The 32% Citation Boost: Why Sharing Your Science Gets You More Citations
Papers that share their data earn 32% more citations, open-access papers earn 26% more, and papers sharing code earn 16% more—after controlling for everything e
32% more citations: the surprising personal payoff for sharing scientific data
The Numbers That Should Change How Scientists Share Their Work
In the four years between 2021 and 2025, astrophysics papers that shared their underlying data received 32 percent more citations than those that did not. Papers with open-access text—freely readable by anyone, anywhere—accumulated 26 percent more citations. And papers that posted their code on GitHub, so anyone could reproduce the analysis, garnered 16 percent more citations than those that kept their software locked away.
These aren't small, marginal effects hiding at the edge of statistical significance. The data advantage carries a p-value of less than —a number so vanishingly small that it's essentially impossible to explain away by chance. Open access text, already the norm in astrophysics at 80.5 percent of all papers, shows a signal so strong () that it's almost embarrassing. Open code, the rarest practice at just 0.6 percent of papers, still registers clearly ().
The findings come from a new study by Parth Joshi and Rupert Croft of Carnegie Mellon University, published on arXiv in July 2026. The researchers analyzed 53,194 peer-reviewed astrophysics papers—a massive sample covering nearly every significant publication in the field over a four-year window. They built a multivariate regression model that controlled for eight separate factors: paper age, number of authors, paper length, grant count, programming language, repository size, and astrophysical sub-field. The goal was to isolate, for the first time, the precise contribution of open-science practices to a paper's citation count while holding everything else constant.
The answer is unambiguous: openness pays. And it pays substantially.
The Science
Science has always depended on building blocks. No experiment exists in isolation; every result sits atop a tower of prior work, borrowed methods, and shared data. But traditionally, much of that tower has been hidden behind paywalls—locked journals that charge thousands of dollars for institutional access, proprietary datasets that require special permissions, and code written for one project that dies when the graduate student who wrote it moves on.
Open science is the movement to change that. It encompasses three distinct practices: making paper text freely available (open access), sharing the underlying datasets (open data), and releasing the software used in the analysis (open code). Each of these has its own community, its own infrastructure, and its own culture. Together, they represent a fundamental shift in how scientific knowledge circulates.
Astrophysics has been ahead of most fields on this front. Since 1991, astronomers have uploaded their preprints to arXiv before formal publication, making the vast majority of the literature freely readable years before it might appear behind a paywall. NASA's Astrophysics Data System (ADS) serves as a kind of Google for the field, with comprehensive bibliographic records that include not just citations but links to external resources: GitHub repositories, Zenodo archives, institutional servers. This infrastructure has made astrophysics a natural laboratory for studying what happens when openness becomes the default.
Joshi and Croft's study is the most comprehensive attempt yet to quantify exactly how much openness helps a paper's impact. They drew their sample from ADS in June 2025, restricting themselves to peer-reviewed papers published between January 2021 and April 2025—chosen specifically to exclude the COVID-19 pandemic years, which disrupted normal publication patterns and open-science culture in ways that would have confounded the analysis. A four-year window was necessary to accumulate enough papers for statistical power, especially for the rarest practices like open code, which appeared in just 337 of the 53,194 papers.
The researchers measured eleven quantities for each paper. Open-access status came from ADS's property field, which flags whether a paper is freely readable. They identified code-sharing by searching for links to GitHub—the dominant platform for scientific software—and data-sharing by looking for links to Zenodo, the open repository maintained by CERN. They retrieved repository sizes and programming language breakdowns from the GitHub API, and grant information from Crossref, the nonprofit that maintains Digital Object Identifiers for scholarly literature. Paper length in characters came from the full text, accessed either through ADS links or the Unpaywall database. Citation counts were pulled directly from ADS, and paper age was calculated from the publication date.
The most novel aspect of the study is how the researchers handled astrophysical sub-fields. Different areas of astronomy have radically different cultures around data sharing. Cosmologists and galaxy researchers routinely release massive simulation outputs—catalogs of thousands of galaxies, synthetic sky maps from the IllustrisTNG project, or spectroscopic surveys like SDSS and DES—through shared community repositories. High-energy astrophysicists working with X-ray and gamma-ray telescopes similarly have a tradition of releasing processed data products. Solar System researchers, by contrast, often work with mission data from instruments like Cassini or New Horizons, distributed through NASA's Planetary Data System rather than generic repositories like Zenodo. These differences matter. If cosmology papers simply share more data, a naive analysis might attribute cosmology's higher citation counts to data sharing when the real driver is the prestige of large collaborations or the inherent citability of big science.
To handle this, Joshi and Croft classified each paper into one of six sub-fields—Solar System, Planets, Stellar, ISM (interstellar medium), High Energy, and Galaxies + Cosmology—using keyword patterns from ADS's normalized subject terms. They then included sub-field as a fixed effect in their regression model, essentially asking: within each sub-field, does openness still predict citations? The answer is yes, for data sharing in every single sub-field.
The statistical approach combined three complementary methods. First, raw comparisons using the Mann-Whitney U test—non-parametric, making no assumptions about the shape of citation distributions—to establish basic patterns before controls. Second, multivariate ordinary least-squares regression with robust standard errors, the workhorse of causal inference in observational studies. Third, partial correlations to verify that the regression coefficients were not artifacts of multicollinearity. All models used the logarithm of (1 + citations) as the response variable, which is standard practice for citation data because raw counts are extremely right-skewed: a few landmark papers accumulate hundreds of citations while most hover in single digits. The log transform brings the distribution closer to normal, making standard errors more reliable.
The model included both a base specification without sub-field fixed effects and a full specification with them. Comparing the two reveals whether the openness effect is driven by field-level differences or persists within each field. The coefficients barely changed, which is exactly what you want: evidence that openness itself—not confounding differences between fields—is driving the citation boost.
What They Found
The headline results are striking enough to state plainly. After controlling for number of authors, paper length, grant count, paper age, and astrophysical sub-field, three independent citation advantages emerge: open data carries a 32 percent premium, open access text carries a 26 percent premium, and open code carries a 16 percent premium.
Open Data Sharing Rates by Astrophysics Sub-field
| Label | Value |
|---|---|
| Solar System | 1.8 |
| Planets | 2 |
| ISM | 1.9 |
| Stellar | 2.2 |
| HEA | 2.8 |
| Gal.+Cosmo | 2.9 |
These numbers represent multiplicative effects on citation counts. A paper that would otherwise earn 10 citations could expect 13.2 if it shared its data, 12.6 if its text were open access, and 11.6 if its code were public. In a field where citation counts influence hiring decisions, grant funding, and academic reputation, these are not trivial advantages.
The raw (uncontrolled) comparisons tell a more complicated story, and one that illustrates why statistical controls matter so much. Before adjusting for anything, papers with open code actually had fewer citations on average (median of 3 versus 5 for closed-code papers). This seems paradoxical until you look at paper age: open-code papers in the sample tend to be more recent, simply because code sharing has become more common over time. They've had less time to accumulate citations. Only after controlling for age does the open-code advantage emerge. This is a perfect example of Simpson's paradox—the direction of a relationship can flip depending on which variables you hold constant.
Open data and open access both show positive raw differences, but even these are partly inflated by confounds. Open-access papers tend to be slightly older on average (reflecting the transition from closed to open publishing), and they tend to have more authors, both of which independently predict more citations. The multivariate regression peels away these confounds to reveal the true signal.
The sub-field breakdown reveals important variation. Open-data sharing rates are highest in Galaxies + Cosmology (2.9 percent of papers) and High Energy Astrophysics (2.8 percent), and lowest in Solar System (1.8 percent) and the interstellar medium (1.9 percent). Open-code rates follow a similar pattern: 0.9 percent in Galaxies + Cosmology versus 0.5 percent in Solar System and ISM. This reflects the different research cultures: cosmology papers routinely release simulation data and survey catalogs that are valuable for the broader community, while Solar System research is often tied to specific instruments whose data live in dedicated archives that weren't captured by the study's Zenodo-based measure.
Citation Advantage from Open Science Practices
| Label | Value |
|---|---|
| Open Data | 32 |
| Open Access | 26 |
| Open Code | 16 |
The open-data citation advantage, however, is not confined to fields with strong sharing cultures. It appears in all six sub-fields. The effect is strongest in Galaxies + Cosmology and ISM, precisely the fields with the most developed data infrastructure, but it is present even in Solar System, where sharing is rarer. This suggests that when data is shared in Solar System research, it commands a substantial citation premium—perhaps because it's relatively uncommon and therefore more notable.
The partial correlations, computed by removing the linear contribution of all control variables from each predictor, confirm the regression results. Open access, open data, and open code all show positive partial correlations with log citations at high significance levels. The magnitudes differ slightly from the regression coefficients because partial correlations capture bivariate relationships, but the ordering is the same: data sharing shows the strongest effect, followed by open access, then open code.
Within sub-fields, the open-data advantage is statistically significant in four of the six categories (Solar System, Stellar, HEA, and Galaxies + Cosmology) and shows positive trends in the other two (Planets and ISM), though the small numbers of open-data papers in some sub-fields limit statistical power. The open-code advantage, meanwhile, is significant only in the full sample—likely because only 337 papers shared code, giving insufficient power for sub-field-level analyses.
The study also examined whether the size of shared code or data repositories matters for citations. Neither code size nor data size shows a significant correlation with citations among sharing papers, suggesting that the act of sharing itself—not the amount of material shared—is what drives the citation benefit.
Sample Distribution Across Astrophysical Sub-fields
| Label | Value |
|---|---|
| Gal.+Cosmo | 9.7 |
| Stellar | 10.2 |
| HEA | 6.9 |
| Solar System | 8.1 |
| Planets | 2.8 |
| ISM | 3.4 |
| No Keywords | 41.1 |
| Other | 17.9 |
Perhaps surprisingly, programming language doesn't predict citations either. Papers shared in Python, IDL, C++, or other languages all show similar citation distributions. This may be because language choice is confounded with sub-field: cosmology favors Python, while X-ray astronomy historically favored IDL. Once you control for sub-field, any effect of language appears to wash out.
Why This Changes Things
For decades, open-science advocates have made a moral and pragmatic case for sharing: science advances faster when knowledge is freely available, when data can be reused in unexpected ways, when code can be inspected for errors. These arguments are correct, but they've often felt abstract—a matter of collective benefit that might come at personal cost.
This study inverts that calculus. It shows that sharing doesn't just help the community. It helps you. A researcher who posts their data on Zenodo, who uploads their analysis pipeline to GitHub, who ensures their paper text is freely readable—they are, according to this evidence, more likely to be cited. And in an academic environment where citations shape hiring, tenure, and grant decisions, that is a concrete personal incentive.
This matters enormously for how we think about scientist behavior. Academic career structures are competitive and individualizing. Researchers respond to incentives. If the incentive structure rewards closedness—say, because unpublished data gives a competitive edge in a race—then even well-intentioned scientists may keep things under wraps. But if openness demonstrably improves citation counts, and citation counts demonstrably improve career outcomes, then the rational choice aligns with the collective good. Self-interest and altruism converge.
The 32 percent premium for open data is particularly striking because data sharing is so rare. Only 2.1 percent of the sample shared data on Zenodo, and the true rate—including papers that share through field-specific archives, institutional repositories, or supplementary materials—is certainly higher. But the fact that the citation benefit is largest precisely where sharing is most developed (Galaxies + Cosmology, HEA) suggests that the benefit grows with ecosystem maturity. When a critical mass of researchers share data and cite each other's shared data, a self-reinforcing cycle emerges. Early sharers get cited more, which incentivizes more sharing, which increases citability further.
The open-access premium of 26 percent is also noteworthy given that astrophysics already leads most fields in open-access adoption. More than 80 percent of the papers in this sample were open access—reflecting the field's decades-long arXiv culture. If the citation benefit still persists at such a high base rate, it suggests that additional openness beyond the norm still carries advantages. Perhaps papers that are open access are more likely to be read and cited by researchers at smaller institutions without journal subscriptions, or by the growing community of scientists in countries where journal access was historically limited.
The 16 percent open-code premium, while smaller in relative terms, may be the most underrated opportunity. Code sharing is vanishingly rare—just 0.6 percent of papers—and the signal is less statistically robust than for data or text. But that's exactly the point. In a landscape where sharing your code is unusual, doing so makes your paper stand out. It signals transparency and reproducibility. It lets other researchers build directly on your methods without reinventing the wheel. It invites collaboration. The low base rate means the competition for attention is less intense. For a researcher willing to invest the effort to clean and document their code, the citation returns may be disproportionately high.
The sub-field differences reveal something important about the sociology of astrophysics. Cosmology and high-energy astrophysics have invested heavily in community data infrastructure: shared simulation outputs, survey catalogs, telescope archives. This infrastructure lowers the barrier to sharing—it becomes natural when there are established platforms for doing so. Solar System and ISM research, by contrast, have less developed community repositories; data often lives in mission-specific archives with their own access protocols. The result is that sharing culture varies not because of any inherent difference in researcher values but because of different material conditions. Building more infrastructure—more repositories, more standards, more incentives—could plausibly raise sharing rates across the board, with corresponding citation benefits for everyone.
The study also has implications for research evaluation. Journal impact factors, which are based on citation counts, remain influential in hiring and funding decisions despite widespread recognition of their limitations. If open-science practices genuinely increase citations, then papers from well-resourced groups with the infrastructure and culture to share may accumulate impact factor credit that smaller groups cannot access. This could amplify existing inequalities in the research system. Acknowledging this doesn't mean abandoning citation counts as a metric—it's too embedded in the system to ignore—but it does mean being thoughtful about what we reward and why.
What's Next
No study is perfect, and Joshi and Croft are appropriately cautious about their limitations. Most importantly, they note that their measures of code and data sharing are almost certainly undercounts. They identify code sharing only when papers link to GitHub and data sharing only when papers link to Zenodo. Papers that share through other platforms—for example, institution-hosted repositories, personal websites, or field-specific archives like NASA's Planetary Data System—are missed entirely. This means their measured sharing rates (0.6 percent for code, 2.1 percent for data) are floors, not ceilings. The true rates are higher, which may mean the true citation benefits are slightly different when estimated against a larger sharing population. But the key result—that sharing is associated with higher citations—would likely hold or strengthen if all sharing were captured, since the undetected sharing papers would simply add noise to the "shared" group.
The grant measure is similarly imperfect. The researchers used the number of grants acknowledged as a proxy for total funding, but Crossref records only grants that have been formally registered, and larger public grants are far more likely to be registered than smaller private ones. This means the sample is probably skewed toward well-funded research, and the relationship between funding and citations—which is probably positive, since more-resourced groups can do more visible work—is not cleanly isolated. Future studies that link grant IDs to funding amounts could sharpen this control.
Causality remains the fundamental challenge. The study demonstrates association robustly, but the direction of causation is not definitively established. It is plausible that sharing causes higher citations—more visibility leads to more readers leads to more citations. But it is also plausible that papers likely to be highly cited for other reasons (novel results, prestigious affiliations, well-known authors) are more likely to share, either because their authors are more conscientious about open science or because highly cited papers face more pressure to share from reviewers and readers. The regression controls go a long way toward addressing this, and the consistency across methods (regression, partial correlations, non-parametric tests) is reassuring. But a true randomized controlled trial—assigning papers randomly to share or not share—is obviously impossible. The best we can do is triangulate across multiple approaches, as this study does, and acknowledge the remaining uncertainty.
Several next steps suggest themselves. Expanding the analysis to other scientific disciplines would test whether the citation benefits of open science are unique to astrophysics or are universal. Existing studies suggest the effect exists in biology and ecology, but comprehensive cross-disciplinary comparisons are rare. Such studies would need to account for field-specific citation cultures—astronomy papers are cited at different rates than biomedical papers—and field-specific sharing infrastructure, which varies enormously across disciplines.
Time trends in sharing and citation benefits deserve more attention. The study notes modest upward trends in open-code and open-data fractions over the 2021-2025 period, consistent with increasing adoption of open-science policies by journals and funding agencies. As sharing becomes more common, the citation premium may change. If everyone shares, the advantage of sharing may diminish—a first-mover benefit that erodes as the market saturates. Alternatively, if sharing becomes expected, not sharing may become penalized, inverting the effect. Longitudinal studies tracking how citation advantages evolve as sharing norms shift would be valuable.
The code-sharing result particularly deserves replication and extension. With only 337 code-sharing papers in the sample, statistical power is limited. Larger samples, drawn from longer time periods or from fields with higher code-sharing rates (ecology, for example, shows higher rates than astrophysics), would clarify whether the 16 percent premium is robust and whether it varies with repository characteristics, documentation quality, or software citation practices.
Finally, the policy implications deserve serious engagement. NASA has already moved toward requiring open access for all funded publications and data under its 2022 Scientific Information Policy. Plan S, the European initiative, imposes similar mandates. If these policies increase sharing rates while also increasing citation counts for individual researchers, they create a virtuous cycle: mandated sharing leads to more citations leads to happier researchers leads to more voluntary sharing. The personal incentive identified in this study could be the mechanism through which top-down mandates translate into lasting cultural change.
Science advances through accumulation. Every result builds on prior work; every new finding must be tested against existing knowledge. That process works best when the building blocks are accessible—when a researcher in Nairobi can read the same paper as a researcher at MIT, when a graduate student in São Paulo can download the same dataset as a postdoc in Munich, when anyone can run the same code and verify a published result. Open science is not merely a matter of fairness or ethics, though those considerations matter. It is a matter of effectiveness. And as this study demonstrates, it is increasingly also a matter of self-interest.
The evidence is clear: sharing your work is good for science, and it is good for you. The remaining question is not whether to open up, but how quickly we can build the infrastructure, norms, and incentives to make it the universal default.
Our findings support the long held notion that public access comes with concrete personal incentives for authors in terms of citations.
Sign in to join the conversation.
Comments (0)
No comments yet. Be the first to share your thoughts.