flowchart LR A["<b>Consultant</b><br/><i>transactional</i><br/>acknowledgment"] B["<b>Collaborator</b><br/><i>sustained</i><br/>authorship"] C["<b>Co-investigator</b><br/><i>named on the grant</i><br/>authorship, effort"] D["<b>Principal investigator</b><br/><i>leads the science</i><br/>first or last author"] A --> B --> C --> D E["authority, time, credit all increase"] -.-> D
4 Team Science for Biostatisticians
The best thing about being a statistician is that you get to play in everyone’s backyard.
John W. Tukey, attributed (c. 1962)
NoteSources
Adapted from author’s lecture notes and supporting materials for a graduate practicum in biostatistics.
4.1 Prerequisites
Answer the following questions to see if you can bypass this chapter. You can find the answers at the end of the chapter in Section 4.17.
- What distinguishes a ‘collaborative’ biostatistician role from a ‘consulting’ role?
- Name three communication skills essential for biostatisticians on interdisciplinary research teams.
- Why does team science succeed or fail on clear expectations about authorship and credit?
4.2 Learning objectives
By the end of this chapter you should be able to:
- Describe the spectrum of biostatistician involvement (consultant, collaborator, co-investigator, PI).
- Explain ICMJE authorship criteria and apply them to a hypothetical paper.
- Hold a structured intake conversation with a new clinical collaborator.
- Write a short scope-of-work document that pre-empts common misunderstandings.
- Maintain a legible portfolio of simultaneous projects and triage which one to advance when several compete for the same day.
- Communicate statistical results to non-statisticians without misrepresenting uncertainty or effect size.
- Recognize common collaboration failure modes (scope creep, last-minute pre-registration, p-hacking pressure) and respond appropriately.
4.3 Orientation
Most biostatisticians spend the greater part of their careers on teams alongside people who cannot read their code and do not share their vocabulary. Effective collaboration across this gap is, we would insist, a learned skill and not a personality trait, and the distinction matters because skills can be taught and temperaments cannot. A statistician described as ‘difficult to work with’ has usually not been rude to anyone. They have failed to establish, in writing and early, what was being asked of them and what they would deliver, and every subsequent disagreement has been argued without a shared record of what was agreed.
That is the shape of nearly every collaboration failure in this chapter: not a clash of personalities but an absent document. This chapter takes up the norms, the conversations, and the documents that prevent the absence.
The chapter is unusually narrative: the technical skills elsewhere in the book are only useful if you can deploy them in a team context, and the team context is governed by social and procedural conventions more than by software.
4.4 The statistician’s contribution
Team science is judgment work end to end.
Recognize the role you are being asked to play. A clinical PI who says ‘we need a statistician’ may be asking for a methodological consult, a hands-on collaborator, a co-investigator on the grant, or a hired labor for the analysis section of a manuscript. The expectations, the timeline, the authorship implications, and the appropriate level of pushback are different for each. Failing to clarify the role before doing the work produces friction at the end.
Pick your battles. Not every methodological choice is worth a confrontation. A non-default reference level in a logistic regression: matter of taste; let it go. A proposal to dichotomize a continuous outcome to make the analysis ‘simpler’: fight, because it loses information and biases inference. Distinguishing preference from principle is what makes you a useful collaborator rather than a pedantic one.
Calibrate your language to the audience. A clinical collaborator does not need to know the difference between Wald and likelihood-ratio confidence intervals. They do need to know whether the treatment effect is clinically significant and how confident you are in that claim. Translating between these registers is a core skill.
Document everything that has been agreed. Verbal agreements about scope, deadlines, authorship, and analytical decisions evaporate. Email summaries after every meeting; written analysis plans before any analysis runs. The document is what you defend when expectations diverge later.
These habits separate biostatisticians who are re-engaged for the next project from those who are not.
4.5 The collaboration spectrum
A useful mental model: biostatistician involvement varies along a spectrum from consulting (transactional) to co-investigation (long-term). Figure 4.1 sets the four common positions against the three things that vary with them, because it is the mismatch between where you sit and where the investigator thinks you sit that produces most collaboration friction.
Common positions:
Consultant. A clinical investigator brings a question or a dataset; the statistician provides analysis or methodological advice; the relationship ends or recurs transactionally. Common in academic statistical consulting services and industry consulting groups. Usually does not produce authorship; deliverables are the analysis or the report.
Collaborator. Sustained involvement across study design, analysis, and writing; the statistician is substantively involved with the science. Authorship is expected when ICMJE criteria are met (below). Most academic biostatistics happens here.
Co-investigator. Named on the grant, with allocated effort; partial authority over the project direction; deeply involved in study design. Usually mid-career and senior biostatisticians on funded research.
Principal Investigator. Leads the project. The biostatistical method is itself the science. Methodological PIs in biostatistics; less common as the lead role in clinical research.
The position determines the appropriate level of authority, the time commitment, and the credit. A common source of tension is mismatch: a statistician treated as a consultant who sees themselves as a collaborator, or vice versa. Clarifying which position is being filled is the first step in avoiding it.
4.7 The intake conversation
The first meeting with a new clinical collaborator sets the trajectory of the collaboration, and it does so mostly through what is not asked. An investigator arrives with a dataset and a question, the statistician takes the work, and the terms nobody raised, what the deliverable is, when it is due, whether there will be authorship, become assumptions that each party holds privately and differently. They surface months later, during the only week in which everyone is busy.
The remedy is a structured intake, which is less formal than it sounds: a fixed set of questions asked in the first conversation and written down afterward. At minimum it covers:
The scientific question. What are they trying to find out? What would the answer enable? Push past ‘is treatment A better than B’ to ‘better in what sense, for which patients, in what setting’. The question determines the design.
The data. What data exist; how they were collected; what can and cannot be shared; whether the data exist yet or are to be collected. This is where most projects’ fate is decided.
The deadline. A grant deadline three months out is different from an exploratory question with no deadline. The deadline shapes what is feasible.
The authorship and acknowledgment expectation. Will the statistician be a co-author? An acknowledgee? Hired help? Establishing this at intake avoids the confrontation later.
The deliverables. A two-page memo, a Quarto report, a published paper, a regulatory submission? The format determines the time commitment.
Red flags. A request to perform a specific test (‘we want a chi-square’) without explaining the hypothesis. A dataset that has already been analyzed multiple times by the team (‘it didn’t show what we hoped’). Pressure to produce results by next week. Vague answers about IRB or data-use agreements. Each warrants more questions before agreement.
A standard intake document, one page, ten questions, formalizes this conversation. Sending it before the meeting often resolves half the issues without the meeting.
4.8 Scope-of-work documents
For any non-trivial collaboration, a written scope of work (SOW) defines:
- Inputs. What data the collaborator will provide, by when, in what format.
- Outputs. What the statistician will produce (analysis report, table for a paper, full manuscript section), by when.
- Methodology. The high-level analytic approach (pre-registered or to-be-pre-registered).
- Revisions. How many rounds of revision are included; what counts as a ‘major’ change requiring renegotiation.
- Authorship and acknowledgment. As discussed at intake.
- Boundaries. What the statistician will not do (e.g., implement methods outside their expertise, perform the literature review).
The SOW is not a legal contract; it is a shared understanding. When the project drifts (as projects do), the SOW is the document everyone returns to.
4.9 Managing a portfolio of projects
The intake conversation and the scope of work govern a single collaboration. The defining structural fact of a collaborative biostatistician’s work is that there is never only one. You will carry five to fifteen active projects at once, at different stages, for different investigators, each of whom is convinced theirs is the one you are working on today. Survey evidence ranks managing multiple simultaneous projects among the most essential skills in the work, second only to coding for those in their first five years (Slade et al., 2023). It is also the skill this book has so far said least about, and the one no software installs for you.
The unit of work is the portfolio, not the project. A junior statistician experiences the job as a queue: finish project A, then start project B. The job is not a queue. Projects arrive on the investigators’ schedules, block on data and approvals you do not control, and then unblock several at once. The competence being asked of you is to hold many partially finished projects in a legible state and move each when it is movable, not to drive one to completion while the rest quietly decay.
Estimate, then double. Statisticians are systematically optimistic about how long an analysis takes, because the estimate imagines the clean path and the work is mostly the mess: the merge that reveals duplicated IDs, the outcome coded three ways across sites, the model that will not converge. Quote timelines with the mess included. A serviceable discipline is to estimate the clean path, double it, and add the one review cycle you forgot. Promising month 2 and delivering in month 2 builds more standing than promising month 1 and delivering in month 3.
Make blocking visible, to yourself and to them. Much of a statistician’s calendar is spent not working but waiting: for the data extract, for the IRB amendment, for the collaborator to say which of two outcomes is primary. A project blocked on someone else is not one you have finished thinking about; it is a debt that comes due, usually at the worst moment. For every active project, track the single thing it is waiting on and who owns it. When a PI asks why their analysis has not moved, ‘I am waiting on the corrected data file I requested on the 3rd’ is an answer; silence is not.
Protect the cost of re-entry. The expensive part of juggling projects is not the analysis; it is the reconstruction paid each time you return to a project you set down three weeks ago and must remember what you were doing and why. This book has, without naming it, spent several chapters lowering that cost. A project that rebuilds with one command (Chapter 10, Chapter 13), whose analytic decisions are fixed in an analysis plan (Chapter 24), and whose history is legible in Git (Chapter 7) can be put down and picked up in minutes. A project held together in your head and one untracked script cannot. Reproducible infrastructure is usually justified as a courtesy to others, the reviewer and the future replicator; its quieter and more immediate payoff is that it is what lets you carry fifteen projects without dropping any.
Batch the switches you can; price the ones you cannot. Context switching is not free, and the statistician who answers every email the instant it arrives pays the re-entry cost a hundred times a day. Where the work allows, batch it: a block for one project’s analysis, a separate block for correspondence across all projects, rather than interleaving at the grain of the inbox. A few switches are forced by genuine urgency and earn their cost; most are not, and the skill is telling the two apart.
Capacity is finite, and saying so is part of the job. The boundaries clause of the scope of work declines work within a project. The same discipline operates across projects: a statistician who never says ‘I can start this in three weeks, not this week’ ends up doing everything late rather than some things on time. Deferring is not a failure of collaboration. Committing to a deadline you cannot meet is.
None of this requires a particular tool. A single list of active projects, each with its current stage, its next action, and the thing it is blocked on, reviewed once a week, is enough. The tool is not the skill; the habit of keeping the portfolio legible is.
4.10 Communicating statistical results
Translating statistical output for a clinical audience is not a matter of simplifying it. It is a matter of reporting the quantity the listener is actually going to act on, which is rarely the quantity the software puts at the top of its output. Four principles follow from that, and they are ordered from the one that changes the most to the one that changes the least:
Effect size first, p-value last. ‘The treatment reduced HbA1c by 0.6 percentage points (95% CI 0.3 to 0.9)’ communicates more than ‘the effect was significant at p = 0.002’. P-values are unintuitive and easily misinterpreted; effect sizes with uncertainty are not.
Plots over tables. A coefficient plot with CIs conveys the same information as a regression table, faster, and is harder to misread.
Calibrated language. Distinguish ‘we found \(X\)’ from ‘the data are consistent with \(X\)’ from ‘we cannot rule out \(X\)’. The verbs matter; the precision matters.
Specific examples. ‘For a 60-year-old patient with baseline HbA1c of 8%, the model predicts a reduction to about 7.4% under treatment, with a 95% CI from 7.1 to 7.7%.’ This is far more useful than the regression coefficient and SE that produce it.
For papers and reports, write the substantive conclusion in plain language and include the quantitative support. For meetings, lead with the substantive conclusion, then the support if asked. The audience determines the order.
4.11 Handling disagreement
When you and the collaborator disagree on a methodological choice, the path forward depends on what is at stake.
Preference. Coding style, default vs. non-default reference level, plotting choices. Defer to the collaborator unless there is a substantive reason not to.
Best practice. Multiple-testing correction, intention-to-treat analysis, missing-data handling. Argue for the standard practice; if overruled, document the dissent in the methods.
Validity. Inappropriate test, data dredging, post-hoc hypothesis selection, dichotomizing continuous outcomes. Push back firmly. If the disagreement persists and the analysis is methodologically wrong, withdraw.
The middle category is where most disagreements live and where judgment is most consequential. Picking too many fights makes you a difficult collaborator; picking none makes you complicit in bad analyses.
4.12 Worked example: the intake document
Project. Effect of post-discharge home health visits on 30-day readmission in heart-failure patients.
Investigator. Dr. X, Cardiology.
Question. Does receiving at least one home health visit within 7 days of discharge reduce 30-day all-cause readmission?
Data. Retrospective EHR cohort, \(n \approx 5{,}000\), obtained from the institutional research data warehouse under existing IRB. Includes demographics, baseline ejection fraction, medication on discharge, indicator for home health use, and outcome.
Deadline. Manuscript draft by month 4; submit month 6.
Methodology (to be confirmed). Propensity- matched cohort with multivariable logistic regression for the primary outcome. Sensitivity analyses with inverse-probability weighting.
Deliverables. Pre-registered analysis plan (month 1), Table 1 and primary results (month 2), manuscript draft (month 4), revisions (month 5).
Authorship. Statistician will be second author. Will be involved in writing methods and results, and revising the full manuscript.
Boundaries. Statistician will not write the introduction or discussion. Statistician will not perform additional analyses requested after month 4 without renegotiation.
Red flags noted. None at intake. To revisit if EHR data quality is worse than expected at first exploration.
This document, agreed in writing at the start, makes the project run smoothly even when the data turn out to be messier than expected.
4.13 Collaborating with an LLM on team-science work
LLMs can help with the writing parts (drafting intake documents, proofreading methods, role-playing skeptical PIs). They cannot replace the judgment.
Prompt 1: drafting an intake document. Describe the project briefly and ask: ‘draft a one-page intake document that I will send to a new clinical collaborator before our first meeting.’
What to watch for. The output will likely be generic and reasonable. Customize it for the specific domain (oncology has different baseline norms than psychiatry).
Verification. Send it to a colleague who has run similar projects; iterate.
Prompt 2: red-flag detection. Paste a short description of a request from a collaborator and ask: ‘are there red flags in this request?’
What to watch for. The LLM is good at flagging explicit issues (no IRB, no analysis plan, vague hypotheses). It is weaker on subtle issues (data has been silently inspected by the team and the ‘hypothesis’ is post-hoc).
Verification. Bring suspicions to a senior biostatistician; their experience patterns the gut calibration.
Prompt 3: drafting an ICMJE contribution statement. Paste author names and roles, ask: ‘draft a CRediT-format contribution statement.’
What to watch for. The output should map roles to ICMJE criteria. Each contribution should be specific and meaningful; vague statements (‘contributed to the project’) are red flags for ghost authorship.
Verification. Have each named author confirm the description is accurate before submission.
4.14 Principle in use
Three habits keep expectations from diverging:
- Document agreements at the time, not at the end. Email summaries after every meeting; written SOWs and intake documents.
- Distinguish preference from principle. Pick the methodological battles that affect validity; defer on the rest.
- Translate between registers. Effect-size language for clinical readers; technical language for methods sections; calibrated uncertainty language for both.
4.15 Exercises
- Take a paper you have contributed to and draft a CRediT-format contribution statement for every author, based only on the paper’s content.
- Write a one-page intake document you would share with a new clinical collaborator before the first meeting.
- Find a published paper whose results are overstated relative to the evidence. Write a two-paragraph critique suitable for a peer review.
- Role-play a difficult conversation: a senior PI asks you to dichotomize a continuous outcome to produce an OR for a press release. Draft your response.
- Audit your own most recent project: which of the ICMJE criteria did you meet? Was your authorship position appropriate?
- List every project you currently have active. For each, write its stage, its single next action, and the one thing it is blocked on together with who owns that thing. Any project for which you cannot name a next action is the one in trouble; say why.
4.16 Further reading
- (Slade et al., 2023), essential team-science skills for biostatisticians.
- (International Committee of Medical Journal Editors, 2024), the defining reference for authorship norms.
- The CRediT taxonomy at
casrai.org/credit, the contribution-role vocabulary for finer-grained attribution. - Working with collaborators in clinical research at the National Academies website, a useful practical guide.
4.17 Prerequisites answers
- Consulting is typically transactional: a question is brought, a statistical analysis is delivered, and the relationship ends. Collaboration is sustained: the statistician is a team member involved across study design, data collection, analysis, and manuscript writing, and shares authorship responsibility for the final product. The middle ground (a few sustained interactions that do not span the whole project) is where most role disputes happen.
- Active listening (understanding the scientific question before proposing a method), precise writing (producing analysis plans and memos that non-statisticians can audit), and calibrated language (distinguishing ‘verified’, ‘likely’, ‘assumed’, and ‘unknown’ in reports). Many biostatisticians’ technical skill outpaces their communication; the gap shows up as ‘difficult to work with’.
- Authorship determines credit in an academic system where credit allocates grants, promotions, and prestige. Unresolved authorship disputes sour collaborations and sometimes terminate them. ICMJE-style agreement at the outset (and written down) avoids the confrontation at the end. The conversation is uncomfortable up front and far more uncomfortable later; have it early.