Epidemiology: Errors, Bias, Confounding, and Effect Modification

What is random error?

Unpredictable variation in measurements caused by imperfections in measurement tools, procedures, or reporting.

How does random error differ from bias?

Random error involves unpredictable variation in measurements, whereas bias is a systematic error that consistently distorts results in a particular direction.

What is inherent variability?

Natural differences between individuals in a population. It is not measurement error and allows epidemiologists to investigate differences in disease risk.

Why does random error occur in epidemiologic studies?

No measurement system is perfect. Errors can arise from instruments, laboratory procedures, human measurements, and participant self-reports.

How can random error affect questionnaire data?

Participants may accidentally select incorrect answers or provide imprecise information, especially when asked to recall information they cannot remember accurately.

How can researchers minimize random error?

By using high-quality instruments, training staff to perform measurements consistently, and designing effective questionnaires.

Can random error be completely eliminated?

No. Researchers can minimize it, but some random error remains in measurements.

How do researchers quantify random error?

They use statistical methods, particularly p-values and confidence intervals.

What is a null hypothesis?

A hypothesis stating that there is no association or difference between the variables being investigated.

What is an alternative hypothesis?

A hypothesis proposing that an association or difference exists between the variables being investigated.

What is a p-value?

The probability of obtaining a result at least as extreme as the observed result, assuming the null hypothesis is true.

What does a p-value tell researchers?

It indicates how compatible the observed data are with the null hypothesis.

What does a p-value NOT tell researchers?

It does not indicate the probability that the null hypothesis is true or the probability that the research hypothesis is correct.

What is statistical significance?

A determination that study findings meet a chosen statistical criterion for rejecting the null hypothesis.

What does it mean to reject the null hypothesis?

The study provides sufficient statistical evidence against the null hypothesis according to the chosen significance criterion.

What does it mean to fail to reject the null hypothesis?

The study has not provided sufficient evidence against the null hypothesis. This does not prove that the null hypothesis is true.

Why should researchers avoid automatically accepting the null hypothesis?

A study may fail to detect a real association because of insufficient information, random variation, or limitations in its design.

What factors influence a p-value?

Sample size, the magnitude of the effect being investigated, and the consistency or variability of the data.

What is frequentist statistics?

An approach to statistics that uses methods such as null hypothesis significance testing to evaluate evidence from observed data.

What is Bayesian statistics?

An approach to statistics that uses probability to quantify uncertainty about hypotheses or events.

What is a confidence interval?

A range of values calculated from sample data to express uncertainty about an estimated population parameter.

What information does a confidence interval provide?

It communicates the uncertainty and precision of an estimate and indicates which population values are compatible with the data under the method's assumptions.

What does a narrow confidence interval indicate?

A relatively precise estimate with less statistical uncertainty.

What does a wide confidence interval indicate?

A less precise estimate with greater statistical uncertainty.

How does sample size generally affect confidence intervals?

Larger samples generally produce narrower confidence intervals, assuming other factors remain comparable.

How are confidence intervals related to statistical significance?

For the corresponding standard tests, a confidence interval that excludes the null value indicates statistical significance at the matching significance level.

What is a Type I error?

Rejecting a true null hypothesis. It is also called a false positive.

What is a Type II error?

Failing to reject a false null hypothesis, thereby missing a real association or difference. It is also called a false negative.

What is statistical power?

The ability of a study to detect a real association or difference when one exists.

What is the relationship between statistical power and Type II error?

Higher statistical power corresponds to a lower probability of making a Type II error.

What does it mean for a study to be underpowered?

The study has insufficient ability to detect an association or difference that genuinely exists.

Why should researchers consider statistical power when interpreting nonsignificant findings?

A nonsignificant result may reflect insufficient power rather than the genuine absence of an association.

What is bias in epidemiology?

Systematic error that distorts study findings, potentially causing an association to be overestimated or underestimated.

How does bias differ from random error?

Random error produces unpredictable measurement variation, whereas bias systematically distorts results.

Can statistical methods eliminate bias?

No. Bias must primarily be addressed through appropriate study design, measurement procedures, and careful interpretation.

Can bias be completely eliminated from epidemiologic studies?

No. Researchers should minimize potential bias and acknowledge any remaining limitations.

How can bias affect a measure of association?

It can produce an estimate that differs from the value that would have been obtained without the bias.

What is internal validity?

The extent to which a study's findings are trustworthy based on its design, measurements, and analyses.

What can threaten internal validity?

Methodological problems, inappropriate measurements, incorrect analyses, and biases that distort comparisons within the study.

What is external validity?

The extent to which study findings can be generalized to the larger population or other relevant populations.

How do internal and external validity differ?

Internal validity concerns whether the study's findings are trustworthy; external validity concerns whether those findings apply beyond the study population.

Can a study have good internal validity but poor external validity?

Yes. A study may be conducted correctly but use a sample that does not adequately represent the population to which researchers want to generalize.

What is selection bias?

Bias arising from how participants are selected, included, retained, or lost from a study.

How can selection bias affect external validity?

The sample may not adequately represent the target population, limiting the populations to which the findings can be generalized.

How can selection bias affect internal validity?

The groups being compared may come from different underlying populations, making their outcomes unfairly or incorrectly comparable.

What questions help researchers identify selection bias?

Who was included? Who was excluded or missed? Did the selection process differ between the groups being compared?

What is healthy worker bias?

A form of selection bias arising because employed people are generally healthier than the overall population, which also includes people too ill to work.

Why can healthy worker bias affect generalizability?

Findings from employed populations may not apply to the overall population, particularly people who cannot work because of illness.

How can loss to follow-up introduce selection bias?

Participants who leave a study may differ systematically from those who remain, distorting the groups being compared.

How can differences in participation introduce selection bias?

If participation differs between groups and participation is related to relevant characteristics, the groups may no longer represent comparable populations.

What is misclassification bias?

Systematic error in which participants are incorrectly classified according to their exposure or disease status.

How can misclassification occur?

Through inaccurate measurements, incorrect self-reports, faulty assessments, or errors made by interviewers or clinicians.

What is nondifferential misclassification?

Misclassification that occurs similarly across the groups being compared.

What is differential misclassification?

Misclassification that occurs differently across study groups, such as when one group reports information less accurately than another.

How do nondifferential and differential misclassification differ?

Nondifferential misclassification occurs similarly across groups, whereas differential misclassification varies according to group membership or disease or exposure status.

Does nondifferential misclassification always bias results toward the null?

No. Its direction of bias cannot always be predicted.

Why is differential misclassification particularly concerning?

It can distort comparisons between groups and seriously threaten a study's internal validity.

What is recall bias?

A form of information bias that occurs when participants remember or report past exposures or experiences differently.

What is interviewer bias?

Bias introduced when an interviewer's behavior, questioning, or interpretation influences the information collected.

What is clinician bias?

Bias arising when a clinician's expectations or knowledge influence the assessment or classification of participants.

What is social desirability bias?

A tendency for participants to report information in ways they believe are socially acceptable rather than accurately describing their behavior or experiences.

How should researchers evaluate self-reported information?

They should consider whether participants can reasonably know the answer and whether they are likely to report it accurately and honestly.

What is missing-data bias?

Bias that occurs when missing information is related to characteristics of participants or study variables, systematically distorting the analysis.

When are missing data less likely to introduce bias?

When the information is genuinely missing at random rather than following a pattern related to participant characteristics or study variables.

Why should researchers investigate patterns of missing data?

Missingness may be associated with the exposure, outcome, or other relevant characteristics, potentially distorting the results.

What is publication bias?

Bias arising when studies with more exciting, positive, or statistically significant findings are more likely to be published than studies with null or less striking findings.

How does publication bias affect the scientific literature?

The published evidence may give a misleading impression of an association because studies with less exciting findings are underrepresented.

What is a sensitivity analysis?

An additional analysis that examines how changing assumptions, methods, or analytical decisions affects study results.

Why do researchers conduct sensitivity analyses?

To evaluate how robust their findings are and investigate whether particular assumptions or potential biases substantially influence the results.

What should researchers do about residual bias?

They should acknowledge plausible remaining biases and discuss their potential effects when interpreting and reporting the study's findings.

What is confounding?

A distortion of the association between an exposure and an outcome caused by a third variable.

What is a confounder?

A variable other than the exposure or outcome that can distort the estimated association between them.

Why is confounding a problem in epidemiologic research?

It can produce a misleading estimate of association and lead researchers to draw inaccurate conclusions about the relationship being studied.

What are the criteria for a potential confounder?

It must be associated with the exposure, cause the outcome, and not lie on the causal pathway between the exposure and outcome.

Why must a potential confounder be associated with the exposure?

Its distribution must differ across exposure groups, creating the potential to distort the observed association.

Does a confounder have to cause the exposure?

No. It must be associated with the exposure, but that association does not necessarily have to be causal.

Why must a potential confounder cause the outcome?

It must have a causal relationship with the outcome to meet the chapter's criteria for a potential confounder.

Does an association between a variable and an outcome automatically make that variable a confounder?

No. The variable must also be associated with the exposure and must not lie on the causal pathway.

Why must a confounder not lie on the causal pathway?

A variable on the pathway is part of the process through which the exposure affects the outcome, rather than a confounder under the chapter's framework.

What is a mediator?

A variable that lies on the causal pathway between an exposure and an outcome and helps explain how the exposure affects the outcome.

How does a mediator differ from a confounder?

A mediator lies on the causal pathway, whereas a confounder does not and instead distorts the estimated exposure-outcome association.

What is a crude or unadjusted measure of association?

A measure of association calculated without accounting for potential confounding variables.

What is an adjusted measure of association?

A measure of association calculated after accounting for relevant confounding variables.

What is confounder control?

The process of reducing or removing the influence of confounding variables to obtain a more appropriate estimate of the exposure-outcome association.

What are the main study-design methods for controlling confounding?

Restriction, matching, and randomization.

What is restriction?

Limiting study eligibility to participants who share a particular characteristic of a potential confounder.

How does restriction control confounding?

It removes variation in the restricted characteristic, preventing that characteristic from being distributed differently across exposure groups.

What is a limitation of restriction?

It can limit recruitment and reduce the generalizability of findings to people outside the restricted group.

What is matching?

Selecting participants or controls so that they have comparable distributions of a potential confounding variable.

How does matching help control confounding?

It makes the distribution of the matched variable more comparable between the groups being studied.

What is randomization?

Assigning participants to study groups by chance rather than allowing assignment to depend on participant characteristics or investigator choice.

How does randomization help control confounding?

It helps distribute confounding variables comparably between study groups.

What are the main analytical methods for controlling confounding?

Stratification and regression.

What is stratification?

Dividing participants into subgroups based on a potential confounder and examining the exposure-outcome association separately within each subgroup.

Why do researchers perform stratified analyses?

To assess whether a variable is confounding the association between an exposure and an outcome.

What is regression in the context of confounder control?

A statistical modeling approach that estimates an association while accounting for other variables, including potential confounders.

How is regression related to stratification?

Both methods can account for other variables when estimating an association; regression provides a modeling-based approach to adjustment.

How can researchers identify confounding by comparing crude and stratum-specific associations?

Confounding is suggested when the stratum-specific measures are similar to each other but differ from the crude measure.

What does it suggest if the stratum-specific measures are similar to each other but differ from the crude measure?

The variable may be acting as a confounder, meaning the crude estimate is distorted.

What is residual confounding?

Confounding that remains after adjustment because the variable has not been controlled adequately.