Industry contribution to scientific publishing

Industry firms regularly conduct and publish scientific research for multiple strategic reasons [9], including accessing external knowledge and resources, attracting and retaining scientific talent [10,11,29], building absorptive capacity [11,30–32], supporting downstream innovation [12,31,33], enhancing firm reputation, and gaining legitimacy and trust from stakeholders [10,31]. At the same time, scientific publications are a public good that is broadly accessible beyond firm boundaries [34], potentially reducing appropriability and creating a tension between the benefits of openness and the need to capture commercial value from knowledge [9,18,31].

Since this openness-appropriation trade-off may vary across scientific fields, industry’s share of publishing can differ substantially from field to field. Industry publishing is particularly prominent in Pasteur’s Quadrant fields, such as Biotechnology, Computer Science, and Artificial Intelligence, where advances in scientific research are tightly coupled with technological innovation [35]. Therefore, industry firms’ engagement with the scientific frontier in Pasteur’s Quadrant fields can directly support the firms’ downstream innovation [31]. Although firms may invest in basic research to build absorptive capacity [31], firms largely tend to publish more in applied fields than in basic fields, where scientific output often has more apparent near-term relevance for problem solving and technological innovation [17,18,36,37]. Since applied research is closer to commercialization, firms can more readily capture private returns through commercialization and complementary downstream capabilities, making openness less expensive. In contrast, in basic fields, where advances are farther from application, scientific research in those fields may generate weaker immediate private returns relative to the cost of openness. Firms may also publish in fields associated with controversial products or technologies (e.g., genetically modified crops in Biotechnology, deepfakes, and facial recognition in Artificial Intelligence) to demonstrate responsible conduct and strengthen trust among external stakeholders [13]. In such fields, commercialization is not hindered because firms can maintain a commercial advantage through IP protection and downstream capabilities, and publishing disclosures may relate to foundational, methodological, or normative components rather than the most commercially sensitive components of their innovations [13]. Additionally, firms are especially likely to publish in fields that generate theoretical foundations for how technological components function and interact, helping firms make sense of uncertain technologies and guide subsequent problem solving and recombination [23,38]. For instance, research by firms on the structure of cell receptors helped scientists design molecules that could bind to the receptor sites and activate or inhibit the functioning of specific body cells, and led to the development of the drug Prozac, which selectively inhibits the action of specific brain cell receptors, which helps treat anxiety and depression [39].

Measuring novelty

Despite the practical relevance of understanding the drivers of novel science, no single novelty measurement approach or indicator captures all its dimensions [40], and relatively few studies have systematically examined the multifaceted nature of recombinant novelty [40–43]. Existing approaches of measuring novelty can broadly be grouped into four categories, each reflecting distinct aspects of the nature and magnitude of novelty. Categorical measures operationalize novelty in binary terms, indicating whether a previously unseen knowledge component or combination is present or absent [44,45], configurational approaches capture the degree of distance or divergence in features [46,47], structural measures adopt a network perspective, emphasizing the position of recombined knowledge in the broader knowledge network [45,48], and combinatorial measures operationalize novelty as arising from rare combinations [2,3,45,49,50].

We follow the combinatorial perspective of novelty, which views novelty as unconventional recombinations of prior knowledge elements [2,3,49]. In the context of scientific publications, novelty has primarily been operationalized by examining the distribution and co-occurrence of patterns of referenced journals in the bibliography of publications [2,3]. Some publications incorporate mildly atypical combinations of journals that, while novel, remain adjacent to dominant knowledge streams. In contrast, others integrate highly disparate or rarely connected domains, indicating a deeper level of conceptual innovation [2,3,51]. Novel recombinations can also vary in scope, ranging from broad integrations that span multiple ideas or disciplines to narrow conceptual shifts [2]. These differences in the conceptual distance and scope of recombined knowledge streams align with Guilford’s conceptualisation [52] of the creative production processes of originality (the statistical rarity of responses) and fluency (total responses generated). These patterns of conceptual distances and the scope of recombined knowledge streams provide a basis for quantifying the degree of novelty in scientific work.

However, atypical recombinations require cognitive leaps and significant departures from established knowledge [2,3,49]. As a result, most publications make incremental advances, and only a small proportion exhibit novel recombinations [2,3]. Therefore, the occurrence of novel recombinations is rare and may confound the patterns for the degree of novelty. This rarity of novel recombinations is especially salient in our field-level analysis, where aggregate patterns may mask whether the observed effects are driven by the frequency of novelty occurrences or by the degree of novelty of publications within fields. Therefore, in this work, we examine the occurrence of novelty in addition to the degree of novelty.

Novelty occurrence: Novelty occurrence of a publication captures whether a publication involves an unexpected recombination of knowledge and indicates a new cognitive leap, and is consistent with the categorical operationalization of novelty [44,45].

Degree of Novelty: Consistent with Guilford’s (1976) [52] view, we characterize the degree of novelty along two dimensions: novelty breadth, which captures the scope of knowledge recombinations, and novelty distance, which captures the extremities of cognitive distance spanned by those recombinations.

Novelty breadth refers to the overall scope of novel recombinations within a publication. It reflects the extent to which diverse and previously unconnected knowledge components are brought together. We operationalize this by summing the intellectual distances of all novel knowledge combinations, thereby capturing the cumulative novelty embedded in the recombination process. Consequently, higher novelty breadth reflects a publication’s engagement with a wide range of conceptually diverse domains through novel recombinations.

Novelty distance, in contrast, captures the extremity of conceptual divergence among the recombined knowledge elements, following Koestler’s conceptualisation of creative bisociation as the intersection of separate thought matrices [53]. Specifically, it measures the intellectual distance between the two most conceptually distant knowledge components within a publication’s set of novel recombinations. This dimension reflects the difficulty of integrating the most disparate ideas, highlighting a publication’s capacity for radical knowledge integration.

Both aspects of novelty, novelty occurrence and the degree of novelty (novelty breadth and novelty distance), are distinct dimensions and are linked to knowledge breakthroughs. Novelty occurrence indicates a new cognitive leap and is linked to breakthrough publications [2]. Many transformative scientific advances have emerged from a broad scope of novel recombinations, i.e., high novelty breadth [2]. A representative example is the study of the self-organization of Microtubules, which integrates concepts from Physics, Chemistry, and Biology to explain cellular formation processes [54]. This kind of interdisciplinary recombination demonstrates how a broad scope of recombinations can advance scientific understanding by connecting several disparate ideas. Several groundbreaking scientific breakthroughs have also emerged by recombining conceptually distant ideas—i.e., high novelty distance—often originating from curiosity-driven inquiry. For instance, Chaos Theory emerged from Edward Lorenz’s curiosity-driven exploration of nonlinear equations in the context of weather prediction, leading to profound implications across disciplines [55]. Similarly, Prospect Theory was developed through curiosity-driven research by Daniel Kahneman and Amos Tversky by recombining ideas from Economics and Psychology. Such curiosity-driven research or science for the sake of science is an important aspect of the scientific enterprise [4].

Since these dimensions of novelty embody distinct characteristics, their associations with antecedents and their impacts may be complex and multifaceted. We explore how industry publishing contribution to a field relates to (novelty occurrence) and the degree of novelty captured by novelty breadth and novelty distance of publications within the field (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Dimensions of Novelty examined in this study.

https://doi.org/10.1371/journal.pone.0346227.t001

Industry contribution to a field and novelty

Scientific research is organized into “fields,” which are research communities with shared interests, goals, norms of language, approaches, and institutional environments to knowledge creation [56,57]. These community-level differences shape different aspects of knowledge created within these fields, such as novelty, impact, and collaboration patterns [58].

Since university research is primarily driven by scientific curiosity and the academic reward system, while industry research is primarily shaped by commercial objectives, scientific research produced by these two entities often differs [18,23,36,59–61]. For instance, university typically pursue long-term basic research, whereas industry tends to prioritize short-term, application-oriented research aligned with market needs [18,36,61]. A high degree of industry contribution to publishing within a field often signals greater interaction and knowledge exchange between industry and university. Since there are differences in research by university and industry, a greater extent of industry contribution to a field may have important implications for the novelty of scientific publications within the whole field.

Industry tends to focus on practical and complex scientific problems [36,62], frequently requiring interdisciplinary research and diverse expertise [63]. Such a setting could enable novel recombinations as scientists from different backgrounds collaborate [63], potentially forging unexpected links between theories or paradigms [23]. This process can introduce a broader and more distant range of ideas into a field. Industry publications may sometimes themselves exhibit novel knowledge combinations [22,64–66], thereby introducing new ideas and approaches into a field. For example, Karl Jansky’s breakthrough discovery of extraterrestrial radio emissions, while investigating radio interference at Bell Laboratories, later stimulated extensive research in Radioastronomy [31]. Even when they do not display high novelty, the problems they introduce are often motivated by practical constraints and commercial applications, which may stimulate novel recombinations in follow-on research in the field. For example, semiconductor firms documented several engineering challenges associated with transistor scaling and the limits of Moore’s Law in scientific publications [67]. These discussions highlighted practical constraints in device performance and fabrication, which subsequently stimulated academic research on alternative transistor architectures, materials, and scaling approaches, many of which were developed by university researchers [67]. As a result, greater industry presence in a field can encourage novel recombinations by expanding the idea pool and offering new problems and perspectives that scientists can integrate with existing knowledge, thereby increasing the propensity for the occurrence of novel recombinations in their publications.

1. H1: Industry publishing contribution to a field has a positive relationship with the novelty occurrence of publications within the field.

While industry publishing contribution to a field can provide a larger pool of ideas driving the propensity for novel recombinations, it could also drive the magnitude and nature of novelty through other mechanisms. To explore this, we focus on two key dimensions of novelty breadth and novelty distance that capture distinct yet complementary aspects of how knowledge elements are recombined in scientific research:

Greater industry contribution to a field is associated with greater industry influence on the field that can shape the direction of scientific inquiry by steering research toward commercially-driven agendas [13–17,68]. On the one hand, research driven by commercial agendas can stimulate work on complex, real-world problems that often require interdisciplinary collaboration and the integration of diverse knowledge domains [36,62,63]. This may encourage researchers within the field to pursue broad-scope novel recombinations spanning several knowledge elements, thereby increasing novelty breadth in fields with high industry publishing activity.

On the other hand, growing industry influence can narrow research to topics of interest to industry firms, limiting the diversity of research within a field. For instance, in the field of Artificial Intelligence, studies show a narrowing of research focus over time, with increased alignment to commercially lucrative areas [69,70]. Extensive industry’s incentives to patent commercially valuable discoveries may also promote secrecy [17,71–74], delay the publication of novel findings [18], and shift the locus of disclosure from academic publications to patents [17–21]. Such dynamics contribute to the “marketization” of science [75], potentially restricting the free flow of knowledge and slowing collective progress. This could lead to a decrease in novelty breadth of publications within the field. Therefore, the net effect of industry publishing contribution on novelty breadth of publications within a field is ambiguous and warrants empirical examination.

1. H2a: Industry publishing contribution to a field has a positive relationship with the novelty breadth of publications within the field.
2. H2b: Industry publishing contribution to a field has a negative relationship with the novelty breadth of publications within the field.

Several scientific breakthroughs have emerged from curiosity-driven blue-sky research that recombines highly intellectually distant knowledge. Such research is characterized by high novelty distance, is inherently risky [58], and may have commercial relevance only in the long term. Although greater industry involvement in fields may promote broad, novel recombinations that can help solve complex applied problems that require interdisciplinary approaches, it may simultaneously divert academic attention in those fields from exploratory research driven by pure scientific curiosity. As a result, although the breadth of novel recombinations may increase with the rise in industry publishing intensity within a field, it could simultaneously discourage the depth of intellectual distance of novel recombinations (novelty distance). Therefore, the benefit of increased scope for novel recombinations in fields with greater industry publishing contribution may be accompanied by diminished novelty distance of publications within the field.

Additionally, similar to novelty breadth, greater industry influence within a field could lead to a decrease in novelty distance of publications by steering research in specific directions and encouraging secrecy within the field.

1. H3: Industry publishing contribution to a field has a negative relationship with the novelty distance of publications within the field.

In addition to testing these hypotheses, we examine whether the observed effects are driven by university or industry-authored publications and whether they vary for top-ranked research institutions. This extension allows us to assess heterogeneity in the mechanisms underlying our hypotheses and to draw implications for science policy.

Dependent variable: Novelty

Consistent with a recombinant perspective, we conceptualize novelty as arising from unconventional combinations of prior knowledge elements. Since our research question is situated at the field level, we require a measure of the knowledge elements available for recombination that is comparable across fields. The set of journals cited in a publication provides such a metric. Therefore, we use journal combinations as a proxy for knowledge recombinations, employing the measures of recombinant novelty as outlined by Wang et al. (2017) [2].

We examine two facets of a publication’s novelty: novelty occurrence, and the degree of novelty captured through novelty breadth, indicating the scope of novel recombinations, and novelty distance, indicating the extremity of conceptual divergence among recombined knowledge.

Novelty occurrence ()

A novel combination of journals in a publication signifies a novel recombination of existing knowledge (novelty occurrence). We identify ‘novel pairs’ of journals by filtering journal pairs in the publication’s references that appear for the first time in the 20 years leading up to the publication year. To incorporate only ‘useful’ novel recombinations, we include journal pairs that have been cited in at least one publication in the following three years. Furthermore, to ensure that each journal combination represents a substantive and credible instance of knowledge recombination, we include only journal combinations in which both journals fall above the 75^th percentile of the journal impact factor distribution for that year. This threshold helps in focusing on recombinations involving widely recognized sources of scientific knowledge. The 75^th percentile journal impact factor cutoff varies between 0.80 and 1.35 over the 15-year period.

takes a value of one if a publication has a novel journal pair and zero otherwise.

Degree of novelty

To evaluate the degree of novelty, we first calculate the intellectual distance associated with each novel journal pair referenced in a publication. Specifically, we compute the complement of the cosine similarity between the two journals, based on their citation co-occurrence patterns with other journals in the preceding three years. The cosine similarity captures the extent to which the two journals appear in similar citation contexts. A high cosine similarity indicates that the journals are frequently cited alongside the same set of other journals, suggesting a well-established relationship and low intellectual distance between them. In contrast, a low similarity score indicates that the journals originate from more distant knowledge areas, implying greater intellectual distance between them.

For each journal in a novel journal pair, we construct a co-citation profile that records the number of publications over the past three years in which the journal has been co-cited with every other journal. We compute the dot product of the two journals’ co-citation profiles to measure the overlap in co-citation patterns. This involves identifying all journals common in the two journals’ co-citation profiles, multiplying their respective publication counts, and summing these products.

We then calculate the cosine similarity between the two journals by scaling this dot product by the Euclidean norms of the two co-citation profiles. The Euclidean norm reflects the overall magnitude of a journal’s co-citation activity—how frequently it is co-cited with other journals. This normalization ensures that the similarity score captures the pattern of co-citation rather than just the volume, thereby indicating how closely related the two journals are in the prior citation landscape.

where Ji and Jj represent the co-citation profiles of the journals involved in the novel journal pair Ji-Jj.

Finally, to capture the intellectual distance of each novel recombination, we subtract the cosine similarity from 1. This complement transformation ensures that higher values indicate greater conceptual distance between the journals. A score of 1 for a novel journal pair indicates that the two journals have no prior co-citation overlap, suggesting they come from highly disparate knowledge areas and are difficult to recombine. A score of 0 implies a cosine similarity of 1, indicating the journals are frequently co-cited in similar contexts and, therefore, easy to recombine.

We illustrate this calculation using an example where a publication cites only three journals: J1, J2, and J3 in its references (Fig 1(a)). The resulting journal pairs for this publication are J1–J2, J2–J3, and J1–J3. Suppose that J2–J3 is a novel pair, i.e., the two journals have not been cited together in any publication over the past 20 years. Let us also assume that the universe of journals in our dataset consists of J1, J2, ..., J7. Fig 1(b) displays the co-citation profiles for J2 and J3. For instance, J2 has been co-cited with J1 in 20 publications, with J5 in 3, with J6 in 7, and with J7 in 4, and has not been co-cited in any with any other journals in the past three years. A similar co-citation profile is also created for J3. Among the co-cited journals of J2 and J3, J1 and J5 appear alongside both J2 and J3, making them the shared co-citation links between the two. The resulting dot product of the co-citation profiles of J2 and J3 is 127 (20×5 + 3 ×9), and the corresponding Euclidean norms are 21.77 and 14.35, respectively. This yields a cosine similarity of 0.32, and the intellectual distance between the journals J2 and J3 thus becomes 0.68.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. An illustration of a publication’s referenced journal pairs and their co-citation profiles for Novelty measurement.

(a) An illustration of journal pairs (J1-J2, J2-J3 and J1-J3) referenced in a publication. (b) An illustration depicting co-citation profiles of journals J2 and J3 in a novel journal pair (J2-J3).

https://doi.org/10.1371/journal.pone.0346227.g001

Novelty breadth ()

We evaluate the metric for novelty breadth of a publication by adding the intellectual distances between all new journal combinations in the references of the publication.

This metric reflects the cumulative conceptual distance involved in the publication’s knowledge integration. A higher novelty breadth indicates that novel recombinations in publications engage with a broader, more diverse set of knowledge elements.

Novelty distance ()

We evaluate novelty distance of a publication through the intellectual distance between the most disparate novel journal combination.

A higher novelty distance reflects that the publication integrates highly disparate ideas.

Independent variable: Industry publishing contribution to a field

Industry publishing contribution is our key variable of interest. We use the percentage of publications with at least one industry author to indicate industry publishing contribution to a field.

A publication can be assigned to more than one field. This violates the independence assumption in regression models. We address this problem by assigning a single field to each publication, flattening the measure so that each publication has a unique value for the field’s industry publishing contribution. We assign the field corresponding to the ‘maximum’ value of industry publishing contribution across all the fields of a publication as the field of the publication. We also run an additional analysis by assigning the field associated with the ‘minimum’ value of industry publishing contribution among the fields of the publication to the publication to get a lower bound on our estimates for robustness (Table S6.1: SI Appendix S6).

Consequently, the measure of industry publishing contribution to a field, corresponding to a given publication in year y takes the following form:

where, s is the set of fields the publication belongs to, and the field f assigned to a publication is given by,

We compute additional field-level metrics for a given publication corresponding to the field f in year y.

Since research takes time to develop and be published, and our focus is on how industry publishing contributions shapes research within fields, we lag the measure of industry publishing contribution by two years. Additionally, we lag all field-level control variables by two years in our analysis for consistency. We also use four-year lagged field-level variables for robustness (see SI Appendix S7 for details).

We also account for potential confounding factors at both the publication and field levels (see SI Appendix S3 for details). At the field level, we control for the field’s size and appliedness. At the publication level, we include controls for various characteristics such as the number of authors, the number of references cited, the number of fields in which the publication is classified, whether the publication involves international collaboration, the number of organizations affiliated with the authors, and the institutional affiliations of the authors. A brief description of variables is provided in Table 2.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Description of variables.

https://doi.org/10.1371/journal.pone.0346227.t002

Novelty occurrence (novelty occurrence)

Models 1–2 in Table 3 present the Logistic regression estimates for the relationship between industry publishing contribution to a field and novelty occurrence. In model 1, we begin by including the control variables. The findings for the control variables indicate that in line with prior work, the likelihood of novelty occurrence increases with team size (T_size) highlighting the importance of teams in driving novel recombinations [85], and decreases when the publication involves international collaboration [86] or a collaboration between authors from several organizations (N_orgs). In addition, a higher number of references in a publication (N_refs) is positively associated with novelty occurrence, potentially reflecting a greater scope for knowledge integration. We also find a positive relationship between the number of fields associated with a publication (Pub_fields) and novelty occurrence, suggesting that interdisciplinary works may increase the chances of new recombination. A higher journal impact factor (JIF) is associated with a lower likelihood of novelty occurrence, in line with prior work [2]. Further, on average, publications with some authors affiliated with university and some with industry (UI) have higher odds for novelty occurrence than those authored by solely university- or solely industry-affiliated authors (This reflects a higher potential of industry to introduce new ideas within a field.).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Regression estimates for the relationship between industry publishing contribution and different aspects of novelty.

https://doi.org/10.1371/journal.pone.0346227.t003

We add our key variable of interest (IndPubContribution_f) in model 2. Our findings support H1 by revealing a positive relationship between industry publishing contribution to a field and the odds of publications within the field being novel (Between models 1 and 2, the AIC decreases from 3082295 to 3044583 (>10), and the BIC decreases from 3082665 to 3044968 (>10), indicating a better fit of model 2 over 1. The sign reversal of the Industry coefficient between Models 1 and 2 reflects industry’s systematic selection into specific fields, which is expected given that industry-authored publications are more likely to occur in fields with high industry publishing contribution.). Specifically, a one standard deviation increase in the share of industry-authored publications within a field is associated with a 46.8% increase in the odds of a publication being novel.

Degree of Novelty (novelty breadth and novelty distance)

Next, we examine how industry contribution to publishing relates to the degree of novelty, focusing on novel publications within the fields. The estimates for control variables in models 4 (for novelty breadth) and 7 (for novelty distance) (Table 3) are largely consistent with those observed for novelty occurrence, with the exception of publication affiliation. On average, solely university-authored publications exhibit greater novelty distance than those from industry, indicating recombination of disparate ideas by university. No significant difference is observed between the two for novelty breadth.

We expect the data to reveal the direction of the relationship between industry publishing contribution and novelty breadth, and a negative relationship with novelty distance. However, our main results in models 5 and 8 (Table 3) uncover a nuanced trade-off: industry publishing contribution within a field is associated with an increase in novelty breadth but a decrease in novelty distance of publications within the field (Between models 4 and 5, the AIC decreases from 1050241 to 1049388 (>10), and the BIC decreases from 1050373 to 1049531 (>10), indicating a better fit of model 5 over 4. Between models 7 and 8, the AIC decreases from 125030 to 122523 (>10), and the BIC decreases from 125163 to 122666 (>10), indicating a better fit of model 8 over 7.). The findings align with H2a and H3. In terms of the effect size, a one standard deviation increase in the industry publishing contribution to a field is associated with a 4.4% rise in novelty breadth and a 2.6% decline in novelty distance of publications within the field.

A larger effect size for novelty occurrence suggests higher odds of producing novel research within a field. In contrast, the degree of novelty—captured by novelty breadth and novelty distance—shows more modest effects. It is important to note that, while the occurrence of novelty is relatively rare, achieving a high degree of novelty is even rarer. Therefore, even modest effect sizes for novelty breadth and novelty distance are meaningful and substantively important.

Our main results suggest that while greater industry publishing in a field can foster novel recombinations by expanding the idea pool, introducing new perspectives, and promoting the integration of diverse knowledge to address complex, industry-relevant problems, it may simultaneously divert academic attention away from exploratory, blue-sky research that depends on combining conceptually distant knowledge.

It is worth noting that our findings for all three measures, novelty occurrence, novelty breadth, and novelty distance, hold even after controlling for the degree of appliedness of a field. Appliedness of a field refers to the extent of commercial application of research within that field (see SI Appendix S3.1). While appliedness of a field captures the problem orientation of research, industry publishing contribution reflects the participation of firms as knowledge producers within the scientific field. These two dimensions, therefore, capture distinct aspects of the research environment. An explanation for the persistence of our results after accounting for appliedness is that, in fields with low appliedness, science can provide a theoretical understanding of the technological components and their interactions [23]. This suggests that the observed relationship between industry publishing contribution and novelty is not simply driven by the applied nature of the field, but also reflects the role of industry actors participating in scientific knowledge production.

University vs industry: Which entity drives novelty?

Having analyzed the association between industry publishing contribution and novelty, an important question that remains is whether the observed patterns are predominantly driven by university or industry-authored publications. If the observed relationship between industry publishing contribution and novelty is driven primarily by industry-authored publications, its effect would be limited to industry outputs. However, if the relationship persists across the entire field—including university-authored publications—it would suggest that industry publishing activity plays a formative role in shaping the field’s overall novelty. We explore this by comparing the relationship of novelty across the three groups of affiliations: solely university-authored (all authors affiliated with university), UI-authored (with some authors affiliated with university and some with industry), and solely industry-authored (all authors affiliated with industry) publications.

The negative coefficient estimates for the interaction of industry publishing contribution with UI- and solely industry-authored publications in model 3 (Table 3) reveal that the relationship between industry publishing contribution and novelty occurrence within a field is the strongest for solely university-authored publications. This suggests that university-authored publications are the primary drivers of the positive relationship between industry publishing contribution and novelty occurrence within fields. The effect sizes indicate that a one standard deviation increase in industry publishing contribution to a field is associated with a 58.7% increase in the odds of a solely university-authored publication exhibiting novelty occurrence, while the odds increase by 20.6% for UI-authored publications, and by only 2.3% for solely industry-authored publications.

Similar to novelty occurrence, the findings in model 6 (Table 3) indicate that university contributes the most to the positive relationship between industry publishing contribution and novelty breadth within fields. A one standard deviation increase in industry publishing contribution to a field corresponds to a 5% increase in novelty breadth for solely university-authored publications, a 2.6% increase for UI-authored, and a 1.3% increase for solely industry-authored publications.

However, Model 9 (Table 3) reveals that solely university-authored publications also exhibit the largest reduction in novelty distance as industry publishing contribution increases. This pattern suggests that university plays a central role in driving the negative relationship between industry publishing contribution and novelty distance. In terms of effect size, a one standard deviation increase in industry publishing contribution to a field is associated with a 3.1% decrease in novelty distance for solely university-authored publications, a 1.5% decrease for UI-authored, and a 0.1% decrease for solely industry-authored publications. Additionally, we also observe a cross-over pattern in the interaction effects: university-authored publications exhibit lower novelty occurrence and novelty breadth compared to other groups in fields with less than ~25% industry publishing contribution but higher novelty breadth in fields with more than ~25% industry publishing contribution (largely Bio-medical research related fields).

Therefore, the relationship between industry publishing contribution and all three measures of novelty is not exclusively driven by industry-authored publications, and university is an important driver of the relationship. These results highlight the importance of greater industry publishing contribution in driving the novelty of entire fields.

Institution research ranking

We additionally examine whether the relationship between industry publishing contribution and the three novelty dimensions differs for publications affiliated with high-ranking compared to other research institutions. To do so, we classify a publication as affiliated with a top-500 institution (Top 500) if at least one author’s institutional affiliation appears in the top-500 of the SCImago Institutions Rankings: Research Indicator [87].

Table 4 shows that high-ranking institutions are, on average, associated with lower novelty occurrence (Model 1) and novelty breadth (Model 3), but with higher novelty distance (Model 5). This pattern suggests that high-ranking institutions tend to recombine knowledge components that are cognitively distant from one another.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Regression estimates of the relationship between industry publishing contribution and novelty dimensions across top-500 (Top 500) and other research institutions.

https://doi.org/10.1371/journal.pone.0346227.t004

The interaction term Top 500×Ind PubContribution_f is not statistically significant in the models for novelty occurrence (Model 2) or novelty breadth (Model 4), suggesting that the relationship between industry publishing contribution and these novelty dimensions is similar for publications from top-ranked and other institutions. In contrast, the positive coefficient for the interaction Top 500×Ind PubContribution_f in Model 6 suggests that the negative relationship between industry publishing contribution and novelty distance is less pronounced for publications by high-ranking institutions compared to other publications.

The effect sizes indicate that, for publications from top-ranked institutions, a one standard deviation increase in the share of industry-authored publications within a field is associated with a 2.9% decline in novelty distance, compared to a 3.2% decline for other publications within the same field.

Therefore, while the associations between industry publishing contribution and both novelty occurrence and novelty breadth are similar across top-ranked research universities, the negative association with novelty distance is less pronounced for publications from top-ranked research universities. This suggests that high-ranking institutions are able to capture the benefits of increased novelty occurrence and novelty breadth, while partially mitigating the decline in novelty distance. One possible explanation is that top-ranked institutions possess stronger research infrastructure, greater autonomy in agenda-setting, and more stable access to long-term funding, which may allow researchers to pursue more exploratory and conceptually distant lines of inquiry, even in fields with substantial industry publishing. These findings remain consistent when the comparison is conducted using the top 100 research institutions versus others (SI Appendix S12).

We conduct additional analysis by stratifying the operationalization of industry publishing contribution to a field into two parts: industry publishing contribution emanating from the percentage of publications authored by university and industry scientists together within the field, and that emanating solely from industry-authored publications (see SI Appendix S5 for details). The findings (Table S5.1, SI Appendix S5) remain largely consistent with the main results. UI publishing contribution is positively associated with novelty occurrence and negatively associated with novelty distance, although the association with novelty breadth is not statistically significant. Similarly, sole-industry publishing contribution is positively associated with both novelty occurrence and novelty breadth, while its association with novelty distance is not statistically significant. These variations in statistical significance in some of the robustness tests may stem from the field-level nature of our analysis, where limited within-field variation could affect the precision of some estimates.

We also conduct several robustness checks for our main results. Our estimates remain consistent across alternative specifications for industry publishing contribution (SI Appendix S6 and S11), lag of field-level variables (SI Appendix S7), field assignment (SI Appendix S8), author affiliation (SI Appendix S9), and novelty (SI Appendix S10).