What a university ranking can tell you about a research partner

Suppose a company wants a research partner for a battery-recycling project. A university’s strong industry-collaboration score puts it on the shortlist. Before approaching a team, the research lead needs to know what earned that position and whether the evidence relates to the proposed work.

A ranking can help you find institutions to investigate. Selecting a research partner requires evidence about the particular team and project. My recommendation is to use the ranking as a lead, then make the selection on that evidence.

What makes a ranking useful for a decision?

Calling a ranking “biased” can conceal several different objections. I would separate three questions.

First, is the record accurate? A wrongly assigned affiliation is a data problem. Second, what does the method value? Counting total output and counting a share of output answer different questions. Third, does that question match yours? A carefully measured institutional characteristic can still be insufficient for selecting a team.

This distinction matters when deciding how much to trust the result. An error calls for correction. A weighting choice calls for an explanation and a sensitivity check. A mismatch with the decision calls for additional evidence. Treating all three as a general legitimacy problem makes the next step less clear.

The research on university-industry metrics already makes the measurement problem explicit. In a study of 750 large research-intensive universities selected by publication output, Tijssen, Yegros-Yegros, and Winnink compared measures of university-industry links and alternative composite rankings. Absolute co-publication counts and their share of university output produced different ranking pictures. The authors recommended a transparent scoreboard with separately selectable metrics, including a separation of measures that depend on institutional size from those that do not. Tijssen et al., 2016¹.

Vernon, Balas, and Momani approached a different question: whether rankings support research improvement. Their review examined 13 eligible systems using editions from 2015 or 2016. They assessed transparency, replicability, consistency, and the measures included, and concluded that existing indicators inadequately evaluated research outcomes. This is a historical evaluation of those systems and that purpose. It does not establish the validity of a new index or predict the outcome of a company partnership. Vernon et al., 2018².

My application of this literature is simple: before using a rank, write down the decision it is supposed to inform. “Find universities with visible industry research activity” and “choose the team most likely to solve our problem” require different evidence.

Check the source and version before deciding how far the evidence transfers.

What does a collaboration index actually observe?

The 2026 Industry Collaboration Index provides a current example. Its published scoring method uses verified university-company co-authored publications from 2021–2025. Four pillars combine co-creation, durability, network scale, and partner-field breadth. Scale explicitly reflects network size.

The method’s verification is not error-free. The captured methodology reports both retained false links and missed true links. Its later audits used AI assistance, and an independent human check was still pending on 11 October. Treat “verified” as the publisher’s classification under its rules, not a guarantee that every affiliation is correct.

The method also publishes rank sensitivity ranges under specified changes to evidence and weights. These are not formal tests of equal positions, and they do not include every source of error, such as affiliation, company-grouping or sector-classification errors. When the ranking results are available, use the reported ranges alongside the components to judge whether a difference in position warrants further investigation. Read the stated limits alongside each range.

Edition 1 excludes confidential contract research, consultancy, licensing, and patents. Sector rankings use their own formula and eligible population. ICI methodology³.

For a company, this suggests a useful search signal: a record of publicly documented joint research. It also sets a boundary. A team working mainly through confidential projects could be relevant without leaving the publication trail this method observes. That possibility needs investigation; absence from the measured record does not establish absence of capability.

Equally, a university could have an extensive industry network while the team you need has no capacity for a new project. The institution and the proposed delivery team are different units of assessment. This is the inference I would challenge in a selection meeting.

What evidence should follow the shortlist?

Build the shortlist through parallel searches in relevant publications, specialist networks and team directories as well. Ranking eligibility and visible co-authorship should not become requirements for discovering a suitable team.

Return to the battery-recycling example. Imagine that one shortlisted university has broad industry activity, while another has a smaller group with relevant methods and access to the required pilot equipment. Those are hypothetical conditions, not findings about named universities.

I would compare the teams against the project brief before assigning either an advantage. The following table is my proposed decision aid, not a validated predictor of partnership success.

Project question	Evidence from the team	What the rank leaves open
Does the expertise fit?	Relevant work, methods, equipment access	Expertise in the specific problem
Is the timing feasible?	Named contributors, committed time, facility slots	Current capacity
Do the working terms fit?	Data access, publication, intellectual property, responsibilities	Compatibility of the arrangement
How is success assessed?	Research question, milestones, acceptance criteria, stop rules	This project’s outcome

An unanswered question should remain visible in the selection record. Substituting institutional standing for a missing answer creates confidence without resolving the uncertainty.

When should the institutional score carry more weight?

There is a reasonable counter-case. A company seeking a long-term relationship across several research areas may care about institutional breadth, recurring industry activity, and the surrounding network. In that search, an institution-level measure could be closer to the decision than it is for a single technical project.

The scope of the decision determines which evidence matters. I would retain relevant ranking components and investigate the underlying work, and keep the overall position in its stated measurement scope. A sector view may help direct that investigation, but a move between overall and sector ranks needs its own methodological explanation.

For the battery-recycling project, the next useful step is a conversation about methods, people, access, and a bounded first piece of work. The ranking earns its place if it helps locate that conversation. The selection decision should be supported by what the conversation and subsequent evidence establish.

Sources and scope

This note reads two peer-reviewed papers and the ICI methodology captured on 11 October 2026, before the scheduled release of its ranking results. It evaluates the inference from a published method to a partner-selection decision. It contains no analysis of unreleased ranks, independent audit of ICI data, or estimate of partnership success.

References

1. Tijssen, R. J. W., Yegros-Yegros, A., & Winnink, J. J. (2016). University-industry R&D linkage metrics: validity and applicability in world university rankings. Scientometrics, 109(2), 677–696. https://doi.org/10.1007/s11192-016-2098-8
2. Vernon, M. M., Balas, E. A., & Momani, S. (2018). Are university rankings useful to improve research? A systematic review. PLOS ONE, 13(3), e0193762. https://doi.org/10.1371/journal.pone.0193762
3. Global Research Partnerships. (2026). Industry Collaboration Index: Methodology, Edition 1. Snapshot retrieved 11 October 2026. Publisher methodology