Human Subjects Research
High-Yield Summary
- The experimental approach manipulates a variable directly, using randomization, blinding, and statistical adjustment to isolate its effect.
- The observational approach examines exposures and outcomes without manipulation: cohort (prospective), cross-sectional (single point in time), case-control (retrospective, best for rare outcomes).
- Hill's criteria help judge whether an observational association is likely causal — but observational findings should still be called correlations, not proven causes.
- Selection bias, detection bias, observation (Hawthorne) bias, and confounding are four distinct ways human subjects research gets distorted.
Ensuring Reliable Results in Experiments
- Randomization
- Randomly assigns participants to treatment/control to evenly distribute confounders — so any effect is more likely due to the treatment itself.
- Single-blind
- Participants don't know their group, minimizing the placebo effect.
- Double-blind
- Neither participants nor researchers know group assignment, further reducing researcher bias.
- Statistical adjustment
- Regression models etc. used to adjust for confounders like age or baseline health differences between groups.
Observational Study Types
| Study type | Definition |
|---|---|
| Cohort (prospective) | Follows a group forward in time to assess outcome rate by exposure — good for establishing temporality (e.g., tracking smokers vs. non-smokers for 20 years) |
| Cross-sectional | Assesses exposure and outcome at a single point in time — good for measuring prevalence |
| Case-control (retrospective) | Compares exposure history between those with the outcome (cases) and without (controls) — best for rare outcomes |
Hill's Criteria for Causality (Observational Studies)
- 1Temporality — the exposure must occur before the outcome.
- 2Strength — a stronger association better supports causality.
- 3Dose-response relationship — more exposure → greater/more severe outcome.
- 4Consistency — the association holds across multiple studies and populations.
- 5Plausibility — there's a biologically reasonable mechanism.
- 6Specificity — a specific exposure links to a specific outcome.
- 7Coherence — the association fits with existing knowledge.
- 8Experiment — experimental evidence, where possible, strengthens the causal link.
- 9Consideration of alternative explanations (a.k.a. "analogy" in Hill's original 9 viewpoints) — rule out confounders and other explanations.
Bias Types
- Selection bias
- The sample isn't representative of the population (e.g., recruiting only healthy volunteers for a chronic-illness study).
- Detection bias
- Outcomes are searched for inconsistently across groups (e.g., screening smokers for lung cancer more than non-smokers).
- Observation bias (Hawthorne effect)
- Participants change behavior because they know they're being watched (e.g., eating healthier while monitored).
- Confounding
- A third variable is linked to both exposure and outcome, creating a false association (e.g., ice cream sales and drowning both driven by hot weather).
Common MCAT Trap
- Even a strong observational association that satisfies several of Hill's criteria is still only a correlation — Hill's criteria raise confidence, they don't prove causation the way a controlled experiment can.
- Don't mix up detection bias (searching for the outcome unevenly) with selection bias (the sample itself is unrepresentative) — one is about who's enrolled, the other about how outcomes are measured after enrollment.
Quick Recall
A researcher compares smoking history in current lung cancer patients vs. cancer-free controls. What study type is this?
Diet-study participants eat healthier simply because they know they're being watched. Which bias is this?