Small Sample Sizes
Few animals or biological replicates → Low power & unstable estimates
No Power Calculation
Sample size chosen arbitrarily → Study may be underpowered
P-values Just Below 0.05
Results clustering around p = 0.04–0.049 → Possible p-hacking or selective analysis
Multiple Outcomes Without Correction
Many biomarkers or endpoints tested → Higher false-positive rate without correction
Pseudoreplication
Cells or slices counted as true biological replicates → True experimental unit ignored (often the animal)
Missing Randomisation Details
“Randomised” without explaining how → Risk of selection bias
No Blinding
esearchers know treatment groups → Observer bias in measurements
Extremely Large Effect Sizes
Dramatic improvements → Often shrink in replication studies
No Replication, Excessively Controlling Experimental Conditions
Only one experiment or cohort → Findings may be unstable
Overstandardisation → Limit scope of findings, limit reproducibility
Suspicious Graphs
Bar plots without raw data points → Distribution and variability hidden
Reliable preclinical science requires randomisation, blinding, adequate power, and transparent reporting.
Avoid costly statistical pitfalls by consulting an experienced biostatistician at the study design stage. We provide biostatistics mentoring for biotech and medtech teams, drawing on 35+ years of experience.
https://lnkd.in/eT6DdMre
