Epidemiology Midterm #2

What is random error?

Random variation or inaccuracies in measurements that occur by chance.

What is the difference between random error and bias?

Random error occurs by chance, while bias is systematic error that consistently distorts results.

What is inherent variability?

Natural differences between individuals in a population, such as differences in height, heart rate, or GPA. It is not itself random error.

Why is inherent variability important in epidemiology?

It allows researchers to identify which individuals or groups may have a higher risk of disease.

What can cause random error in measurements?

Imperfect measuring instruments, measurement techniques, human error, and inaccurate or imprecise self-reported information.

Can random error be completely eliminated?

No. It can be minimized through better instruments, staff training, and improved questionnaires, but it cannot be eliminated entirely.

How can researchers reduce random error?

Use high-quality measuring instruments, train staff, and design clear, effective questionnaires.

How does the quality of a measuring instrument affect random error?

Instruments that measure more consistently and precisely generally produce less random error.

How can questionnaires introduce random error?

Participants may misunderstand questions, provide imprecise estimates, or accidentally select the wrong response.

What question can help researchers assess random error in self-reported information?

Can participants realistically remember or accurately report the information being requested?

What is a null hypothesis?

A hypothesis stating that there is no difference or association being tested.

Why do researchers use a null hypothesis?

It provides a starting point for statistical testing to assess whether the observed results are compatible with no difference or association.

What is a p-value?

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

What does a small p-value indicate?

The observed results would be relatively unusual if the null hypothesis were true.

Does a p-value tell you the probability that the null hypothesis is true?

No. It describes the probability of the observed results under the assumption that the null hypothesis is true.

What does a p-value of 0.04 mean under the textbook's interpretation?

Assuming the null hypothesis is true, there is a 4% probability of obtaining results at least as extreme as those observed.

What is statistical significance?

A determination that a result meets the chosen statistical threshold for rejecting the null hypothesis.

What significance threshold is commonly used in public health and clinical research in the textbook?

A p-value of 0.05 or lower.

What does it mean to reject the null hypothesis?

The results provide sufficient statistical evidence against the null hypothesis according to the chosen significance threshold.

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

The study has not provided sufficient statistical evidence against the null hypothesis.

Why shouldn't researchers say they have accepted the null hypothesis simply because a result is not statistically significant?

A nonsignificant result does not prove that no difference or association exists. The study may lack sufficient sample size or precision.

Is the commonly used 0.05 significance threshold an absolute scientific boundary?

No. It is an arbitrary conventional threshold, and results just above and below it may be very similar.

What three factors influence a p-value, according to the textbook?

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

How can sample size affect statistical significance?

Larger samples can make it easier to detect a difference because they generally reduce uncertainty due to sampling variation.

Why can a very small effect still be statistically significant?

A sufficiently large sample may provide enough statistical evidence to detect a small effect.

What is a Type I error?

Incorrectly rejecting a true null hypothesis; commonly described as a false-positive result.

What is a Type II error?

Failing to reject a false null hypothesis; commonly described as a false-negative result.

What is statistical power?

The probability that a study will detect an effect when the specified effect truly exists.

How is statistical power related to Type II error?

Higher statistical power corresponds to a lower probability of making a Type II error, under the same study assumptions.

Why is a confidence interval useful?

It communicates uncertainty around an estimate and helps researchers assess its precision.

What does a narrow confidence interval generally indicate?

Greater precision and less uncertainty about the estimated value.

What does a wide confidence interval generally indicate?

Less precision and greater uncertainty about the estimated value.

What does a confidence interval that includes the null value generally suggest?

The result is not statistically significant at the corresponding level for the relevant statistical test.

What is the difference between statistical significance and practical significance?

Statistical significance concerns evidence against a statistical null hypothesis, while practical significance concerns how meaningful or important the result is in real-world settings.

What is internal validity?

The extent to which a study's findings are valid for the participants being studied.

What is external validity?

The extent to which study findings can be generalized to other populations or settings.

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

Yes. Its findings may be valid for the study participants but not generalizable to other populations.

What is bias in epidemiology?

Systematic error in the design, conduct, measurement, or analysis of a study that distorts its findings.

What are the two major categories of bias emphasized in the textbook's discussion of study bias?

Selection bias and misclassification bias. The chapter also discusses publication bias.

What is selection bias?

Bias that arises when the selection of participants or the availability of their data systematically distorts the study's results.

How can the selection of study participants introduce bias?

The people who enter or remain in a study may differ systematically from those who do not, potentially distorting the exposure-outcome association.

What is nonresponse bias?

Bias that may occur when people who do not participate differ in relevant ways from those who do participate.

How can loss to follow-up introduce selection bias?

If participants who leave a study differ systematically from those who remain, the final results may be distorted.

What is healthy worker bias?

A form of selection bias in which employed people tend to be healthier than the general population, making comparisons with the general population potentially misleading.

Why can comparing workers with the general population underestimate occupational health risks?

The general population includes people who may be too ill to work, making workers appear healthier in comparison.

What is misclassification bias?

Systematic error caused by incorrectly classifying participants according to their exposure or outcome status.

What is exposure misclassification?

Incorrectly classifying whether a participant was exposed to a factor.

What is outcome misclassification?

Incorrectly classifying whether a participant experienced the outcome or disease being studied.

What is nondifferential misclassification?

Misclassification that occurs similarly across the groups being compared with respect to the other variable under consideration.

What is differential misclassification?

Misclassification that differs between groups defined by another study variable.

Does nondifferential misclassification always bias results toward the null?

No. Although it often does under certain common conditions, the direction of bias is not guaranteed in every situation.

What is recall bias?

A type of information bias in which participants remember or report previous exposures differently depending on their outcome or disease status.

Why is recall bias a concern in case-control studies?

People with a disease may remember or report past exposures differently from people without the disease.

What is interviewer bias?

Bias that occurs when an interviewer's behavior or questioning influences the information collected.

What is observer bias?

Bias that occurs when an observer's expectations or knowledge influence how observations or measurements are made or interpreted.

What is social desirability bias?

The tendency for participants to provide answers they consider socially acceptable rather than answers that accurately reflect their behavior.

How can blinding help reduce bias?

It can prevent researchers or assessors from knowing information that might influence how they measure or interpret study variables.

What is publication bias?

Bias that occurs when studies with certain results, especially statistically significant or positive findings, are more likely to be published or made available.

Why is publication bias a problem?

Published studies may not represent all research conducted, potentially distorting conclusions drawn from the available evidence.

What is sensitivity analysis?

An approach that examines how study findings change when assumptions or analytical choices are varied.

Why is increasing sample size insufficient to eliminate bias?

A larger sample may reduce random error, but it does not automatically correct systematic problems in participant selection or measurement.

What is confounding?

Distortion of the estimated association between an exposure and an outcome by another variable.

What are the three traditional criteria for identifying a confounder?

It is associated with the exposure, is a risk factor or predictor of the outcome, and is not an intermediate step in the causal pathway between exposure and outcome.

Why must a potential confounder be associated with the exposure?

Its relationship with the exposure is necessary for it to distort the exposure-outcome association through the traditional confounding mechanism.

Why must a potential confounder be related to the outcome?

It must be capable of contributing to differences in the outcome.

What is a mediator?

A variable that lies on the causal pathway through which an exposure affects an outcome.

What is the difference between a confounder and a mediator?

A confounder can distort the exposure-outcome association, while a mediator is part of the pathway through which the exposure may affect the outcome.

Why is a mediator not considered a traditional confounder?

Because it lies on the causal pathway between the exposure and the outcome.

What is a crude or unadjusted estimate?

An estimate of the exposure-outcome association that has not been adjusted for other variables.

What is an adjusted estimate?

An estimate that accounts for one or more additional variables in the analysis.

Why compare crude and adjusted estimates?

To evaluate whether accounting for other variables changes the estimated association.

If the crude and adjusted estimates differ meaningfully, what might this suggest?

Confounding may be present, although differences should be interpreted in light of the measure and analytical method used.

What is confounding control?

The use of study design or analytical methods to reduce or account for the influence of confounding variables.

What are three common methods for controlling confounding during study design?

Randomization, restriction, and matching.

How can randomization help control confounding?

Random assignment tends to distribute known and unknown confounding factors across study groups.

Does randomization guarantee perfectly balanced groups?

No. Chance imbalances can still occur, especially in small studies.

What is restriction as a method of controlling confounding?

Limiting eligibility to a specific category of a potential confounder, such as enrolling only nonsmokers.

What is a limitation of restriction?

It can limit generalizability and prevent researchers from examining associations across excluded categories.

What is matching?

Selecting participants so that comparison groups have similar distributions of specified characteristics.

What is a limitation of matching?

It can complicate recruitment and analysis, and matched variables must be handled appropriately during analysis.

What are two common analytical methods for controlling confounding?

Stratification and multivariable statistical adjustment.

What is stratification?

Dividing participants into groups according to a variable and examining the exposure-outcome association separately within each group.

How can stratification help researchers assess confounding?

It allows them to compare stratum-specific estimates with the crude estimate to see whether the overall association is distorted by the variable.

What is multivariable adjustment?

A statistical approach that accounts for multiple variables simultaneously when estimating an association.

What is residual confounding?

Confounding that remains after adjustment because a variable was not measured or controlled adequately.

How can inaccurate measurement of a confounder lead to residual confounding?

Inaccurate measurement can prevent the analysis from fully accounting for the confounder's influence.

Can researchers always control every possible confounder?

No. Some confounders may be unknown, unmeasured, or inadequately measured.

Why should researchers not automatically adjust for every available variable?

Some variables may be mediators or otherwise inappropriate adjustment variables, and controlling for them can distort the effect being estimated.

What is overadjustment?

Adjusting for a variable that should not be controlled for in the analysis, such as a mediator when the goal is to estimate the total causal effect.

What is effect modification?

A situation in which the exposure-outcome association differs across levels of a third variable.

What is another term used for effect modification in the textbook?

Interaction.

How does effect modification differ from confounding?

Confounding distorts an association and is something researchers generally try to control. Effect modification means the association differs across groups and is a finding researchers generally want to describe.

Should researchers eliminate effect modification in the same way they try to control confounding?

No. Effect modification is itself an important finding and should be reported.

How can researchers check for effect modification?

Conduct a stratified analysis and compare the exposure-outcome associations across the strata.

What are stratum-specific estimates?

Estimates of an exposure-outcome association calculated separately within each category of a third variable.

What pattern in stratum-specific estimates suggests effect modification in the textbook's approach?

The estimates differ from each other, with the crude estimate lying between them.

What should researchers do when effect modification is present?

Report the results separately for the relevant strata rather than relying only on one overall estimate.

Why can an overall crude estimate be misleading when effect modification is present?

It can conceal important differences in the association between subgroups.

If an association is statistically significant in one group but not another, does that alone prove effect modification?

No. Researchers need to compare the estimates themselves to determine whether the associations differ.

Can the same variable theoretically act as both a confounder and an effect modifier?

Yes, although the textbook notes that this is rarely seen in practice.