JMIR Publications

ABSTRACT

Background:

Mentorship is thought to shape how researchers adopt research practices, including those related to research integrity and open science, but large-scale evidence is scarce because mentoring relationships are rarely recorded. Bibliometric data could make such relationships visible at scale, provided the methods are validated and their limits understood.

Objective:

To test whether a bibliometric algorithm detects documented senior-junior research relationships, using formally recorded PhD supervision as the reference standard, and whether its classification differs across groups.

Methods:

We applied an algorithm built on Dimensions publication and grant data, which scores the strength (mentometer) and direction (seniority score) of co-authorship relationships, to 213 PhD candidate–supervisor pairs curated from theses defended in 2021-2022 at four Dutch university medical centres. We measured recall, examined which signals drive the score, compared classification by inferred gender, medical centre and supervisor seniority, and measured collaboration after the thesis.

Results:

The algorithm identified 212 of 213 pairs (99.5%); 187 (87.8%) were classified as very high likelihood. The score correlated with early-career co-authorship (ρ=0.71) and joint publications (ρ=0.43) more than with supervisor seniority (ρ=0.14). In 52.6% of pairs, candidate and supervisor published together after the thesis year, more often in very high pairs (56.5% vs 24.0%).

Conclusions:

The algorithm reliably detects formal supervisory relationships in biomedicine, and its score depends on the relationship rather than on the supervisor's seniority. It does not measure the quality of mentoring and cannot see mentoring that leaves no trace in publications or grants. Within these limits, it offers a tested basis for large-scale studies of how research relationships relate to research practices. Clinical Trial: N/A