Table 1 Strengths and inherent causal biases of different study designs

From: Evidence triangulator: using large language models to extract and synthesize causal evidence across study designs

Study Design

Key Strengths

Main Potential Biases

Implications for Causal Inference

Randomized Controlled Trials (RCTs)

- Gold standard for causality due to randomization;

- Rigorous control of intervention and outcome measurement

- Restricted generalizability;

- Attrition bias (dropout);

- Short-term follow-up limits capturing long-term effects

- High internal validity;

- May not reflect real-world applicability if study population is unrepresentative

Observational Studies (OS)

- Large and diverse cohorts;

- Longitudinal tracking of real-world behaviour;

- Often captures rare outcomes or exposures

- Confounding bias due to non-random exposure assignment;

- Selection bias;

- Reverse causation

- Broad external validity;

- Vulnerable to systematic biases if confounding variables and study populations are not carefully managed

Mendelian Randomization (MR)

- Reduces confounding and reverse causation via genetic instruments;

- Biological plausibility tests for causal effects

- Horizontal pleiotropy (genetic variant affects multiple traits);

- Population stratification;

- Weak instrument bias

- Offers quasi-experimental conditions;

- Violations of core assumptions can undermine reliability of the causal estimate

Back to article page