Three conflicting studies: build an evidence verdict
The hardest expert task is not choosing the most positive study. It is weighing consistency, methods, population, effect size and uncertainty when papers disagree.
Build an evidence table before a verdict
Record for each study the population, randomisation, comparator, formulation, primary outcome, follow-up, attrition, effect size, uncertainty and adverse events. Comparing only sample sizes or headlines misses what the trial measured. Keep topical cosmetics and prescription medicines in separate evidence categories.
Three conflicting studies: build an evidence verdict — a published visual reference for the topic, not SESH product-test data.
Open full-size image ↗More evidence is not a simple vote
Two small uncontrolled studies agreeing may be weaker than one rigorous controlled trial, but a single high-quality trial can still be imprecise or atypical. A systematic review should address why studies differ, whether outcomes are comparable and whether combining them is sensible. Forest plots display estimates and intervals; they do not automatically prove a result.
Distinguish strength from certainty
A large reported change from a fragile design may inspire follow-up, not an immediate promise. Assess bias, indirectness, inconsistency, imprecision and possible publication bias. Translate evidence into calibrated language: 'suggests', 'is supported in this population' or 'insufficient to conclude'.
Use conflicting results to improve questions
When studies conflict, ask whether active forms, concentrations, regimen or baseline skin differences explain it. Sometimes research genuinely remains inconclusive; that is an acceptable scientific endpoint. A brand must not suppress less favourable results while citing one shiny chart.
These curves show how changing mean and variance affects a statistical model. They are theoretical teaching data, not skin or fragrance-panel results.
Open full-size image ↗Evidence lab · hypothetical data
Hypothetical: A (n=18, uncontrolled) finds 40% visual improvement; B (n=180, randomised) finds a 4% between-group difference with narrow uncertainty; C (n=60) finds no clear difference and wide uncertainty. How would you summarise?
Defensible interpretation
Do not average headline percentages. Give most inferential weight to the valid comparison, note C's imprecision and A's bias, and state whether 4% is clinically meaningful. The studies do not demonstrate 40% benefit attributable to the active.
Research references
Educational material, not medical advice or a product promise.
