Scientific Reasoning and Data Interpretation
Scope
This section gives a short study description, table or graph and asks what can legitimately be concluded. It tests study design, bias, screening statistics and numerical interpretation rather than memorised facts.
Core rules
- Design and causality: randomised controlled trial > prospective cohort > case–control > cross-sectional > ecological. Only designs measuring exposure before outcome can support temporality.
- Effect measures: case–control studies yield an odds ratio, not incidence; cohort studies yield a risk ratio.
- 2×2 screening metrics: sensitivity = TP/(TP + FN); specificity = TN/(TN + FP); positive predictive value = TP/(TP + FP). PPV depends on prevalence; sensitivity and specificity do not.
- Bias: selection (who entered the study), information or recall, and confounding (a third factor linked to both exposure and outcome). Randomisation and adjustment address confounding; blinding addresses observer bias.
- Statistics: a p-value is the probability of the data under the null hypothesis, not the size or importance of an effect; confidence intervals show precision.
Worked example with reasoning
10,000 people are screened; prevalence is 2%, sensitivity 90%, specificity 95%. Affected = 200, so TP = 180 and FN = 20. Unaffected = 9,800, so FP = 5% × 9,800 = 490 and TN = 9,310. PPV = 180/(180 + 490) = 180/670 ≈ 26.9%. Even with good test characteristics, three of four positives are false when disease is rare.
Traps that cost marks
- Reporting sensitivity when the question asks for predictive value.
- Ignoring base rates in screening scenarios.
- Claiming causation from a cross-sectional survey.
- Reading an axis as counts when it is a rate per 100,000.
- Equating "not statistically significant" with "no effect".
Transfer tip
Write the denominator you are using before answering; most errors in this section are denominator errors.