Do the workshops change academic careers?
An observational study of 187 international academic-writing workshops and a resolved roster of 634 participants. It links the programme record to open publication data to ask what changed around first participation—and what the public record cannot show.
Partner
Humanities & social sciences academy
Discipline
Impact Analysis
Scale
International
Year
2025
Who can be measured
The private programme record supplies the roster, workshop dates, shared attendance and country-level participant geography. Only anonymous totals are published here.
ORCID is a researcher-controlled identity record. It helps distinguish people with similar names; its identifiers are used for linkage and withheld from the public release.
OpenAlex is an open index of scholarly works and authors. It supplies the publication histories, co-authorships, recorded affiliations, topics and present-day citation metadata used in this analysis.
Crossref metadata helps confirm DOI-backed works and group versions of the same output into one publication family. DOI identifiers are used during analysis and withheld from the public release.
The observability funnel
An accepted OpenAlex identity is a canonical author link retained after review of names, affiliations, dates and corroborating public records. ORCID-linked records without a sufficiently complete OpenAlex history remain partial; unresolved records have no publication source accepted at the project’s confidence threshold. Neither category means that the person had no publications. The main comparison requires observations for years −3 to +3 around first participation, ending by 2025. 44 roster members meet that definition; 36 accepted identities do not yet have the full follow-up window.
Workshop roster
Accepted OpenAlex identity
Complete −3…+3 window
Sources: Writing-workshop programme corpus, OpenAlex, ORCID
The trajectories
The trajectory fan
Three paired tests summarise the same before-and-after differences in different ways. The Wilcoxon signed-rank test, which uses the direction and ranked size of each person’s change, gives p = 0.029. The paired t-test, which tests the mean change, gives p = 0.055. The sign test, which considers direction alone, gives p = 0.117. The evidence of decline therefore depends on how the change is summarised.
Each line joins one participant’s average annual number of publication families in years −3…−1 to the corresponding average in years +1…+3. To keep the fan legible, only trajectories changing by more than five publication families per year are drawn; the summary covers all 44. The raw mean change is −1.11 publication families per person per year, with a bootstrap 95% interval from −2.24 to −0.02; the median change is −0.67 publication families per person per year. This describes a decline in the observed cohort. It does not attribute that decline to the workshops.
Sources: Writing-workshop programme corpus, OpenAlex, ORCID, Crossref
Observed output versus adjusted event-time coefficients
The lower panel compares each event year with year −1 after accounting for stable differences between people and conditions shared within each calendar year. The model uses 308 person-years—one person observed in one year. The pre-trend test gives p = 0.185, while the combined test for years +1…+3 gives p = 0.963. Every post-workshop interval crosses zero: no common workshop-timed shift is isolated, and the intervals stay wide enough to include meaningful change in either direction.
Observed mean output above; adjusted differences from year −1 below, each with a 95% confidence interval.
Sources: Writing-workshop programme corpus, OpenAlex, ORCID, Crossref
Estimates against zero
The raw row compares all 44 records before and after first participation. The cohort rows repeat that descriptive comparison by first-workshop year. The DOI-backed row tests whether counting only DOI-backed publication families changes the picture. The matched row compares 11 participants in one eligible cohort with similar participants whose workshops came later; it is exploratory and cannot represent the programme as a whole. The main adjusted estimate is −0.06 publication families per person per year, with a 95% interval from −2.49 to +2.37. The point estimate is close to zero; the interval is too wide to conclude that the average change was small.
Observed before / after
44 people
−1.11
DOI-backed outputs only
44 people
−0.50
Exploratory matched comparison
11 people
−2.39
Cohort 2020
11 people
−1.55
Cohort 2021
19 people
−0.35
Cohort 2022
14 people
−1.81
Adjusted average
44 people
−0.06
change in publication families per person per year
Each square is an estimated annual change; each horizontal line is its 95% confidence interval. The rows answer different questions and should not be read as repeated estimates of one effect.
Sources: Writing-workshop programme corpus, OpenAlex, ORCID, Crossref
Other signals in the publication record
Citation visibility is the present-day OpenAlex citation count divided by the age of a publication family; it is not a historical citation series. New co-authors, recorded publication affiliations and OpenAlex topics are also incomplete traces. The publication record sees some changes more readily than others, and none of these measures is a complete account of an academic career.
Citation visibility
p-value 0.005
−3.05
New co-authors
p-value 0.357
−1.06
New recorded affiliations
p-value 0.304
−0.50
New topics
p-value 0.060
−1.06
raw mean change per person per year, years +1…+3 vs −3…−1
These rows show raw mean changes in four publication-derived signals across the 44 records. Every bootstrap interval crosses or touches zero, so the size and average direction of each change remain uncertain. For citation visibility, the Wilcoxon signed-rank test gives p = 0.005, indicating that within-person changes tend downward when their ranked sizes are considered, even though the mean decline is imprecisely estimated. These are secondary traces in the publication record, not evidence of workshop effects.
Sources: Writing-workshop programme corpus, OpenAlex, ORCID
Workshop peers in the publication record
The peer-pair field
All workshop peer pairs · 1,138
accepted identities for both people · 18 (1.6%)
The 18 publication-linked pairs — magnified
Within the 18 publication-linked pairs, 2 first appear as co-authors after their shared workshop: 11.1%, with a Wilson 95% interval of 3.1%–32.8%. The remaining 98.4% of workshop-peer pairs cannot be assessed using this measure because the pair does not have accepted OpenAlex identities at both ends. That is a limit of the record, not evidence that no relationship existed.
Sources: Writing-workshop programme corpus, OpenAlex, ORCID
Two anonymous collaboration sequences
Participant A × Participant B
Two workshop peers with accepted publication identities. No co-authorship appears before their shared 2022 workshop in the available record.
- 2
- years to first co-authored publication family
- 3
- co-authored publication families observed
- 2024–25
- publication years
Participant A × Participant B
Two workshop peers with accepted publication identities. No co-authorship appears before their shared 2023 workshop in the available record.
- 1
- year to first co-authored publication family
- 1
- co-authored publication families observed
- 2024
- publication year
One pair shared a workshop in 2022, first appears as co-authors two years later, and has three observed joint publication families in 2024–25. The other shared a workshop in 2023 and has one observed joint publication family in 2024. No earlier co-authorship appears for either pair in the available record. The sequence is consistent with a relationship emerging after the workshop, but publication lead times and other routes to collaboration mean the record cannot explain why it formed.
Sources: Writing-workshop programme corpus, OpenAlex, ORCID
Findings, limits and next directions
Shared-workshop geography
The lines join country pairs represented within the same workshops; they do not show later co-authorship. Larger circles indicate more participants, wider lines more shared workshops, and ringed markers locate territories too small to draw at world scale. No person-level geography is published.
Sources: Writing-workshop programme corpus, Natural Earth
What the analysis shows
Coverage is the first finding. The resolved roster contains 634 people across 187 workshops. 80 have accepted OpenAlex identities; 44 have the complete seven-year window used for the main comparison. An unresolved identity or incomplete window means that the publication record is insufficient for this analysis—not that the person had no publications.
The 44 publication trajectories do not move as one. 15 rise, 26 fall and 3 are unchanged. Their raw mean annual publication-family count falls by 1.11 publication families per person per year, while the person- and calendar-year-adjusted estimate is −0.06, with a 95% interval from −2.49 to +2.37 publication families per person per year. No common workshop-timed shift is detected, and the interval remains too wide to determine whether the average change was small.
The collaboration evidence narrows again. Only 18 of 1,138 workshop-peer pairs can be assessed at all. 2 first appear as co-authors after their shared workshop. That timing is visible; its cause is not. The selective denominator cannot support a programme-wide collaboration rate.
The workshops connected places. Resolved participant records span 60 countries and territories and form 133 shared-workshop country-pair links. This is a geography of participation, not evidence of later cross-country collaboration.
Where this analysis stops
The public record is not the career. It sees publications more readily than confidence, mentoring, rejected manuscripts, informal exchange or collaborations that never become co-authored outputs. Same-year publications may have begun before the workshop; recorded affiliations and topics are incomplete proxies; and OpenAlex can miss works or split and merge identities. Recent participants are also observed for less time.
This remains an observational study. Fixed effects account for stable differences between people and conditions shared within calendar years, but they cannot remove every changing influence or the selection that shapes who participates and whose publications are visible. The matched comparison is exploratory and limited to one eligible cohort. The limit is clear: no common adjusted shift was detected, but absence of detection is not proof that every participant experienced no effect or that the average effect was exactly zero.
What evidence should come next
A stronger answer would begin with planned follow-up for the whole roster: consented identity linkage, a longer common observation window, outputs beyond indexed publications, and direct measures of writing confidence, submissions, acceptances, mentoring, career progression and collaboration. Interviews could reveal mechanisms that bibliographic traces cannot see.
Estimating causation would also require a credible comparison with what might have happened without participation: ideally randomised or phased access, or a pre-specified matched group with strong baseline and follow-up data. That design would need explicit checks for pre-existing trends, attrition and unequal publication visibility. Repeated releases could then distinguish short- from long-term change. Until then, the defensible conclusion is narrower: varied publication trajectories; no detected common adjusted shift; two post-workshop co-authorship sequences among the 18 assessable pairs; and broad participation geography. The record leaves open changes it was not built to observe.
What this work demonstrates
The study resolves a roster of 634 participants across ORCID, OpenAlex and Crossref, groups versions of the same output into publication families, and compares the three years before and after first participation. Identities that could not be matched with confidence stay outside the comparison rather than being counted as absent.
Where this approach could be applied
The approach could help funders and programme teams reconstruct participation, reach and institutional relationships across comparable initiatives, and establish what an open publication record can and cannot show about a programme.
Continue the investigation