Because the end effect is less dramatic in the longer astro-ph announcements, the subsequent analysis focuses on hep-th and hep-ph.Source: §2 · p. 4
• In hep-th, median citation differences between L/2 and both 1-NE and L-NL are significant at the 1% level, distinguishing visibility and reverse-visibility contrasts.Source: §2 · pp. 5–6
• In hep-th, L-L submissions have 19.5 median citations versus 15 for 1-E, but the 4.5 difference is not significant (MWU P=0.1138).Source: §2 · p. 7
• In hep-ph, 1-NE, L-NL, and L-L each have 12 median citations, while 1-E submissions have 20.Source: §2 · p. 7

3 Geographic/Timezone effects

• The geographic classification uses submitter email domains as a rough location proxy, which can miss cross-continental collaborations or a submitter’s later relocation.The analysis also assigns .com and .org addresses to North America and uses a political Europe–Asia boundary.Source: §3 · p. 8
• The L/2 bin contains a majority of European submissions and a larger Asian share than the first- and last-position bins.The geographic distributions are based on the submitter’s email domain; North America and Asia are out of phase, while South America contributes under 10% in each bin.Source: Figure 4 · p. 8, §3 · p. 9
• Asian articles have aggregate median citations about 30% below the combined Europe–North America medians in astro-ph, hep-th, and hep-ph.The reported medians are 7, 7, and 6 for Asia, versus 10, 9, and 10 for EuNA, respectively.Source: §3 · p. 9, Table 1 · p. 9
• European and North American citation medians are generally not statistically different, except for hep-ph at L/2; the hep-th L/2 medians also do not differ significantly.The hep-ph medians are 8 for North America and 6 for Europe, significant by the Mann–Whitney U test but not the Kolmogorov–Smirnov test.Source: §3 · p. 10
• At L/2, Asian submissions have median citations similar to European and North American submissions, so their concentration there does not explain the lower middle-position median.For hep-th, the Asian median is 8, matching EuNA; for hep-ph, the Asian median is 7 versus 6 for EuNA, with no significant regional difference at L/2.Source: §3 · p. 10, Table 2 · p. 10
• Adding Asian submissions lowers the 1-NE median by one citation, but does not change the L/2 median; the 1-NE signal is weak and based on about 80 Asian submissions.The 1-NE difference is significant by only one of the two tests at the 5% level; comparisons at the 1-E and L-L positions are unavailable because each has fewer than 10 Asian submissions.Source: §3 · pp. 10–11

4 Discussion

• The alternative tests for reverse-visibility and procrastination had medians in the expected directions, but their samples were too small for statistically significant conclusions, even with a broader definition of middle positions.Source: §4 · p. 11
• The analyzed effects arose from unintentional properties of the announcement system during the studied period, though many positional effects may remain possible in the current system.Source: §4 · p. 11
• Randomizing list order could disserve readers who benefit from preferentially seeing self-promoted or procrastinated articles; the authors favor personalization using registered preferences and opt-in usage data.They suggest personalization could mitigate the global resonance of research communities viewing the same material in the same order.Source: §4 · p. 12
• Citation and readership differences by list position can be as large as a factor of two and are mostly independent of paper quality, complicating assessment of individuals and organizations.The authors use medians because heavy-tailed citation distributions make means sensitive to highly cited articles.Source: §4 · p. 12
• Further work could examine other arXiv disciplines, anonymized reader behavior, later citation-stabilized datasets, and additional short-term visibility effects.Source: §4 · p. 12