Methods
How we compare mission costs with tracked publications and citations.
Data as of Nov. 18, 2026.
The challenges of bibliometrics for space science missions
Space is hard, as the saying goes. So is measuring a space mission’s scientific output.
There is no perfect collection of science publications associated with a mission; there are only degrees of correctness. Large missions, like the Hubble Space Telescope, dedicate staff time to identifying science papers and distinguishing results from engineering descriptions and speculative work. A European team studying the European Space Agency’s (ESA) space science missions describes a transition from manual review to machine learning and natural language processing, with scientists still validating the selections. Such assessments can require access to the full text of large numbers of publications, often shielded by paywalls.
Exactly how one defines “science” varies, and many judgment calls are required. For this study, we used a simpler approach also used in many peer-reviewed publications, such as Fiore & Elvis (2026): querying NASA’s Science Explorer (SciX) database by mission name and the names of its primary instruments. SciX is the leading repository of space science publication and citation references. This approach assesses the signal of a mission within the abstract literature, using publications as a proxy for scientific productivity and citations to those publications as a proxy for impact.
We tailor each query to filter out obvious false positives from name or acronym collisions: TRACE, for example, is both a space mission and a very common word. We also compared our query outcomes to the managed collections of several missions with hand-curated publication sets. They generally agree within 10%–25% of publication records, with a single exception, GALEX, differing by a wide amount due to the pattern of not reporting the mission name in many abstracts. Since we’re primarily comparing mission costs across orders of magnitude, this is well within acceptable limits. We tested our outcomes with curated mission collections and found the overall comparative outcomes remained consistent.
Our aim is to identify broad patterns in the research associated with these missions. We call these records “tracked publications” to make their scope clear: they capture the papers we found, while leaving room for omissions and corrections.
What we included
This dataset contains data related to 143 projects built or funded by NASA’s science directorate from 1980 to 2024, spanning all five divisions.
The default comparison allows papers the publication year and three following calendar years to accrue citations.
Comparisons stay within each division. Research communities differ in size, publishing practices, and citation rates, so a higher count in one field does not mean its science is more valuable.
Tracked publications
The publications we tracked are the peer-reviewed research papers associated with each mission by searching their titles, abstracts, and keywords via NASA’s Science Explorer digital library (SciX). Publications are time-gated by the start of mission operations and continue through the full mission’s lifetime, plus two years to accommodate general publication times.
Our SciX queries use mission and major instrument names along with topic restrictions to distinguish ambiguous acronyms. The query for each mission is listed on each mission’s page.
This approach will miss papers that use mission data without naming the source in those fields. A name match alone also does not establish that a paper used the data.
As mentioned earlier, this approach is effectively measuring a mission’s signal strength in the SciX abstract fields. This is a common approach for bibliometrics, though more complex approaches are possible given access to the full text of journal articles and advanced machine learning analysis (see De Marchi, G., & Parmar, A. N., 2024).
What we exclude
We have general exclusions for nonrefereed material, conference proceedings, news articles, and datasets or data catalogs. We have additional mission-by-mission exclusions and additions based on known false positives reviewed with AI and human oversight.
Citation comparison windows
Older papers have had more time to collect citations, which would artificially inflate their metrics compared to more recent ones. To account for this, all metrics in this dashboard (except for the explicit lifetime metrics) use windowed citation counts over four calendar years: each paper’s publication year and the three years after it.
This means that missions launched after the full citation comparison window are not included in this study. A mission still operating limits publications to those that were published within this citation window.
The count runs by calendar year: a paper collects citations from papers dated its own publication year through three years later, so every paper gets four calendar years of exposure, from just over three years for a December paper to almost four for a January one. The length is a trade-off, and it’s ultimately an editorial decision. A longer window would count more of each paper’s citations but exclude more recent missions, since a mission’s papers count only once their windows have closed.
Eligibility for the main analysis follows the maturity of each mission’s prime-mission publication window. Failed missions with no recorded publications can qualify earlier under a waiting period after mission end. The “Missions awaiting inclusion” lists show lifetime data for launched missions that have not yet qualified; they do not contribute to the comparative totals. Inclusion dates are estimates based on the current mission dates and rules. When a required date is missing, the inclusion date remains unknown.
Mission costs
Costs come from NASA’s reported life-cycle costs at launch. We include documented partner contributions where available. These generally cover development, payloads, launch, and prime operations. Values are adjusted to 2025 dollars using NASA’s New Start Inflation Index.
Running averages
The citations-per-dollar charts on the overview and division pages show a running average. Each mission’s rate is its windowed citations per $100M of its cost, and a mission that failed or published nothing counts as zero. At each cost, we average the rates of missions within a factor of 10 of that cost, weighting the closest missions most heavily. We only draw the average where at least three missions contribute.
The shaded band is a 95% range from resampling the missions. To compare small and larger missions, we average the rates on each side of $150M and only say one side did better when the 95% range of the difference excludes zero and the result holds with any single mission left out.
Citation-weighted output (N50)
Our analysis uses both threshold counts to determine high-impact papers and a citation-weighted count, defined as: Nc = papers + citations ÷ c, where c is a citation cutoff. We set c to each division’s top-50% cutoff in the Active Mission Window and call the result N50. Because each division has its own c, N50 normalizes outputs across different divisions.
Each division’s cutoffs rank its distinct papers in the Active Mission Window by their citations in that window. A cutoff is the citation count of the paper at that point in the list, rounding the rank up: halfway down for c, and a tenth and a hundredth of the way down for the top-10% and top-1% cutoffs that define high-impact papers. Papers tied at a cutoff share the places left, so each counts as a fraction of a paper.
How we used AI in this project
AI was used extensively in this project to help write and review code, to refine and improve our SciX queries, and to identify papers for exclusion based on our internal criteria. All AI-driven query suggestions and exclusions were reviewed by humans. In accordance with The Planetary Society’s policy, beyond basic copyediting, AI was not used to write our analysis or narrative text. Some descriptive, functional text, such as tooltips on this dashboard, is AI-generated.
Corrections
If you find a missing or misattributed paper, a wrong date or cost, or any other error, write to casey.dreier@planetary.org.
Special thanks
The authors wish to acknowledge the staff and services of NASA’s Science Explorer digital library, who provided feedback on early drafts of this project and increased API access to build out our datasets. We are grateful for their work and the data provided by this service.
Image credits
NASA/ESA.
Further reading
- Space science & the space economy
Fiore, F., & Elvis, M. (2026). Space Policy, 75, 101713.
- A Measure of Total Research Impact Independent of Time and Discipline
Pepe, A., & Kurtz, M. J. (2012). PLOS ONE, 7(11), e46428.
- Assessing your Observatory’s Impact: Best Practices in Establishing and Maintaining Observatory Bibliographies
D’Abrusco, R., et al. (2024). The Open Journal of Astrophysics, 7.
- Computing and Using Metrics in the ADS
Henneken, E. A., et al. (2014). arXiv:1406.4542.
- Lessons from a High-Impact Observatory: The Hubble Space Telescope’s Science Productivity between 1998 and 2008
Apai, D., et al. (2010). Publications of the Astronomical Society of the Pacific, 122(893), 808–826.
- On the calculation of percentile-based bibliometric indicators
Waltman, L., & Schreiber, M. (2013). Journal of the American Society for Information Science and Technology, 64(2), 372–379.
- ESA Mission Publications and Their Impact
De Marchi, G., & Parmar, A. N. (2024). In ESA Science Programme Missions: Contributions and Exploitation, ISSI Scientific Report Series. Springer Nature Switzerland.
- Bibliometrics: The Leiden Manifesto for research metrics
Hicks, D., Wouters, P., Waltman, L., de Rijcke, S., & Rafols, I. (2015). Nature, 520, 429–431.