Do I need a written agreement for an informal collaboration?

For two academic groups sharing analysis with no money moving and no personal data, a written summary of what you agreed is normally enough, and it should still exist as an email everyone has seen. The moment funding transfers between institutions, personal data or samples move, or a company is involved, the institutional agreement stops being optional and your research office has to be involved.

How should we decide author order?

Agree the rule rather than the list, and agree it before the work starts. Adopt an explicit standard such as the ICMJE criteria for who qualifies as an author, state what first, last and corresponding author mean in this collaboration since conventions differ by field, and set the points at which the order will be reviewed. Contributions change, and a rule survives that change where a fixed list does not.

Who owns the data in a research collaboration?

Usually an institution rather than an individual, determined by employment terms, funder conditions and the ethics approval under which the data was collected. That is why this is a question to ask your research office rather than to settle between researchers. What you can and should agree between yourselves is access: who can see what, in which form, and for how long.

What should we do if the collaboration breaks down?

Go back to what was written down at the start, which is the main reason to write anything down at all. Deal with the concrete questions separately: authorship for the work actually done, what happens to unpublished data, and whether the paper still goes ahead. If the dispute is about authorship on a submitted paper, journals expect the authors to resolve it, and institutions rather than editors are the body that arbitrates.

When should we have this conversation?

Before any work is done, and ideally on the first call. Everything on this list is easy to discuss while nothing exists and difficult once there is a dataset, a draft and an implied author order. If you are already mid-project and never had the conversation, having it late is still far better than never, and the natural moment is before the first full draft circulates.

The Research Collaboration Discussion Checklist

Five things to settle out loud before the work begins: who does what, how authorship will be decided, who controls the data, what the timeline commits each side to, and what happens when it ends.

This page is a checklist for a conversation, not a contract and not legal advice. Nothing here is a model agreement, and it is not a substitute for one. Where money moves, where personal data or materials are transferred, where a company is involved or where a funder requires it, the written agreement is drafted by your institution's research or legal office, and established models exist for exactly that: the UK Intellectual Property Office publishes the Lambert Toolkit of model collaboration agreements, and EU consortia commonly start from the DESCA model consortium agreement. What follows is the set of questions that decide whether a collaboration goes well, most of which never make it into any contract because they are about expectations rather than obligations.

What this checklist is, and when you need a real agreement instead

Two different things are often called a collaboration agreement, and confusing them causes real harm. One is a legal instrument between institutions. The other is a shared understanding between researchers. You frequently need both, and this page is only the second.

Talk to your research office before the work starts, not after, if any of the following apply: funding transfers between institutions, personal or patient data is shared, materials or samples move, a commercial partner is involved, or intellectual property with commercial potential is likely.

In those cases the instruments have names and are drafted by professionals: a collaboration agreement, a consortium agreement, a data sharing or data transfer agreement, a material transfer agreement, a memorandum of understanding.

Funders often make it mandatory. Most EU collaborative projects require a consortium agreement between the partners before the grant starts, which is why models such as DESCA exist.

A discussion checklist does not replace any of that. What it does is prevent the disputes that contracts do not cover: authorship order, who presents at which conference, who answers the reviewers, and what happens when someone stops replying.

Write the outcome down anyway, even informally. Five lines in an email after the call, agreed by everyone, resolves more future arguments than a signed document nobody reads.

Roles: who does what, and who decides

Start with the work rather than with the credit. Most authorship disputes are really disagreements about contribution that were never made explicit while the work was being planned.

List the tasks, then put a name against each one: study design, data collection, analysis, code, writing, figures, ethics approval, project administration. The CRediT contributor roles taxonomy is a ready-made vocabulary for this and is now an ANSI/NISO standard used by many journals.; Name who leads the project overall, and who is the corresponding author for the resulting paper. They are not always the same person and the difference should be deliberate.; Say who owns each deliverable, meaning who is accountable for it existing on time, not who is allowed to touch it.; Agree how a disagreement about interpretation gets resolved, before you have one. In practice this usually means naming the person whose call it is on a given part of the work.
Be explicit about students: whose student does which part, who supervises that part day to day, and whether the work is expected to appear in a thesis.

Authorship: agree the rule, not the list

Agreeing a fixed author list at the start is fragile, because contributions change. Agreeing the rule by which the list will be decided survives the changes, and it is the conversation almost nobody has at the right time.

Adopt an explicit standard and say so. The ICMJE criteria require substantial contribution, drafting or critical revision, final approval, and accountability for the work, and they are the reference in many fields well beyond medicine.; Decide what first, last and corresponding author mean in this collaboration, because the conventions differ by field and by country and both partners often assume their own.
Say when the list will be revisited: at minimum before the first draft is circulated and again before submission. A contribution that grows or shrinks should move the position.; Decide what happens to people who contribute but do not meet the authorship standard, which normally means a named acknowledgement with their permission.; Agree in advance what happens with a second paper from the same work, since the second paper is where most disputes actually appear.; Put the agreed rule in writing in a shared document. Not a contract, just a paragraph everyone has seen.

Data: ownership, access and sharing

Data is the part where informal collaborations break, because ownership, access and permission to share are three different questions and researchers routinely treat them as one.

Ownership is usually institutional rather than personal, and it is set by your employment terms and your funder, not by agreement between the two of you. Establish who owns what before promising anyone anything.
Access is separate: who can see the raw data, who can see derived data only, and under what conditions. A collaborator who can analyse but not export is a normal and often necessary arrangement.; Consent and ethics set hard limits. If your approval does not cover sharing with a third institution or a third country, no agreement between researchers can create that permission, and an amendment takes time you should plan for.; Funder obligations apply regardless of what you agree between yourselves. The NIH requires a data management and sharing plan for applications generating scientific data under its 2023 policy, and UKRI expects research data to be made openly available with as few restrictions as possible under its open data expectations. Know which apply to each partner, because they may differ.
Agree the practical details that cause friction: where the data lives, who has the credentials, what format, what documentation, and who is responsible for the deposit at publication.; Agree an embargo if one side needs it, with a date rather than a vague understanding, and agree what happens to unpublished data if the collaboration ends before publication.

Timeline and effort: what each side is actually committing

The most common cause of a stalled collaboration is not bad faith, it is two people with different implicit assumptions about pace. Making the assumptions explicit takes ten minutes.

Fix the dates that are real: grant deadlines, ethics submissions, conference abstract dates, the intended submission window for the paper.; Say how much time each side is committing in a form that can be checked, such as one day a week of a named person, or a student for six months.; Agree a meeting cadence and stick to it. Monthly is enough for most projects; ad hoc means nothing happens between crises.; Agree turnaround expectations for drafts, which is where most irritation accumulates. Two weeks for comments on a full draft is a common and workable default.
Say what happens when someone slips: who is told, how soon, and what the fallback is. Slippage is normal and a plan for it is not an accusation.

Exit: how the collaboration ends

Every collaboration ends, most of them uneventfully and a few of them badly. The exit conversation is the one that feels most awkward to have at the start and is worth the most later.

What happens if someone moves institution, which is the most likely ending of all. Does the work follow the person, the institution, or the data?
What happens if a partner withdraws before publication: what they may reuse, what they may not, and whether their contribution to date earns authorship or an acknowledgement.; What happens if you disagree about interpretation at the end. Options range from a joint paper stating both readings to two separate papers, and deciding the mechanism in advance is easier than deciding the outcome later.; What happens to unpublished data and code, and whether either side can use them in a new project without the other.; How to decline a second project without damaging the first. Saying at the outset that this agreement covers one piece of work makes the later no straightforward rather than personal.

Frequently asked questions

No. This page is a checklist for a conversation between researchers and is not legal advice or a model contract. Where money, personal data, materials, intellectual property or a commercial partner are involved, the written agreement is drafted by your institution's research or legal office. Published models exist for those cases, including the UK Intellectual Property Office's Lambert Toolkit and the DESCA model consortium agreement used by EU consortia.