Author Career Stage and Role: Behavioral Evidence

Last updated: 2026-05-17

Synthesis

Empirical research on scholarly writing converges on a strong asymmetry by career stage: novice and expert academic writers do different things when they revise, and attend to different parts of the manuscript when they write. The foundational study (Sommers 1980, n=15) found that student writers treat revision as a lexical/rewording activity, while experienced writers treat it as a recursive, structural activity that reshapes meaning throughout the writing process [1]. Geisler (1994) extended this into academic contexts and identified the mechanism: experts construct a two-tier representation — a domain-content layer plus a separate rhetorical/authorial-positioning layer — while novices collapse the two and treat texts as transparent information transfer [2]. For a system that generates feedback on manuscripts, this means a “revise the introduction” instruction lands on these two populations as substantively different tasks.

The career-stage asymmetry is sharpest in the rhetorical positioning Swales (1990) labeled the CARS model (“Create A Research Space”): three rhetorical moves in research-article introductions — (1) establishing territory, (2) identifying a gap or niche, (3) occupying the niche [3]. Junior writers — grad students, postdocs, junior faculty — disproportionately invest in Move 1 (territory through citation density) and Move 3 (concrete contribution); senior writers disproportionately invest in Move 2 (gap construction and novelty framing), where the high-risk rhetorical work lives. This pattern aligns with the broader incentive structure: Foster, Rzhetsky & Evans (2015) found scientists skew strongly toward tradition over innovation in research-strategy choice [4], and Tripodi et al. (2025, n≈12,000 across 15 disciplines) found this skew is sharpest pre-tenure and relaxes post-tenure as researchers shift toward “more novel and higher-risk research” with declining citation impact [5].

Importantly, much doctoral-writing behavior that is read as skill deficit is better explained as identity work. Kamler & Thomson (2006, 2014) argue persuasively, and Casanave (2002), Aitchison & Lee (2006), and Paré (2011) corroborate from different angles, that PhD students’ defensiveness about findings, hedging into invisibility, methods-section overdevelopment, and resistance to feedback are explained more parsimoniously as identity protection in an unsafe rhetorical space than as writing-skill gaps [6, 7, 8, 9]. Paré frames the supervisor relationship as one where the supervisor “revises the writer” as much as the text [9]. This matters for persona design because it predicts that the same comment (“the framing is weak”) lands as a craft note for a senior PI and as a challenge-to-belong for a grad-student first author — and the simulated author’s feedback will therefore differ in priority assigned to acting on it, not only in interpretation.

Evidence and frameworks

The novice/expert revision asymmetry

Sommers (1980) is the foundational empirical study [1]. Across 8 freshman and 7 experienced adult writers with three compositions and multiple revision cycles plus interviews, the results split cleanly: students revise by changing words; experienced writers revise by reshaping arguments. Experienced writers also treat revision as recursive and embedded throughout drafting, not as a discrete post-drafting stage. The sample is small and from general composition (not scholarly writing), but the pattern has been widely cited as load-bearing and replicates in adjacent literatures.

Geisler (1994) extended the asymmetry into academic philosophy specifically, using rhetorical analysis, protocol analysis, and classroom-discourse analysis across undergraduates, advanced students, and professional academics reading the same texts [2]. Her finding — that experts maintain a two-tier representation (content + rhetorical/authorial-positioning layers) that novices have not yet constructed — provides the mechanism for the Sommers result. Novices literally do not “see” the rhetorical layer that senior writers are revising.

CARS, stance, and what gets attended to by stage

Swales’ (1990) CARS model [3] gives the rhetorical structure of research-article introductions: establish territory → identify gap → occupy niche. Hyland’s (1999, 2005) corpus work [10, 11] across eight disciplines documents how experts deploy stance (hedges, boosters, attitude markers, self-mention) and engagement (reader pronouns, directives, questions, asides) to calibrate community-specific norms. The verbatim framing in Hyland (2005) is that these features are how writers “project their textual voice” and “signal solidarity with readers” [11].

Empirically, novice writers — including doctoral and early-career authors — tend either to over-boost (assertive claims unsupported by hedging) or to over-hedge into evasion. Hyland’s (1999) study of 80 research articles plus interviews shows citation practice is similarly stratified: senior writers cite to position themselves rhetorically (aligning, distancing, gap-creating); novices cite to demonstrate they did the reading [10].

Identity work in doctoral writing

Kamler & Thomson (2006, 2014) [6] and the companion empirical critique Kamler & Thomson (2008) [7] are the most influential book-length treatment. Their core claim: doctoral writing is “identity formation” rather than skill acquisition, and the dissertation-advice book industry obscures this by reducing writing to skills and rules. The implication is that doctoral writers experience reviewer pushback as personal rejection, not as craft feedback, because they are still constructing the discursive identity that would let them receive the comment otherwise.

Casanave (2002) [12] provides longitudinal case studies across the entire enculturation arc — undergraduates, master’s, doctoral candidates, new faculty, bilingual faculty, seasoned professionals. Her metaphor of academic writing as a “game-like social and political as well as discoursal practice” stresses that the rules are enacted, not transparently followed. Aitchison & Lee (2006) [8] document the anxiety, isolation, and negative self-perception in the doctoral writing experience and the role of writing groups as communities of practice. Paré (2011) [9] argues supervisors “are writing teachers” [TODO verify exact wording] and that their work is relocating the student into the disciplinary community as a “rhetorical subject” capable of joining the conversation.

Junior-vs-senior career-stage incentives

Foster, Rzhetsky & Evans (2015) [4] analyzed MEDLINE biomedical abstracts at scale and found scientists’ research-problem choices skew strongly toward tradition over innovation — the expected additional reward of risky/novel strategies does not compensate for additional risk, except for prize-winners, where occasional gambles dominate. Tripodi et al. (2025, PNAS) [5] is the most recent and most directly relevant empirical finding: across ~12,000 researchers in 15 disciplines, publication rates rise sharply on the tenure track, peak just before tenure, and diverge afterward. Post-tenure, researchers shift toward more novel, higher-risk work with declining citation impact.

The synthesis: junior faculty are empirically “convergent, high-output, gap-filling” writers; post-tenure / senior PIs are empirically “more novel, more dispersed, lower-hit-rate” writers. This is a research-strategy finding, not directly a writing-behavior finding, but the rhetorical implications follow: junior faculty are pushed toward CARS Moves 1 and 3 (safe territory, concrete contribution), while senior PIs have the latitude to take Move 2 risks (bolder gap claims, more contested framing).

Camerer, Loewenstein & Weber (1989) [13] established the curse of knowledge experimentally in economic settings: better-informed agents systematically fail to ignore their private information when predicting less-informed agents’ judgments. The extrapolation to scholarly writing — that senior PIs over-elide methodological detail, under-explain motivation steps, and assume in-group jargon — is theoretically motivated and consistent with Geisler (1994) but not directly measured in academic writing contexts. Treat as a load-bearing pattern with strong indirect support, not a directly-replicated scholarly-writing finding.

Authorship-position behavior

Despite extensive literature on authorship credit (see credit-taxonomy-authorship), empirical work measuring how first authors vs middle authors vs last authors actually divide labor on a manuscript — time spent, sections revised, priorities held — is surprisingly thin. The discipline-by-discipline conventions of authorship order are documented (biomedicine: first = lead, last = senior PI; alphabetical in math/economics; hyperauthorship in particle physics — see credit-taxonomy-authorship). What is not well documented is the corresponding writing-behavior asymmetry. This is a genuine empirical gap, not a search failure, and any persona modeling first-author vs senior-author behavior must extrapolate from career-stage research (above) rather than from authorship-position-and-behavior research specifically.

Postdocs as a distinct stage

Casanave (2002) [12] explicitly treats the transition stages between doctoral student and new faculty, and shows the post-PhD posture differs from both. Career-development literature consistently describes the postdoc as a transitional independence-building stage where the writer must shift from supervised co-author to lead/corresponding author. The empirical writing-behavior literature on postdocs specifically is thinner than for doctoral students or tenured faculty. Any persona for a postdoc first author must extrapolate from doctoral and junior-faculty research, with explicit acknowledgment of the evidence gap.

How this informs scriptorium

This evidence base grounds the author-persona work proposed in #42, #43, #44 in three ways:

Personas should differ on attentional priority, not only voice. A grad-student first author persona that simply “writes like a grad student” misses the load-bearing finding: grad students attend to different parts of the manuscript than senior PIs, and revise at a different layer (local/lexical vs structural/rhetorical). The persona spec should encode revision-layer preference and CARS-move attention, not only register/voice.
Feedback uptake is asymmetric by stage. The same skill output (“the framing is weak”) will be evaluated very differently by a grad-student persona (identity-threatening, defended against) and a senior-PI persona (craft note, acted on). Persona-driven feedback loops must model the uptake differential, not just the interpretation differential — otherwise convergence across personas on a critique looks like signal when it’s actually flattening real differences.
Convergence-across-personas is signal precisely because incentives differ. Mirroring the diversity-of-attention argument in reviewer-archetypes-evidence: if a grad-student-first-author persona and a senior-PI persona both flag a revision suggestion as worth acting on, that’s a much stronger signal than either alone — because they’re filtering through different attentional priorities and different incentive structures. The Foster (2015) / Tripodi (2025) evidence is what licenses treating that convergence as load-bearing.
The postdoc persona is intentionally less evidence-grounded. When constructing the postdoc persona, acknowledge in the spec that it extrapolates from doctoral and junior-faculty findings rather than from direct empirical work, and flag what the persona may get wrong as a result.
Expert blind spot has implications for the senior-PI persona. A senior-PI persona should under-explain motivation, over-elide methodological detail, and assume in-group jargon — and should resist (or rationalize) suggestions to expand those — because that is the empirically supported expert pattern. The persona that catches “the methods are over-condensed” is doing real work that a more accommodating persona would miss.

Implementation priority for scriptorium

Verdict: Yes (knowledge layer for v0.2 persona docs and v0.3 personalization skills). No (not a skill itself).

Yes — grounding layer: This document is the load-bearing evidence base for the author-persona work tracked in #42 (concept docs), #43 (voice-profile skill), and #44 (persona-calibration skill). The persona specs that live outside the repo (so that skills cannot Goodhart against them) should cite this document for each role-differentiated behavior they model.

No — dedicated skill: Career-stage classification is not itself a useful skill output. The findings here shape how other skills calibrate, not what they produce. No author-stage-classifier skill is justified.

If Maybe later: if persona-calibration trend data (#44) shows that role-differentiated feedback is systematically improving manuscripts in a measurable way, promote the persona system from a testing tool to a production feature that authors opt into when they want “what would a senior PI say about my framing here?”

Open questions / weak evidence

Postdoc-specific writing behavior is empirically underdetermined. Most studies lump postdocs with PhD students or skip ahead to tenure-track. Casanave (2002) is the strongest source on the transition itself but specifically empirical studies of postdoc writing behavior are scarce. Personas for postdocs will need to extrapolate and acknowledge it.
Authorship-position-and-behavior research is largely conceptual. What first authors vs middle authors vs last authors actually do at the manuscript level is not well measured. Personas must extrapolate from career-stage research, with the caveat that “first author = grad student” is a biomedical-default that misfits other fields (see credit-taxonomy-authorship).
“Junior faculty are more risk-averse in writing” is theoretically motivated but only indirectly empirically supported. Foster (2015) and Tripodi (2025) establish the research strategy skew, and the publish-or-perish literature documents the incentive. The direct empirical link to writing/framing behavior specifically (not topic choice) is inferred, not measured. Persona design should treat this as a defensible-but-not-decisive pattern.
Sommers (1980) is small (n=15) and from general composition, not academic scholarly writing. Treat as a load-bearing pattern that downstream research supports, not as a single decisive study.
The “curse of knowledge in scholarly writing” extrapolation is plausible but not directly measured. Camerer et al. (1989) is experimental in economic settings; the extension to senior-PI revision blindness in academic writing rests on theory plus consistency with Geisler (1994). Treat similarly to the risk-aversion finding above.
Verbatim phrasing for several attributions is not pinned to a page-level citation — Paré’s “supervisors are writing teachers” formulation; Kamler & Thomson’s exact phrasing on “new ways with words”; Bazerman’s exact wording on socialization-into-scientific-activity. The substance of these is correct and supported by the references list, but exact quoted wording should be checked against the source page if scriptorium prose ever quotes them directly. Until then, these are treated as paraphrase-grade attributions, not quoted attributions.
Cultural/disciplinary scope is narrower than ideal. The foundational sources are Anglophone composition studies (Sommers), US academic philosophy (Geisler), and broadly Anglophone EAP (Swales, Hyland). Cross-cultural generalization beyond these contexts is plausible but not directly established. EAL scholar experience is treated separately in eal-academic-writing-evidence.

References

Sommers, N. (1980). Revision Strategies of Student Writers and Experienced Adult Writers. College Composition and Communication, 31(4), 378–388. DOI: 10.2307/356588.
Geisler, C. (1994). Academic Literacy and the Nature of Expertise: Reading, Writing, and Knowing in Academic Philosophy. Lawrence Erlbaum. ISBN: 978-0805810684.
Swales, J. M. (1990). Genre Analysis: English in Academic and Research Settings. Cambridge University Press. ISBN: 978-0521338134.
Foster, J. G., Rzhetsky, A., & Evans, J. A. (2015). Tradition and Innovation in Scientists’ Research Strategies. American Sociological Review, 80(5), 875–908. DOI: 10.1177/0003122415601618.
Tripodi, G., Zheng, J., Qian, Y., Murray, D., Jones, B. F., Ni, C., & Wang, D. (2025). Tenure and research trajectories. PNAS, 122(30), e2500322122. DOI: 10.1073/pnas.2500322122. PMID: 40694335.
Kamler, B., & Thomson, P. (2014). Helping Doctoral Students Write: Pedagogies for Supervision (2nd ed.). Routledge. ISBN: 978-0415823494. (1st edition: 2006.)
Kamler, B., & Thomson, P. (2008). The Failure of Dissertation Advice Books: Toward Alternative Pedagogies for Doctoral Writing. Educational Researcher, 37(8), 507–514. DOI: 10.3102/0013189X08327390.
Aitchison, C., & Lee, A. (2006). Research writing: problems and pedagogies. Teaching in Higher Education, 11(3), 265–278. DOI: 10.1080/13562510600680574.
Paré, A. (2011). Speaking of Writing: Supervisory Feedback and the Dissertation. In L. McAlpine & C. Amundsen (Eds.), Doctoral Education: Research-Based Strategies for Doctoral Students, Supervisors and Administrators. Springer. DOI: 10.1007/978-94-007-0507-4_4.
Hyland, K. (1999). Academic attribution: citation and the construction of disciplinary knowledge. Applied Linguistics, 20(3), 341–367. DOI: 10.1093/applin/20.3.341.
Hyland, K. (2005). Stance and engagement: a model of interaction in academic discourse. Discourse Studies, 7(2), 173–192. DOI: 10.1177/1461445605050365.
Casanave, C. P. (2002). Writing Games: Multicultural Case Studies of Academic Literacy Practices in Higher Education. Lawrence Erlbaum. ISBN: 978-0805835311.
Camerer, C., Loewenstein, G., & Weber, M. (1989). The Curse of Knowledge in Economic Settings: An Experimental Analysis. Journal of Political Economy, 97(5), 1232–1254. DOI: 10.1086/261651.
Bazerman, C. (1988). Shaping Written Knowledge: The Genre and Activity of the Experimental Article in Science. University of Wisconsin Press. ISBN: 978-0299116941. (Cited for theoretical framing of writing-as-discipline-membership.)