The Erratic Dose-Response Run
A doctoral candidate in life sciences is running a dose-response experiment on a plant-derived extract against a bacterial strain. The first run shows a clean minimum inhibitory concentration at 250 µg/mL. A repeat run, using a newly opened bottle of the same reagent grade, produces erratic readings: 125 µg/mL shows no inhibition, 250 µg/mL shows complete inhibition, and 500 µg/mL shows partial inhibition. The candidate has a conference deadline in three weeks and proposes dropping the erratic wells as 'outliers' and reporting only the first run. The supervisor notices that the new bottle's lot number differs and that the incubator log shows a two-hour power interruption during the second run. The candidate argues that repeating again will delay the abstract and that 'everyone smooths data like this.' The laboratory has a signed data-integrity policy requiring raw records for five years and a co-author at another institute who has not seen the second run.
1. Is dropping the erratic wells as outliers defensible? Justify your answer using principles of scientific inference and data integrity.
Read the worked answer
Conclusion: not defensible as proposed. Outlier removal is legitimate only when there is a documented assignable cause that is independent of the treatment effect and when the rule is fixed before analysis. Here the two anomalies - a new reagent lot and a recorded power interruption - are plausible confounders, not random noise. Dropping the wells would (i) conceal a failed replication, (ii) inflate the apparent potency of the extract, and (iii) violate the laboratory's raw-record policy. The argument that 'everyone smooths data' is a precedent claim, not a scientific justification. A defensible route is to treat the second run as an invalid assay rather than as data to be trimmed, and to repeat the full series with both lots under uninterrupted incubation. If the first result is reported at the conference, it must be labelled preliminary, with sample size and conditions stated. The co-author must see both runs before any abstract is submitted.
2. What should the supervisor do in the next ten working days, and what should be communicated to the co-author and the conference organisers?
Read the worked answer
The supervisor should act in this order: (1) quarantine the new reagent lot and record both lot numbers; (2) rerun the full dose series in parallel with old and new lots, including a no-template control and a viability control; (3) inspect and repair the incubator, and attach the temperature/power log to the record; (4) document the failed run and the interruption in the lab notebook; (5) hold a supervision meeting that separates deadline pressure from the decision. For the conference, submit only if the repeat is concordant with the first run; report both runs, with the failed run in supplementary material. If the repeat is discordant, withdraw the abstract or convert it into a methods poster. The co-author should receive the raw data and a short written note explaining the discrepancy. The supervisor must not sign off on a 'clean' figure assembled by dropping wells.
3. If the candidate insists on publishing the first run alone, what institutional and publication consequences should the supervisor explain?
Read the worked answer
The supervisor should state that the manuscript would misrepresent the evidence and that the laboratory's data-integrity policy requires unresolved concerns to be referred to the institutional research integrity officer. The co-author must be informed because authorship carries accountability for the data. If the paper were published and the omission later discovered, likely consequences include correction or retraction, loss of credibility for the whole group, and possible misconduct proceedings. The candidate should also be told that a preliminary, clearly labelled report is not a career failure, whereas a retraction is. The teaching point is that a deadline does not convert a confounded assay into an outlier problem; the correct response to a failed replication is investigation, not deletion.