Types of bias

Association and causation

• Association: Two variables appear related (e.g. more coffee = more energy).

• Causation: One variable directly affects another (e.g. smoking → lung cancer).

  • Only experimental designs can suggest causation.

Systematic bias

• Anything that inaccurately influences the conclusions drawn in a study

• Can be controlled for methodologically

Selection bias

• Happens when participants are not randomly selected or groups aren’t comparable.

  E.g. Letting patients choose whether they want the new treatment or not.

Use randomisation and clear inclusion/exclusion criteria

Sampling bias

• Certain types of people are more likely to be included.

• Only healthy volunteers sign up for a study—results don’t apply to sick people.

Use a representative, random sample. Avoid only recruiting “easy” participants.

Allocation Bias

• Researcher improperly assigns participants to groups

• Placing sicker patients in control group

Use allocation concealment (don't let researchers choose who goes where).

Performance Bias

• Differences in care between groups other than the intervention.

  E.g. Control group finds out they didn’t get the real drug and seeks extra help.

Use blinding of participants and healthcare providers.

attrition bias

• Results are affected because people drop out of the study.

  E.g. People with side effects stop treatment, skewing the results.

Do intention-to-treat analysis (analyze everyone as originally assigned).

Measurement (information bias)

• Inaccurate measurement of outcome

  E.g. Faulty blood pressure monitor gives wrong readings.

Use validated tools, train researchers, and calibrate equipment.

Detection bias

• Outcome assessor is influenced by knowing which group participant is in.

  E.g. A doctor knows the patient got the treatment and looks harder for improvement.

Blind the assessors when possible.

Recall bias

• Participants don’t remember past events accurately.
  E.g. Parents overreport autism symptoms after MMR vaccine due to media scare.

Use objective records if possible (e.g., medical files instead of memory).

Response Bias

• Participants answer the way they think the researcher wants.
  E.g. Patient lies about exercising every day in a diary.

Make surveys anonymous; ask neutral questions.

confounding factors

• A third variable affects both the exposure and the outcome

  E.g. People who drink coffee also smoke more → smoking, not coffee, causes heart disease.

Use randomisation, matching, or statistical adjustments (e.g., regression).