The 0.06 That Would Not Cooperate
Dr. R. supervises Ms. K., a full-time PhD scholar whose project - funded by a ₹8 lakh grant, of which a diagnostics firm contributes ₹3 lakh - tests whether a 12-session counselling module reduces relapse in 84 outpatients. The primary outcome returns p = 0.06 (two-tailed). In a meeting, Ms. K. proposes three changes before submission: exclude three participants who 'did not really engage' with the module; re-analyse the primary outcome as one-tailed because the direction was stated in her synopsis; and lead the paper with a secondary outcome that did reach p < 0.05. Her fellowship renewal is due in six weeks and requires at least one publication, ideally in the conference proceedings published by the co-funding firm. Dr. R. also chairs the departmental research committee that must clear Ms. K.'s six-month extension. The scholar says, 'Everyone does this; otherwise three years of work is wasted.'
1. Assess each of the three proposed changes. Which, if any, can be defended, and under what conditions? Give the statistical and ethical reasoning, not just a verdict.
Read the worked answer
Separate what is arguable from what is not.
Excluding the three 'non-engagers'. Not defensible as the primary analysis. In a randomised comparison the primary analysis must be intention-to-treat: everyone randomised is analysed as allocated. Engagement is plausibly a consequence of the treatment, or of relapse, so conditioning on it removes precisely the people in whom the module failed and inflates the estimate. It becomes defensible only if the exclusion rule was written down before data lock, applied without sight of outcomes, and rests on a reason unrelated to treatment (such as a documented diagnostic error). Even then it is a sensitivity or per-protocol analysis reported alongside ITT - never instead of it.
Switching to one-tailed. Partly arguable, currently misused. A pre-specified direction is a legitimate justification, but a synopsis is not an analysis plan, and the decision is being taken after seeing p = 0.06. It is defensible only if a dated pre-specification exists, both one- and two-tailed results are reported, and the switch is disclosed in the paper. As proposed, a reviewer will read it as a rescue attempt.
Leading with the significant secondary outcome. Not defensible. Promoting a secondary outcome to the headline after the primary missed is outcome switching. The primary estimate (0.06 with its confidence interval) must be reported first and in full; the secondary finding may be reported as pre-specified and explicitly labelled exploratory.
The honest conclusion is that the trial is inconclusive - a finding, not a failure.
2. What should Dr. R. do about authorship, the funder-published proceedings and her own dual role?
Read the worked answer
Authorship. A funder cannot be an author merely because it paid. Authorship requires a substantial intellectual contribution, involvement in drafting or critical revision, final approval, and public accountability for the work; money and a publishing platform meet none of these. If the firm's staff did contribute, that must be documented and declared; if not, the firm is acknowledged and named as a funder in the competing-interests statement. The proceedings being published by the co-funder creates a guest/ghost authorship and capture risk that must be disclosed to the conference and to the university.
Funder review. The contract should permit the funder to check the manuscript only for confidential or proprietary information, with a written undertaking that publication cannot be vetoed and that the results will be published whatever their direction. Without that clause in place, Dr. R. should not let the paper go to the funder's venue.
The dual role. As chair, Dr. R. would adjudicate an extension partly caused by an inconvenient analysis. She should declare the conflict and recuse herself from that item, with another member chairing; if recusal is impractical, the disclosure and the reasons for the decision must be minuted. She should also take the fellowship clause - renewal contingent on a publication - to the research dean, because it is an institutional inducement to the very practices she is being asked to prevent.
Do not bargain. Ms. K. must not be led into trading authorship for a favourable review.
3. Draft a six-week plan that gives the scholar a legitimate output without corrupting the record.
Read the worked answer
Week 1 - lock and pre-specify. Freeze the dataset, write a dated statistical analysis plan, deposit it with the doctoral committee, and bring in an independent statistician who has not seen the outcome tables.
Week 2 - analyse honestly. Intention-to-treat as primary, with a 95% confidence interval; sensitivity analyses (per-protocol, multiple imputation for missing data); all pre-specified outcomes tabulated, significant or not.
Week 3 - write the full report. Draft the paper with the null result as the finding, plus a limitations section that states the study was powered for a larger effect. This is publishable; journals and registered-report formats exist precisely for it.
Week 4 - take the legitimate short output. The conference abstract should be on the intervention's feasibility and the outcome-measurement protocol, not on a claim of efficacy. A preprint secures priority and gives the funder something citable.
Week 5 - extension. Apply with the analysis plan attached, showing what the extra six months buys (follow-up completion, not a re-run of the test).
Week 6 - internal review and submission. Send the funder the timeline now, not the result. Early expectation-setting is what prevents the pressure from resurfacing at submission.