Prominent but Less Productive: The Impact of Interdisciplinarity on Scientists' Research

PRODUCTION PENALTIES OF INTERDISCIPLINARY RESEARCH

• The section contrasts these production-side challenges with reception penalties: audiences tend to penalize offerings that are difficult to classify.Source: PRODUCTION PENALTIES OF… · p. 6
• The section locates production costs in the cognitive and collaborative challenges of developing mastery across domains, including technical and logistical hurdles.Disciplinary differences may also create epistemological or methodological conflict, raising coordination costs, tension, and role strain that can slow progress toward publication.Source: PRODUCTION PENALTIES OF… · p. 7
• Interdisciplinary papers may face a slower, more challenging review process because experts from different fields can disagree about their merits, extending the road to publication.The passage suggests these review-stage difficulties may carry earlier production delays through to the final product.Source: PRODUCTION PENALTIES OF… · pp. 7–8
• Interviews with successful interdisciplinary scholars offer initial evidence that this work takes extra time, commitment, and mental effort, delaying progress from conception to publication.Interviewees described moving between distinct intellectual demands and said that thinking through two disciplinary concerns could make producing a finished article take longer.Source: PRODUCTION PENALTIES OF… · p. 8
• The authors hypothesize that greater engagement in interdisciplinary research is associated with lower research-article productivity.They argue that the time and effort required to address two disciplinary concerns can leave scientists feeling less productive.Source: PRODUCTION PENALTIES OF… · p. 8

RECEPTION BENEFITS OF INTERDISCIPLINARY RESEARCH

• IDR is hypothesized to have higher citation-based visibility, even after controlling for the larger prospective audience created by targeting multiple fields.The authors also expect no reception penalty, defined as being received poorly in the academic marketplace.Source: RECEPTION BENEFITS OF… · pp. 8, 10
• IDR is hypothesized to have greater variability in visibility, making it a high-risk, high-reward activity that can yield prominence or be ignored.The authors expect IDR to be well received on average, while also increasing the probability of breakthroughs and failures.Source: RECEPTION BENEFITS OF… · p. 10
• The authors distinguish spanning categories from spanning distant categories, hypothesizing that a measure of IDR that omits distance will have weaker effects on productivity and visibility.They expect both the challenges and the value of spanning categories to increase with distance between domains.Source: RECEPTION BENEFITS OF… · pp. 10–11

• Survey data show that 36% of life-sciences dissertations self-classify as interdisciplinary, compared with 26% in engineering and 21% in math.Source: INTELLECTUAL CONTEXT and IDR · p. 11
• Using a cognitive-dissimilarity measure from 0 to 1, prior research reports mean interdisciplinarity of 0.654 for Biotechnology and 0.664 for Medicine, versus 0.53 for Electrical Engineering (Porter and Rafols 2009; Porter et al. 2007).Source: INTELLECTUAL CONTEXT and IDR · p. 11
• Because actors tend to value familiar work, IDR is expected to gain more visibility in highly interdisciplinary fields, where it is more customary and appreciated.Source: INTELLECTUAL CONTEXT and IDR · pp. 11–12
• Researchers in highly interdisciplinary fields may face fewer production challenges because they have more IDR models, potentially better training, and more amenable, productive review processes.Source: INTELLECTUAL CONTEXT and IDR · p. 12
• H5a predicts a stronger positive association between IDR and visibility in highly interdisciplinary fields, while H5b predicts a weaker negative association between IDR and productivity.Source: INTELLECTUAL CONTEXT and IDR · p. 12
• The paper expects IDR’s benefits to emerge mainly as increased citations, while its penalties arise during production as greater cognitive and communicative challenges and reduced productivity.Source: INTELLECTUAL CONTEXT and IDR · p. 12

Sample

• The scientists were affiliated with 52 NSF-funded industry/university-cooperative research centers housed at universities and conducting research of interest to, and partly supported by, industry.The centers spanned areas including biosurface physics, civil and environmental engineering, and architectural science.Source: Sample · p. 12
• The analysis covers 854 PhD-level scientists with publication records as of 2005, selected from IUCRC members spanning doctoral students, post-docs, and faculty at all career stages.The broader member information was collected as of 2003; scientists without publications were excluded because IDR, productivity, and visibility could not be computed for them.Source: Sample · p. 13
• The archival dataset includes approximately 32,000 center-affiliate articles and their references, whose Web of Science subject categories were used to measure interdisciplinarity.Article information included authors, institutions, journal, and citations; the main analyses used archival data, while supplementary analyses also drew on a survey, journal turnaround times, and working papers for some authors.Source: Sample · p. 13
• Supplementary analyses used a survey of center-based scientists, public journal-turnaround data, and unpublished working papers for a subset of authors.The main analyses relied on archival data.Source: Sample · p. 13

Productivity

• Productivity is the total number of WoS-journal articles published from each scholar’s first publication through 2005, excluding book chapters and articles in less internationally recognized journals.The authors characterize this as a conservative measure.Source: Dependent Variables · p. 13
• Article counts are standard in scientific-productivity research, and the authors assess robustness using co-author-weighted publication counts.Article counts are highly correlated with co-author-weighted counts, including as collaboration increases.Source: Dependent Variables · p. 13
• JIF-weighted productivity is excluded because it conflates quantity and quality, distinct outcomes hypothesized to respond differently to IDR; article counts are analyzed by person and person-year.Source: Dependent Variables · p. 13

Visibility

• Visibility is measured as forward citations accrued by each WoS-indexed article through 2010, with journal impact factor controlled in all models.Source: Dependent Variables · p. 14
• Selected analyses use mean person- and person-year citation visibility; skewed aggregates are log-transformed or modeled with negative binomial counts, and H3 citation variability is measured by citation standard deviation.Source: Dependent Variables · p. 14

Independent Variables

• Porter and colleagues’ paper-level IDR measure integrates the variety, balance, and similarity of subject categories pooled from a paper’s references.Similarity is based on the cognitive distance between categories; the measure is a parameterization of the Sterling Index.Source: Independent Variables · p. 14
• For person-year analyses, the authors average papers’ IDR scores and exclude zero-score observations because they occur in low-productivity years.Zero scores also arise when a single paper references only one subject category; the authors say these years obscure the hypothesized IDR-related production penalty.Source: Independent Variables · pp. 14–15
• Including zero scores leaves visibility analyses robust but makes productivity effects curvilinear, since zero and high IDR scores are both associated with low productivity.The authors attribute the zero scores to low-productivity years and state that high IDR scores are also associated with low productivity, as hypothesized.Source: Independent Variables · p. 15
• To test whether distance matters beyond variety (H4), the authors compare Porter’s measure with the total number of unique subject categories in scholars’ pooled references.The IDR score is only mildly related to reference count (r=.15), and controlling for reference count leaves reported results unchanged.Source: Independent Variables · pp. 15–16
• r=0.18, the IDR score correlates positively with the subject categories assigned to the focal paper; Science, Nature, and PNAS average 0.71, 0.71, and 0.70, respectively, versus 0.62 in the sample.The three journal means are more than half a standard deviation above the sample average.Source: Independent Variables · p. 16
• Field-level IDR is measured continuously as the average IDR score across the nine broad fields in the sample, and categorically by life sciences (highest) and electrical engineering (one of the lowest).Source: Independent Variables · p. 17

Control Variables

• Person-level controls account for gender, professional age, and status, with gender assessed from names and website information and professional age measured from PhD year to 2005.The papers are nested within persons.Source: Control Variables · p. 17
• Institutional status is proxied by faculty membership in the National Academy of Sciences, while PhD-granting institution rank serves as a rough measure of scholar quality.The latter is reverse-coded so higher values indicate higher quality and helps address differences in scholars’ likelihood of engaging in IDR.Source: Control Variables · pp. 17–18
• Journal and field controls include scholars’ mean journal impact factor, field journal turnaround time for productivity models, and average field citations per paper for visibility models.Source: Control Variables · p. 18
• To account for audience size and paper reach, visibility analyses control for field size and the number of subject categories classifying the focal paper.The authors treat papers classified in multiple fields as potentially reaching a larger body of scholars.Source: Control Variables · p. 18
• Paper-level and person-year analyses account for team size and whether collaborators had published together before.Yearly analyses use average author counts and the proportion of papers with repeat collaborators.Source: Control Variables · p. 18
• The analyses also control for lagged cumulative publication experience and citations, plus publication year to account for older papers having more time to accrue citations.Source: Control Variables · pp. 18–19

Statistical Approach

• Path analysis models productivity and visibility together, estimating their interrelationship and both direct and indirect effects of interdisciplinarity.The person-level analysis covers 854 scientists.Source: Statistical Approach · p. 19
• Structural equation modeling uses full information maximum likelihood to include available data and avoid listwise deletion or imputation.The authors also report that results are the same without multiple imputation for variables with missing data.Source: Statistical Approach · pp. 19–20
• Fixed-effects regressions model productivity at the person-year level (n=8,779) and visibility at the paper level (n=29,782), using within-person variation and controlling for unmeasured person-level characteristics.Source: Statistical Approach · pp. 19–20
• For field-level IDR, the study uses random-effects models for visibility and population-averaged negative binomial models for productivity because field characteristics are time-invariant.Fixed-effects models would drop field IDR and other time-invariant field characteristics.Source: Statistical Approach · p. 20

Productivity results

• A 10% increase in IDR predicts 7.7% lower career productivity (coefficient -0.77); the penalty persists under coauthor weighting (b=-0.99**, SE=0.39) and person-year fixed effects (b=-0.143**).The person-year result compares years in which scholars do more interdisciplinary work with their other years.Source: RESULTS · p. 21

Visibility results

• A 10% increase in IDR suggests 6.9% higher average citations (+0.69***); IDR has a positive, significant visibility effect in paper-level models with person and year fixed effects, also holding after group authors are excluded.Source: RESULTS · p. 21

Variation in citations

• Citation variability was higher for scientists’ high-IDR than low-IDR papers (mean=30.4 vs mean=24.2; p<.0015; n=647).The comparison uses scientists with at least two papers of each type.Source: RESULTS · p. 22

Cognitive distance

• Removing cognitive distance from the IDR measure preserved a positive citation association but removed the productivity penalty; simply drawing on more subject categories did not hinder productivity.The authors infer that producing and publishing work spanning unrelated fields is more difficult than work spanning related fields.Source: RESULTS · p. 23

Field differences

• IDR’s productivity penalty was weaker in highly interdisciplinary life sciences than in less interdisciplinary electrical engineering, where it was not statistically significant; field differences in visibility were unsupported, possibly due to low power.The comparison supports H5b but not H5a.Source: RESULTS · p. 23

Field-level interaction

• At the person-year level, field-level IDR and its interaction with individual IDR positively affected productivity and visibility, strengthening IDR’s citation benefit and lessening its productivity penalty.The authors suggest that more receptive fields may help scholars manage IDR’s cognitive challenges or may be more amenable to IDR in peer review.Source: RESULTS · pp. 23–24

Control variables

• The reported effects held with controls for gender, professional age, and status, and women were not more likely than men to engage in IDR, contrary to perceptions and prior research.Source: RESULTS · p. 24

RESULTS

• Using NAS membership at a scientist’s institution as the status measure, IDR follows an inverted U: scientists at low- and high-status universities engage in IDR, while those at middle-status universities conform.The authors interpret the middle-status position as precarious, with the prospect of recognition and the threat of delegitimation encouraging disciplinary conformity.Source: RESULTS · p. 24
• Controls for field size, field citation averages, journal turnaround time, and lagged productivity or visibility do not alter or erase the main findings.These controls have intuitive and often significant effects, but the main findings remain.Source: RESULTS · p. 24

Cognitive and Collaborative Challenges

• b=1.51*, se=.077: working repeatedly with a similar set of collaborators reduces the IDR productivity penalty.This pattern is consistent with collaborators learning to work together over time.Source: Drivers of the Productivity Penalty · p. 25
• IDR production involved marginally significant challenges: communication was less clear (p=.065), more difficult (p=.083), and of lower quality (p=.109), while teams had more difficulty generating ideas (p=.103).Source: Drivers of the Productivity Penalty · p. 25
• Under an alternative multidisciplinarity measure, scientists conducting multidisciplinary research were more productive (b=0.58*, SE=0.31); only scientists publishing IDR were less productive.The measure pools subject categories across a scientist’s publication record, rather than first calculating IDR at the paper level.Source: Drivers of the Productivity Penalty · pp. 25–26

Challenging Review Process

• IDR did not significantly correlate with or predict review time; eventually published working papers were more interdisciplinary than unpublished ones (p=.054), offering little evidence of review-process hindrance.The working-paper comparison is difficult to interpret because those papers were more recent and IDR was trending upward; the authors conclude IDR work is not more likely to be file-drawered or rejected.Source: Drivers of the Productivity Penalty · pp. 26–27

Cognitive and Collaborative Challenges

• IDR’s largest hurdle is the first-stage work of planning, coordinating, and conducting research; review-stage file-drawering or rejection is not more likely.The authors note that these analyses are associations, not causal tests.Source: Drivers of the Productivity Penalty · p. 27

Robustness Checks and Selection Issues

• IDR’s visibility-positive and productivity-negative main effects remain significant around center founding, while interactions with center affiliation are not significant.The analysis uses fixed-effects models of papers published in the five years before and after center founding; it suggests center affiliation neither increases IDR’s visibility benefit nor reduces its cost.Source: Reskin 2003) · pp. 27–28
• Compared with scientists in related fields, the center-based sample is older, more predominantly male, and more productive, but the authors report that these differences do not alter their results.The authors acknowledge that their research design cannot completely rule out distinctive features of center-based scientists.Source: Reskin 2003) · pp. 28–29
• The results hold among younger scientists and early-career papers, and the IDR-by-age interaction is negative and significant but very small.The person-level analysis restricts the sample to scientists no more than 13 years post-PhD, and the paper-level analysis to scientists’ first five career years; the age interaction suggests the penalty may be stronger for older scientists.Source: Reskin 2003) · p. 28
• The IDR results hold for both men and women, and separate survey evidence also finds fewer publications among scholars whose training and work fields differ.The paper additionally notes prior evidence that Stanford faculty productivity declines with a joint appointment.Source: Reskin 2003) · p. 28
• In the 1980s and 1990s, the sample’s papers are almost identical to Web of Science papers in conventionality and slightly lower in novelty, countering the view that center scholars are exceptional innovators.The authors say these comparisons suggest the results are likely to apply more broadly to academic science.Source: Reskin 2003) · p. 29

Quality of IDR papers

• Quality checks support an independent IDR effect on visibility rather than an explanation based on scholars’ unmeasured quality or propensity to engage in IDR.Person fixed effects address individual differences; where those effects are unavailable, the analyses control for the quality of scientists’ PhD and current institutions.Source: Reskin 2003) · pp. 29–30
• IDR’s positive citation effect holds after controlling for journal impact and within low-impact journals, while journal-level turnaround times do not differ by IDR level.The authors take the citation and review-process findings as evidence against IDR papers being inherently higher or lower quality.Source: Reskin 2003) · p. 30

DISCUSSION

• IDR’s direct productivity effect (-0.06) outweighs its reception effect (+0.04): 80th- versus 20th-percentile scholars publish 24 vs. 26 articles and receive 297 vs. 253 citations; total citation effect is -0.28.Source: DISCUSSION · p. 31
• The paper shifts attention from reception penalties to reduced productivity, a distinct production penalty attributed to communication and coordination hurdles rather than review.Source: DISCUSSION · pp. 31–32
• The productivity penalty is stronger in low-IDR fields than high-IDR fields, suggesting field-specific cognitive and collaborative challenges may be mitigated by training or IDR-friendly review.Electrical engineering exemplifies a low-IDR field and life sciences a high-IDR field.Source: DISCUSSION · p. 32
• When IDR reflects variety without cognitive distance, citations rise but the productivity penalty disappears; integrating distant fields within a paper links citation gains with fewer publications.Source: DISCUSSION · p. 33
• Interdisciplinary scientists avoid the reception penalties found among market-takers and gain visibility, but IDR is high-risk, high-reward, with more overall citations and greater citation variation.Source: DISCUSSION · p. 33
• Preliminary analyses suggest high-IDR papers tend to be cited by interdisciplinary papers, but audience-focused research and broader samples remain important extensions.The authors characterize this preliminary pattern as “breadth breeds further breadth” and call for further exploration.Source: DISCUSSION · p. 34
• Although the 850 scientists span fields and university settings, all are affiliated with university research centers; the authors propose prospective career tracking as a complement to this study.Their analyses suggest center affiliation does not drive the reported results.Source: DISCUSSION · p. 34
• IDR scholars produce useful, noteworthy research with greater impact on the scientific community and may publish in top-tier journals (IDR–journal impact factor correlation: 0.25), but their overall productivity is hampered.The authors describe a clear production penalty associated with IDR work.Source: DISCUSSION · pp. 34–35
• The analysis of working papers suggests IDR does not hamper—and may help—subsequent publication, but larger-sample analysis is needed because its penalties may extend further than documented.Source: DISCUSSION · p. 35
• The study does not assess tenure or whether increased citations offset IDR-related productivity losses, so career effects—especially for early-career scientists—remain uncertain.The authors call for more focused study of individual scientists’ career outcomes.Source: DISCUSSION · p. 35
• Interdisciplinary scholars face burdens mastering multiple domains, integrating them into one work, and coordinating across coauthors, although fellow scientists appear to receive their projects well.Source: DISCUSSION · p. 35
• The authors argue that IDR’s societal benefits in useful, valuable science seem clear despite substantial individual costs, and that scholarship evaluation should be reconsidered to encourage continued IDR.Source: DISCUSSION · p. 35

Appendix A1. Construction of the Interdisciplinary Research (IDR) Measure for 3 Hypothetical Articles

• Porter’s Integration measure combines disciplinary variety, balance, and similarity into an IDR score from 0 to 1, with higher scores for references spanning more relatively unrelated disciplines.Disciplines are proxied by 244 Web of Science Subject Categories; similarity is derived from a normalized cosine matrix based on all Web of Science-indexed articles.Source: Appendix A1. Construction of the… · p. 49
• The examples contrast a low-IDR paper citing two highly related categories (Sij=0.35), a high-IDR paper citing all four categories including distant ones (Sij=0.0001), and a paper citing two distant categories (Sij=0.0011) with a higher score than Article 2.Source: Appendix A1. Construction of the… · p. 49
• The three displayed IDR calculations are 0.6854, 0.2639, and 0.3198.They are shown as 1 - ∑ij Sij * Pi * Pj.Source: Appendix A1. Construction of the… · p. 49

Appendix A3. Robustness Checks: Regressions of Productivity and Visibility on IDRa

• IDR coefficients are 0.276** for Model 1 Visibility, -0.312* for Model 2 Productivity, and 0.485*** for Model 3 Visibility.The models use fixed effects; the reported significance markers denote ** p<.05 and * p<.10, while *** denotes p<.01.Source: Appendix A3. Robustness Checks:… · p. 51
• Pre/Post Center Formation (Post=1) coefficients reported in the table are -0.221** and 0.121.Source: Appendix A3. Robustness Checks:… · p. 51
• The table reports Post-Center*IDR coefficients of 0.257 and 0.119.Source: Appendix A3. Robustness Checks:… · p. 51
• The robustness models measure visibility at the paper level and productivity at the person-year level with person fixed effects, using publications from five years before and after center formation.Source: Appendix A3. Robustness Checks:… · p. 51