Balancing Data Use and Citation in Paleontology

PBDB publications were cited at a median rate of 4.28 times each year (median absolute deviation: 3.47), a significantly greater rate than annual citations for data-provisioning publications from the same period of time (1.35/year, median absolute deviation: 1.26; Wilcoxon rank sum test, p-value < 0.0001; Fig. 2; data available in Smith et al. [Reference Smith, Raja, Clements, Dimitrijević, Dowding, Dunne and Gee2023a]). When a citation was credited to each data-provisioning publication within the available subset of data-using publications (151 out of 396 PBDB publications with available data), the citation rate increased to 2.44 each year (median absolute deviation: 1.49). Assuming these 151 publications are a representative sample of all PBDB publications (see Supplementary Material for discussion of assumptions), extrapolating to the entire set of 396 PBDB publications increased the median citation rate for data-provisioning publications to 4.16 annual citations (median absolute deviation: 2.22), statistically indistinguishable from the median rate for PBDB papers (Wilcoxon rank sum test, p-value = 0.2103; Fig. 2)—using the harmonic mean as the summary statistic to downweight outliers (e.g., publications with extraordinarily high citation rates) resulted in a similar pattern (see “Supporting Analyses” in the Supplementary Material). These results suggest data-provisioning publications should be cited at a rate equal to that for the PBDB publications that reuse their data.

Figure 2.

Citation rates for official Paleobiology Database (PBDB) publications and the data-provisioning publications used in those PBDB publications. Only data-provisioning publications from the same time frame (since 2001) as PBDB publications are included to standardize for temporal effects. Citations to data-provisioning publications (i.e., primary literature) are presented as the current rate (i.e., no additions for neglected citations), the projected rate when including citations from PBDB publications where data were available (k = 112; i.e., additions), and the projected rate when making those additions and extrapolating to the entire set of PBDB publications (k = 396; i.e., additions and extrapolated).

It is clear that the status quo (Fig. 1A)—where the professional reward for PBDB publications is three times greater than for data-provisioning publications—does not give adequate recognition to the importance of data-provisioning publications and the effort required to produce them. At a minimum, the outputs of data-provisioning publications represent intellectual input by their authors and, in many cases, represent dozens or hundreds of hours of work and large financial investment (Agnarsson and Kuntner Reference Agnarsson and Kuntner2007; Ebach et al. Reference Ebach, Valdecasas and Wheeler2011; Baker and Mayernik Reference Baker and Mayernik2020; Melville et al. Reference Melville, Chapple, Keogh, Sumner, Amey, Bowles and Brennan2021). As has been broadly recognized in the literature, data producers deserve credit for their foundational work (Agnarsson and Kuntner Reference Agnarsson and Kuntner2007; Payne et al. Reference Payne, Smith, Kowalewski, Krause, Boyer, McClain, Finnegan, Novack-Gottshall and Sheble2012; Penev et al. Reference Penev, Mietchen, Chavan, Hagedorn, Smith, Shotton and Tuama2017; Cousijn et al. Reference Cousijn, Kenall, Ganley, Harrison, Kernohan, Lemberger and Murphy2018, Reference Cousijn, Feeney, Lowenberg, Presani and Simons2019; Kaufman et al. Reference Kaufman, Abram, Evans, Francus, Goosse, Linderholm and Loutre2018; Marwick and Birch Reference Marwick and Birch2018; Silvello Reference Silvello2018; Zhao et al. Reference Zhao, Yan and Li2018; Jones et al. Reference Jones, Grant and Hrynaszkiewicz2019; Lammey Reference Lammey2019; Pierce et al. Reference Pierce, Dev, Statham and Bierer2019; Tomaszewski Reference Tomaszewski2019; Colavizza et al. Reference Colavizza, Hrynaszkiewicz, Staden, Whitaker and McGillivray2020; Dosso and Silvello Reference Dosso and Silvello2020; Dorta-González et al. Reference Dorta-González, González-Betancor and Dorta-González2021; Hood and Sutherland Reference Hood and Sutherland2021). In a hypercompetitive academic environment where many aspects of an individual's career (e.g., reputation, career prospects, funding) are influenced by citation counts, this status quo for citation practice is neither fair nor sustainable (Agnarsson and Kuntner Reference Agnarsson and Kuntner2007; Neylon and Wu Reference Neylon and Wu2009; Payne et al. Reference Payne, Smith, Kowalewski, Krause, Boyer, McClain, Finnegan, Novack-Gottshall and Sheble2012; Piwowar and Vision Reference Piwowar and Vision2013; Tang et al. Reference Tang, Bever and Yu2017; Curry Reference Curry2018; Gingras and Khelfaoui Reference Gingras and Khelfaoui2018; MacRoberts and MacRoberts Reference MacRoberts and MacRoberts2018; Silvello Reference Silvello2018; Pierce et al. Reference Pierce, Dev, Statham and Bierer2019; Stern and O'Shea Reference Stern and O'Shea2019; Colavizza et al. Reference Colavizza, Hrynaszkiewicz, Staden, Whitaker and McGillivray2020; Dosso and Silvello Reference Dosso and Silvello2020; Raja and Dunne Reference Raja and Dunne2022). Without rebalancing the credit distribution (Fig. 1B), emerging “big data” research—considered here as research using large amounts of data (e.g., > 1 TB) including environmental data, images, stratigraphic information, taxonomic records, and more (Leonelli Reference Leonelli2014; Allmon et al. Reference Allmon, Dietl, Hendricks and Ross2018)—in paleontology is at risk of undercutting itself by contributing to a systematic devaluation of the specimen-based work that is foundational to the discipline itself.

The estimated citation rate for data-provisioning papers after the addition of neglected citations demonstrates the fundamental and underappreciated value of specimen-based work in paleontology. One way to acknowledge its value and to incentivize future specimen-based work is to cite the data in a formal way when they are used (Piwowar and Vision Reference Piwowar and Vision2013; Penev et al. Reference Penev, Mietchen, Chavan, Hagedorn, Smith, Shotton and Tuama2017; Cousijn et al. Reference Cousijn, Kenall, Ganley, Harrison, Kernohan, Lemberger and Murphy2018, Reference Cousijn, Feeney, Lowenberg, Presani and Simons2019; Kaufman et al. Reference Kaufman, Abram, Evans, Francus, Goosse, Linderholm and Loutre2018; Silvello Reference Silvello2018; Zhao et al. Reference Zhao, Yan and Li2018; Pierce et al. Reference Pierce, Dev, Statham and Bierer2019; Dosso and Silvello Reference Dosso and Silvello2020; Dorta-González et al. Reference Dorta-González, González-Betancor and Dorta-González2021; Hood and Sutherland Reference Hood and Sutherland2021). Although citations are inherently flawed as a metric and subject to biases (e.g., Gingras and Khelfaoui Reference Gingras and Khelfaoui2018; MacRoberts and MacRoberts Reference MacRoberts and MacRoberts2018; Davies et al. Reference Davies, Putnam, Ainsworth, Baum, Bove, Crosby and Côté2021; Hood and Sutherland Reference Hood and Sutherland2021; Raja and Dunne Reference Raja and Dunne2022), citations in one form or another are likely to continue being used to evaluate researchers (see Hicks et al. [Reference Hicks, Wouters, Waltman, de Rijcke and Rafols2015] for cautionary guidelines and Wilkinson et al. [Reference Wilkinson, Dumontier, Aalbersberg, Appleton, Axton, Baak and Blomberg2016] for discussion of FAIR principles). Citing data producers may be another step toward increased transparency and reproducibility in the pipeline from data production to digital upload and reuse (Wilkinson et al. Reference Wilkinson, Dumontier, Aalbersberg, Appleton, Axton, Baak and Blomberg2016; Escribano et al. Reference Escribano, Galicia and Ariño2018; Hood and Sutherland Reference Hood and Sutherland2021; see also Supplementary Table S1 and “Additional Contributions to the Paleobiology Database” in the Supplementary Material). Consequently, the development of a clear protocol for citing data can set a community-wide standard that preempts many of the shortcomings reported for traditional text citations (e.g., Gingras and Khelfaoui Reference Gingras and Khelfaoui2018; MacRoberts and MacRoberts Reference MacRoberts and MacRoberts2018; Davies et al. Reference Davies, Putnam, Ainsworth, Baum, Bove, Crosby and Côté2021). The recommended best practices for data citation from the broader literature, from which paleontology can draw (Payne et al. Reference Payne, Smith, Kowalewski, Krause, Boyer, McClain, Finnegan, Novack-Gottshall and Sheble2012; Kaufman et al. Reference Kaufman, Abram, Evans, Francus, Goosse, Linderholm and Loutre2018), include two general themes: (1) credit data provisioning by citing the publication in which the data were initially reported, or (2) use new metrics specifically developed for data citation.

Conceptually, the most straightforward way to credit data producers is to cite the publication from which the data were originally reported when the data are reused (Penev et al. Reference Penev, Mietchen, Chavan, Hagedorn, Smith, Shotton and Tuama2017; Cousijn et al. Reference Cousijn, Kenall, Ganley, Harrison, Kernohan, Lemberger and Murphy2018; Hood and Sutherland Reference Hood and Sutherland2021; and as recommended by some databases, e.g., BioTIME, https://biotime.st-andrews.ac.uk/usageGuidelines.php; Neotoma, https://www.neotomadb.org/data/data-use-and-embargo-policy). As the most basic option, this strategy carries the simplifying assumption that all authors participated in data production and credits them equally on this basis (see Pierce et al. [2019] for a counterargument). By virtue of its simplicity, this strategy for citing original publications facilitates ease of use through rapid integration into existing citation metrics, circumventing the need for an independent data citation tool. A prerequisite for using many data citation tools is a unique identifier for datasets (e.g., DOI), which is not available for many past publications (Hood and Sutherland Reference Hood and Sutherland2021) and, in recent publications, continues to be a shortcoming driven by poor adherence to data-sharing recommendations (Gabelica et al. Reference Gabelica, Bojčić and Puljak2022; see Agosti et al. [Reference Agosti, Benichou, Addink, Arvanitidis, Catapano, Cochrane and Dillen2022] for recommendations on use of identifiers). Many of the datasets included in the PBDB do not have unique identifiers, making the application of more complex data citation tools intractable. In alignment with our objective, citing original publications upon data reuse allowed for the most intuitive comparison between data-provisioning and data-using publications with a metric already familiar to academics.

It also bears stating that citing the database itself—in this case study, the PBDB—is necessary but not sufficient. As a consequence of being secondary sources of data, databases indirectly can create a barrier to citation of data-provisioning publications by masking the original data sources. Reflecting this issue, several databases (e.g., BioTIME, Neotoma) provide guidance on citing original data sources and, in the PBDB itself, recommendations toward this end have been made (https://paleobiodb.org/#/faq/how-should-the-paleobiology-database-data-be-cited-; see also Uhen et al. [Reference Uhen, Allen, Behboudi, Clapham, Dunne, Hendy and Holroyd2023] for a current user guide).

Citation practice is developing rapidly, as a host of data citation tools have been proposed—including the Data Citation Index (Clarivate 2023), SageCite (Lyon Reference Lyon2010), Data Usage Index (Ingwersen and Chavan Reference Ingwersen and Chavan2011), and Data Credit Distribution (Dosso and Silvello Reference Dosso and Silvello2020)—and multiple working groups have been convened on this topic (e.g., Scholix, Data Usage Metrics, Data Citation Synthesis Group). One of the driving principles behind the development of these metrics is the idea that data use is complex and therefore requires a tool that captures the nuances of data (Data Citation Synthesis Group 2014; Cousijn et al. Reference Cousijn, Feeney, Lowenberg, Presani and Simons2019; Dosso and Silvello Reference Dosso and Silvello2020; Hood and Sutherland Reference Hood and Sutherland2021). As a scientist's value cannot be distilled to a single metric, using several of these tools in combination with other measures of a person's contributions to science and society may be a more equitable option for evaluating scientists in the future (Neylon and Wu Reference Neylon and Wu2009; Curry Reference Curry2018; Ewers et al. Reference Ewers, Barlow, Banks-Leite and Rahbek2019; Stern and O'Shea Reference Stern and O'Shea2019; Davies et al. Reference Davies, Putnam, Ainsworth, Baum, Bove, Crosby and Côté2021; Hood and Sutherland Reference Hood and Sutherland2021; Westoby et al. Reference Westoby, Falster and Schrader2021). Given the attention to citation practice and alternative metrics in the recent literature and the progress made by working groups on the topic, it may only be a matter of time before data citation metrics become mainstream (Data Citation Synthesis Group 2014; Kaufman et al. Reference Kaufman, Abram, Evans, Francus, Goosse, Linderholm and Loutre2018; Cousijn et al. Reference Cousijn, Feeney, Lowenberg, Presani and Simons2019; Hood and Sutherland Reference Hood and Sutherland2021).

Regardless of the citation approach, attributing credit to data provisioning (and all individuals involved in the process of making data available in digital compilations; see Supplementary Table S1; see also Escribano et al. Reference Escribano, Galicia and Ariño2018; Benichou et al. Reference Benichou, Buschbom, Campbell, Hermann, Kvaček, Mergen, Mitchell, Rinaldo and Agosti2022) in a professionally meaningful way represents a shift in citation practice and credit distribution in paleontology (Payne et al. Reference Payne, Smith, Kowalewski, Krause, Boyer, McClain, Finnegan, Novack-Gottshall and Sheble2012; Kaufman et al. Reference Kaufman, Abram, Evans, Francus, Goosse, Linderholm and Loutre2018; Fig. 1). With the rise of quantitative paleontology and the associated shift away from paleontology's traditional descriptive roots, it is imperative we find reasonable and equitable ways to improve our data citation practices. For example, a single publication might draw data from thousands of primary sources (Supplemental Fig. S4) and, particularly for journals with strict page limits or length-based page charges, it is often not feasible to include citations for each of the data-provisioning publications. Though it will continue to be impractical to cite thousands of papers in printed format, the growing awareness of the importance of data citation and improving digital infrastructure provide a path forward. As a starting point, online archives and preprint servers (e.g., BioArXiv, EarthArXiv, Open Science Framework) can accommodate the long list of references required to cite all data-provisioning publications used by a publication based on the PBDB or another database. These online archives and preprint servers are routinely indexed by aggregators (e.g., Google Scholar, Web of Science). Publication of a reference list with an online repository or preprint server can increase the likelihood that citations are attributed to data-provisioning publications but, critically, the references must be included with the main text in the references section, not placed in the supplementary material. Current processes for aggregating citations do not find references in supplementary material. Raja et al. (Reference Raja, Dimitrijević, Krause and Kiessling2022a) illustrated the feasibility of this approach by publishing their database references in a preprint hosted at Open Science Framework (Raja et al. Reference Raja, Dimitrijević, Krause and Kiessling2022b). To alleviate the burden and facilitate consistency of compiling these large reference lists, future authors can use the R package refer (https://github.com/adamkocsis/refer). This package offers tools to generate a formatted document containing the metadata and reference list that can be used to upload to the aforementioned online archives. The user is required to provide either a text file containing formatted references or a BibTeX file containing the references in their data-using publication along with other generic information (e.g., title of the publication, author affiliations). The package also includes a ready-made template for the formatting of the document, and experienced users can provide their own templates.

Still, this solution is a stopgap measure, and it would be preferable for journals to implement policies and technical changes on their platforms to encourage more equitable citation practices. Although many journals still print hard copies, essentially all journals have online versions, and many journals are now published online-only. Even so, many online publishers have retained strict manuscript length policies, thereby limiting the number of references allowed. Rather than relegating data reference information to the depths of supplementary material where they will not be included in citation counts, online journals can, as a first step, omit the reference list from their imposed page limits. Another option for journals is to require authors to submit a list of references for data-provisioning publications as an appendix and to publish this list with the main text references in the online version (e.g., McGill et al. Reference McGill, Dornelas and Field2016)—a printed issue could still include only the references cited in the main text. Encouragingly, some journals (e.g., Global Ecology and Biogeography [McGill et al. Reference McGill, Dornelas and Field2016], Scientific Data [personal experience, e.g., Raja et al. Reference Raja, Dimitrijević, Krause and Kiessling2022a]) have already made these changes, allowing authors to fully cite their data sources. Admittedly, these changes will be somewhat onerous, as they require managing and formatting thousands of references; however, AI tools and the refer package presented here are viable options for streamlining this process. Whether these changes are adopted more broadly will depend on demand from the community.

Broader Considerations for Paleontology as a Discipline

Improving data citation practice will also have a positive effect on paleontological journals, especially those that publish specimen-based work (e.g., Acta Palaeontologica Polonica, Journal of Paleontology, Journal of Vertebrate Paleontology). Whereas higher-profile outlets (e.g., Science, Nature) tend to publish paleontological articles on charismatic and unusual specimens (e.g., dinosaurs, fossils in amber) or on large data compilations (e.g., latitudinal diversity gradients, extinction), most paleontological papers are published in discipline-specific journals (Raja and Dunne Reference Raja and Dunne2022). As might be expected, publications in these journals traditionally receive fewer citations, and the journals have lower impact factors.

Just as the citation rate for data-provisioning publications increased after accounting for neglected data citations (Fig. 2), the JIF—another flawed but commonly used evaluative metric (e.g., Neylon and Wu Reference Neylon and Wu2009; Stephan et al. Reference Stephan, Veugelers and Wang2017; Curry Reference Curry2018; Stern and O'Shea Reference Stern and O'Shea2019)—increases substantially for paleontological journals (Fig. 3A). Combining our tabulated neglected citations with currently attributed citations used to calculate JIF by Clarivate (https://jcr.clarivate.com/jcr/home), we found that in the last decade (2010–2019), the JIF reported for a journal in a given year (e.g., Journal of Paleontology in 2015) would increase on average by ~0.1, or 5.08%. This is a conservative estimate, as it only includes neglected citations from the 151 PBDB publications for which data were available. Extrapolating to the entire dataset of 396 PBDB publications suggests that any of the 55 journals categorized by Clarivate as a paleontological journal would see an increase in JIF by ~0.2, or 13.3% (see “7_paleo_journal_JIFcalculation.csv” in Smith et al. [Reference Smith, Raja, Clements, Dimitrijević, Dowding, Dunne and Gee2023a] for raw data for all 55 paleontological journals from 1997 to 2021 and additional information on language and country of publishing). The change in JIF from neglected citations was not, however, uniform across journals or through time. Whereas some journals had no (n = 10; e.g., Micropaleontology, Paleoceanography and Paleoclimatology, Stratigraphy) or few (e.g., GFF, Palaios, Zootaxa) neglected citations in our dataset and limited associated changes to JIF, other journals would have substantial increases in JIF in one or more years (e.g., Palaeontologia Electronica, Palaeontology, PalZ). Furthermore, for those journals with large JIF changes, there is a notable increase in the effect of adding neglected citations in more recent years (Fig. 3B). Neglected citations rarely contributed to JIF in the early part of the decade (2010 onward); however, at the end of the decade, the average JIF for 10 highly impacted journals increased by 27% in 2018 and 36% in 2019 when recalculated to include neglected citations. These differences through time are a consequence of the formula for calculating JIF and publishing trends in paleontology (Fig. 3C,D). Because JIF for a given year (e.g., 2020) is based on citations of research published in the preceding 2 years (e.g., 2018 and 2019), a relatively short turnaround time is needed between publication of a data-provisioning study and subsequent PBDB publication using those data. Consequently, many instances of data reuse cannot be incorporated into this metric because of the limited look-back period. Alternatives to the 2-year JIF do exist (e.g., 5-year JIF); however, analyses comparing 2- and 5-year JIFs show minimal differences between the two (e.g., Campanario Reference Campanario2011; Dorta-González and Dorta-González Reference Dorta-Gonzalez and Dorta-González2013). Though it continues to be a widely used metric across many branches of science, JIF performs poorly when capturing reuse of data and undervalues journals where data-provisioning studies are published. Accelerating publication rates in paleontology (Fig. 3C,D) and the shift from printed to online publication appear to have reduced the time between initial data publication and data reuse. Moreover, the number of paleontological journals published in 2021 was 55, more than double the 24 published in 1997 when Clarivate began compiling Journal Citation Reports. As the number of publications and citations in paleontology continues to grow (Fig. 3C,D), so too will the consequences of neglected data citations (Fig. 3A,B). A more equitable future in paleontology requires rapid correction to citation practice.

Figure 3.

The effects of adding neglected citations from data reuse on journal impact factor (JIF; A, B) and general patterns in publishing trends in paleontology (C, D). A, The increase in JIF for the 55 journals categorized to paleontology by Clarivate, for the period of 2010 to 2019. Note, an outlier value of 172% in 2018 for PalZ was not plotted. B, Increases in JIF for the 10 paleontological journals most affected by neglected citations, only including those with complete data for the duration of 2010 to 2019. For raw data for all 55 paleontological journals from 1997 to 2021, see “7_paleo_journal_JIFcalculation.csv” in Smith et al. (Reference Smith, Raja, Clements, Dimitrijević, Dowding, Dunne and Gee2023a). C, The number of citable items published in paleontological journals each year. D, The number of citations to items published in paleontological journals each year.

JIF influences more than how journals rank in comparison to one another; it also influences how authors and the work they publish in those journals are regarded and rewarded professionally (Neylon and Wu Reference Neylon and Wu2009; Stephan et al. Reference Stephan, Veugelers and Wang2017; Curry Reference Curry2018; Stern and O'Shea Reference Stern and O'Shea2019). Despite the poor performance of JIF as an indicator of quality, JIF continues to influence an author's choice of publication venue and contributes to the perceived importance of the papers published in the journal and, more broadly, the discipline (Neylon and Wu Reference Neylon and Wu2009; Curry Reference Curry2018; Stern and O'Shea Reference Stern and O'Shea2019). In lieu of systematic changes in publication practices in science (e.g., Kravitz and Baker Reference Kravitz and Baker2011; Curry Reference Curry2018; Davies et al. Reference Davies, Putnam, Ainsworth, Baum, Bove, Crosby and Côté2021), increasing the prestige of discipline-specific journals is imperative for increasing the profile of paleontology and will benefit all in the discipline.

Data sharing—particularly when credited appropriately—is an equally important component in any effort to strengthen the field of paleontology. As with data citation, the issue of data sharing is commonly discussed in the literature, and there is a consensus that it is incumbent upon authors to share the data they use to produce their results (e.g., Piwowar and Vision Reference Piwowar and Vision2013; Kaufman et al. Reference Kaufman, Abram, Evans, Francus, Goosse, Linderholm and Loutre2018; Marwick and Birch Reference Marwick and Birch2018; Jones et al. Reference Jones, Grant and Hrynaszkiewicz2019; Lammey Reference Lammey2019; Mandeville et al. Reference Mandeville, Koch, Nilsen and Finstad2021). Even so, data sharing is not practiced consistently (Stuart et al. Reference Stuart, Baynes, Hrynaszkiewicz, Allin, Penny, Lucraft and Astell2018; Gabelica et al. Reference Gabelica, Bojčić and Puljak2022; Roche et al. Reference Roche, Berberi, Dhane, Lauzon, Soeharjono, Dakin and Binning2022). Several large publishers (e.g., Elsevier, Springer, Taylor and Francis, Wiley) have data availability policies with multiple tiers, ranging from written recommendations to strict requirements for publishing data, but it remains at the discretion of journals to enact and enforce these policies (Jones et al. Reference Jones, Grant and Hrynaszkiewicz2019)—this may change in the United States, however, with a new mandate for public availability of data produced in federally funded research, beginning in 2023 (National Science and Technology Council 2022). As demonstrated by Gabelica et al. (Reference Gabelica, Bojčić and Puljak2022), who found that data were only available for 6.8% (n = 3,556) of publications in their review of 333 open access journals from BioMed Central, many data-sharing policies are not effective in practice. Data availability was considerably better in the present study, with data accessible for 32% (n = 128) of PBDB publications—note that though data were available, they were not always usable for the analysis conducted here. Encouragingly, when data from PBDB publications were not readily available online, 167 of the 268 (68%) authors who were contacted were responsive, and approximately half (n = 84) of these responses included the requested data. Still, we were unable to recover data for 21% (n = 83) of PBDB publications for myriad reasons. With pushes toward big data science in paleontology and related disciplines, it will be up to the community to influence journal policies toward required sharing rather than relying on unenforceable recommendations (Payne et al. Reference Payne, Smith, Kowalewski, Krause, Boyer, McClain, Finnegan, Novack-Gottshall and Sheble2012; Kaufman et al. Reference Kaufman, Abram, Evans, Francus, Goosse, Linderholm and Loutre2018; Jones et al. Reference Jones, Grant and Hrynaszkiewicz2019).

Improved data sharing requires buy-in from individuals, who may themselves benefit from the practice and enhance the quality of science in paleontology. As reviewed by Marwick and Birch (Reference Marwick and Birch2018), there are many reasons to share data (e.g., reciprocal data sharing by others; reproducibility of research; enabling others to ask new questions) and some associated costs (e.g., time required to clean data; data use without citation). One of the incentives is that data sharing is associated with increased citation of the publication where the data were initially published (Sears Reference Sears2011; Piwowar and Vision Reference Piwowar and Vision2013; Tomaszewski Reference Tomaszewski2019; Colavizza et al. Reference Colavizza, Hrynaszkiewicz, Staden, Whitaker and McGillivray2020; Dorta-González et al. Reference Dorta-González, González-Betancor and Dorta-González2021). For example, Colavizza et al. (Reference Colavizza, Hrynaszkiewicz, Staden, Whitaker and McGillivray2020) reported that when publications included data availability statements with the associated data publicly accessible, those publications saw a 25% increase in their citations compared with publications without available data. As demonstrated here (Fig. 2), the potential citation benefit may be even larger in a discipline like paleontology, where publications on data compilations have become mainstream. Changes to the format on funding proposals, for example, inclusion of a “research outcomes” section that includes datasets by the Deutsche Forschungsgemeinschaft (i.e., German Research Foundation) and a non-publication section in National Science Foundation grant reports, can further encourage data sharing. Of perhaps greater importance, data sharing ensures the reproducibility of scientific results (Piwowar and Vision Reference Piwowar and Vision2013; Altman et al. Reference Altman, Borgman, Crosas and Matone2015; Marwick and Birch Reference Marwick and Birch2018). As has been demonstrated to the detriment of many fields of study (e.g., behavioral ecology [Viglione Reference Viglione2020], food science [van der Zee et al. Reference van der Zee, Anaya and Brown2017], paleontology [Price Reference Price2022], psychology [John et al. Reference John, Loewenstein and Prelec2012]), some researchers have been guilty of misrepresenting their data. Data sharing provides a means to uphold academic integrity and establishes an ethical and practical standard that encourages scientific advancement (Marwick and Birch Reference Marwick and Birch2018; Raja and Dunne Reference Raja and Dunne2022).

Paleontology has not yet crossed the threshold to become a big data discipline (Allmon et al. Reference Allmon, Dietl, Hendricks and Ross2018) but has the potential to do so in the near future. Realizing this potential will expand research horizons in paleontology but, to be done effectively and equitably (e.g., Raja et al. Reference Raja, Dunne, Matiwane, Khan, Nätscher, Ghilardi and Chattopadhyay2022c), it requires a stable foundation in specimen-based work and reckoning with structural biases. Large paleontological databases, including the PBDB, are far from complete. For example, in examining the collections at nine paleontological museums in the United States, Marshall et al. (Reference Marshall, Finnegan, Clites, Holroyd, Bonuso, Cortez, Davis, Dietl, Druckenmiller and Eng2018) estimated that there were 23 times the number of unique localities in only those nine collections than were in the PBDB at the time. Paleontologists should be wary of assuming our databases are comprehensive, as “having a lot of data is not the same as having all of them; and cultivating such an illusion of completeness is a very risky and potentially misleading strategy” (Leonelli Reference Leonelli2014: p. 7). Activating the extensive data held in museum collections (e.g., unpublished specimens; “extended specimen” data; Webster Reference Webster2017; Allmon et al. Reference Allmon, Dietl, Hendricks and Ross2018; Marshall et al. Reference Marshall, Finnegan, Clites, Holroyd, Bonuso, Cortez, Davis, Dietl, Druckenmiller and Eng2018), will require support for the infrastructure sustaining collections and recognition of the importance of specimen-based work that often goes wanting in paleontology and related disciplines (Johnson et al. Reference Johnson, Filkorn and Stecheson2005; Agnarsson and Kuntner Reference Agnarsson and Kuntner2007; Payne et al. Reference Payne, Smith, Kowalewski, Krause, Boyer, McClain, Finnegan, Novack-Gottshall and Sheble2012; Allmon et al. Reference Allmon, Dietl, Hendricks and Ross2018; Marshall et al. Reference Marshall, Finnegan, Clites, Holroyd, Bonuso, Cortez, Davis, Dietl, Druckenmiller and Eng2018; Engel et al. Reference Engel, Ceríaco, Daniel, Dellapé, Löbl, Marinov and Reis2021; Benichou et al. Reference Benichou, Buschbom, Campbell, Hermann, Kvaček, Mergen, Mitchell, Rinaldo and Agosti2022). A critical component to realizing a big data future in paleontology will be increased funding to support museum collections and data repositories, with respect to both maintaining existing materials and to obtaining and curating new materials. Illustrating the scope of the need, Allmon et al. (Reference Allmon, Dietl, Hendricks and Ross2018) estimated that it costs US$1 to digitize each specimen, and digitizing only the currently identified specimens in U.S. collections (as of 2018) would require an investment of US$35 million. That figure increased to US$75 million after including all fossils, not just those with existing taxonomic identifications. Investment at this scale represents a massive increase in funding, as the budget for this type of work in the United States was only US$10 million at the time (Allmon et al. Reference Allmon, Dietl, Hendricks and Ross2018). These monetary estimates also do not account for the costs of data storage and maintenance of data repositories (whether museum-based or external) that provide access to other researchers and the public. Particularly, as complex data become more commonplace (e.g., CT scans, images), infrastructure requirements will be critical to ensuring a big data future in paleontology. Without funding for this fundamental work, growth and advances in paleontology will be slow at best.

The illusion of completeness is elucidated further when considering biases in where data recorded in paleontological databases originated, what organisms are preferentially studied, who contributes to compiling data in databases, and who conducts the research (e.g., Raja and Dunne Reference Raja and Dunne2022; Raja et al. Reference Raja, Dunne, Matiwane, Khan, Nätscher, Ghilardi and Chattopadhyay2022c). Compilations of modern biodiversity data show a clear association between data production and wealthier, more resource-rich countries, particularly those in western Europe and North America (Amano and Sutherland Reference Amano and Sutherland2013; Hughes et al. Reference Hughes, Beas-Luna, Barner, Brewitt, Brumbaugh, Cerny-Chipman and Close2017). The same is true for compilations of paleontological data; a recent study examining data recorded in the PBDB found that 97% of fossil occurrence data were produced by researchers based in high- or upper middle-income countries (Raja et al. Reference Raja, Dunne, Matiwane, Khan, Nätscher, Ghilardi and Chattopadhyay2022c). The same study found a direct link between paleontological data production and socioeconomic factors, such as greater wealth, education level, and political stability (Raja et al. Reference Raja, Dunne, Matiwane, Khan, Nätscher, Ghilardi and Chattopadhyay2022c). These patterns clearly illustrate a global knowledge and power imbalance in paleontological research that can only be rectified by changes to how paleontological research is conducted (Cisneros et al. Reference Cisneros, Raja, Ghilardi, Dunne, Pinheiro, Fernández and Sales2022; Monarrez et al. Reference Monarrez, Zimmt, Clement, Gearty, Jacisin, Jenkins and Kusnerik2022; Raja et al. Reference Raja, Dunne, Matiwane, Khan, Nätscher, Ghilardi and Chattopadhyay2022c).