Reading focus
Syllabus: What Would Justify a Causal Claim?
Use the criteria for triangulation to compare causal questions and likely sources of bias across designs.
Why you’re reading it. Build a cumulative argument without counting significant findings or assuming different methods have independent errors.
A question to consider. For the effect of family income on children’s symptoms, propose two approaches with meaningfully different weaknesses; what agreement or disagreement would change your judgment?
Why this reading
Lawlor and colleagues propose comparing approaches whose main sources of bias differ. Their worked examples make agreement informative only after examining what each approach estimates and why its errors might differ from the others.
Before you read
Bias is a systematic distortion of an estimate. Triangulation brings approaches with different assumptions to bear on a common causal question. A larger number of studies does not necessarily provide more independent evidence if they share the same weakness.
Read for
- Use the proposed criteria to compare the causal question, timing of exposure, likely biases and expected direction of those biases.
- Examine why estimates from different exposure durations need not have the same magnitude.
- Ask whether apparently different methods can still share an important source of bias.
Consider afterwards
- Two cohorts and an experiment all find that an intervention is beneficial. What would you examine before treating that agreement as stronger causal evidence?
Editorial context
This is a methodological proposal illustrated with epidemiological examples. It offers a way to organize a causal argument, not a mechanical vote or a guarantee that biases cancel. The examples depend on assumptions about exposure duration, measurement and the comparability of questions. Claims about those examples should not be generalized to every population or treated as current clinical recommendations. Treur and colleagues’ 2024 planning guide extends the discussion to prospective study design.