Bias, Confounding and Interaction Flashcards
Random Error
- “accidental” error that occurs because of variability in measurement
Systematic Error
- error in measurement that occurs in the development of an instrument
- no matter how many times you measure it will always be wrong
- or repeated mis-use of that instrument
Selection Bias
- happens when some feature of data collection makes the participation of some people in the study, less likely than others
Non-Response
- Subjects declining to participate, or don’t further agree with the study
Information Bias
- the data that we collect leads to research participants being misclassified
- also called misclassification bias
Types of Misclassification Bias (2)
- Differential Misclassification
2. Nondifferential Misclassification
Differential Misclassification
- happens when one study group or the other is misclassified MORE frequently than the other
Nondifferential Misclassification
- errors in both/any study group (ie both case + controls)
- mainly a problem to do with data collection
- less likely to find an association between groups
Reporting Bias
- participants are less likely to report an exposure that may-well be related to their disease, but which is socially undesirable
- “covering up” or “creating” a new one
Information Bias comes from (6)
- interviews
- surveillance
- recall bias
- reporting
- data taken from records
- surrogate bias (info from close family on behalf of loved one)
Confounding
- is not an error found in the study, but a phenomena that is identified
- an extraneous factor identified
How can you control for confounding? (4)
- Randomization
- Restriction
- Stratification
- Matching
Randomization
- to ensure equal distributions of characteristics between groups
Restriction
- remove potential influences of a confounder by eliminating that characteristic in our sample population
Stratification
- create study groups that are less vulnerable to confounding
Ex. if age is a confounder, use study groups with a smaller age range
Matching
- match subjects in groups according to the value of the known confounding variables to ensure equal distribution in each group.
Interaction
- is concerned with the effects of multiple exposures, such that the effect of any given exposure is amplified / mitigated by the presence of another factor
T/F - Exposure to a single causal factor, incidence rates will be higher than the background risk
True
T/F - If interaction is not happening, the incidence rates for those with both exposures will simply be the risk
True
T/F - If the incidence rate in those exposed to both factors is higher than the risks added together, we can conclude that interaction is occurring, such that exposure to both factors multiplies the risk
True
Synergism
- a situation in which the combined effect of two or more factors is greater than the sum of their solitary effects
Antagonism
- combined effect of two or more factors is SMALLER or WEAKER than that expected from the effect of one factor in the absence of the others
