M09.01.11 Bias in study design

Bias in research is the deviation of results from the truth, often affecting the validity of a study. It can arise intentionally or unintentionally and manifests in various forms. A comprehensive understanding of bias is crucial in ensuring accurate and reliable research outcomes.

Key Concepts

Reliability

• Definition: The consistency of a test in measuring an attribute.
• Types:
  • Test-Retest Reliability: Consistency across different testing occasions.
  • Split-Half Reliability: Consistency within a single test.
  • Inter-Rater Reliability: Consistency across different evaluators.

Validity

• Definition: The degree to which a test accurately measures the intended attribute.
• Reliability vs. Validity: Reliability is a prerequisite for validity, but reliability alone does not guarantee validity.

Types of Bias

1. Selection Bias (Sampling Bias)
   • Definition: Occurs when the sample is not representative of the population.
   • Examples:
     • Selecting subjects from a health club to predict heart disease rates.
     • Using only hospital records (Berkson’s bias).
     • Including people with certain characteristics and excluding others (nonrespondent bias).
   • Solution: Employ random, independent sampling and apply data weighting.
2. Measurement Bias
   • Definition: Information is gathered in a way that distorts findings.
   • Examples:
     • Asking leading questions, e.g., “You don’t like your doctor, do you?”
     • Subjects alter behavior because they know they are being studied (Hawthorne effect).
   • Solution: Use control groups to minimize bias.
3. Experimenter Expectancy (Pygmalion Effect)
   • Definition: Experimenter’s expectations influence the subject’s responses.
   • Solution: Use a double-blind design where neither the subject nor the researcher knows who receives the treatment.
4. Lead-Time Bias
   • Definition: Early diagnosis gives the illusion of increased survival.
   • Solution: Use life expectancy or “back-end” survival measures for accurate assessments.
5. Recall Bias
   • Definition: Participants fail to accurately remember past events.
   • Examples: Difficulty recalling specific behaviors from the past.
   • Solution: Confirm information using multiple sources.
6. Late-Look Bias
   • Definition: Individuals with severe disease are less likely to be part of a study due to early mortality.
   • Solution: Stratify participants by disease severity.
7. Confounding Bias
   • Definition: External factors obscure the relationship between variables.
   • Example: In a study on exercise and heart disease, age differences between groups may influence outcomes.
   • Solution: Conduct meta-analyses and control for confounders.
8. Design Bias
   • Definition: Elements of the study do not align to answer the research question effectively.
   • Example: Comparing effects of a treatment in non-comparable groups.
   • Solution: Random assignment of participants to control and treatment groups.

Table: Types of Bias, Associations, and Solutions

Points to Remember

• Reliability is necessary for validity but not sufficient by itself.
• Bias can significantly alter research outcomes and must be minimized for valid and reliable results.
• Using random sampling, control groups, double-blind designs, and multiple data sources are effective strategies for reducing bias.

Bibliography

• Last, J.M. (2001). A Dictionary of Epidemiology. Oxford University Press.
• Porta, M. (2014). A Dictionary of Epidemiology. Oxford University Press.
• Rothman, K.J., Greenland, S., Lash, T.L. (2008). Modern Epidemiology. Lippincott Williams & Wilkins.