Session 4 Flashcards
Confounder
A variable that contributes to confounding, that is a common cause of the exposure and the outcome under study.
Confounding
Non-exchangeability that arises from common causes of the exposure and the disease.
Confounding by Indication
Confounding that arises when the exposure is a treatment and the outcome, or the severity of the outcome is an indication for the treatment. For example, in a study of the effects of anti-depression medication on suicidal ideation it may be difficult to disentangle the effects of the medication on the outcome from the effects of the indication for the medication (depression) on the outcome.
Collider Bias
Bias that arises from conditioning on a common effect of the exposure and outcome (or variables associated with the exposure and outcome). This is the way that selection bias is depicted in a DAG and is often used as a synonym for selection bias.
Comparability
Often used as a synonym for exchangeability.
De-confounder
A term developed by Pearl to mean a variable whose control will mitigate confounding but which is not a common cause of the exposure and the outcome.
Direction of Confounding
Positive vs. Negative
We label a confounder “positive” if the uncontrolled association is more positive than the true causal effect.
We label a confounder “negative” if the uncontrolled association is more negative than the true causal effect.
Exchangeability
A situation where the probability of disease in the unexposed equals the counterfactual risk of disease in the exposed (the risk of disease in the exposed if they were unexposed).
When the proportion of doomed and protective response types are the same in the exposed and unexposed people
Internal Validity
The association found in a study sample equals the causal effect that the exposure actually had on the exposed in the source population. Source population is defined as the exposed and unexposed people in the study at T1 who gave rise to the cases.
Measurement Error
Non-exchangeability that arises from errors in how variables are measured. This type of non-exchangeability is between the exposed and unexposed as they are labeled, rather than in terms of their actual response type.
Monotonicity
In the context of this class, we often invoke a monotonicity assumption. This assumption states that if an exposure has a harmful effect on some individuals in the study, it does not have a protective effect on others. This assumptions is reasonable for some risk factors but not reasonable for others. If we cannot make this assumption, then our average causal effects can only tell us the difference between the proportion of causal and protective types. If we can make this assumption, then we can estimate the proportion of causal types.
Negative Relationship
When an increase in one variable predicts a decrease in the other. For example, when vitamin D supplementation decreases the probability of rickets. Risk ratios, rate ratios and odd ratios below 1 indicate a negative relationship; risk differences above 0 indicate a negative relationship. For dichotomous exposures, we say that the exposure prevents the outcomes (rather than causes it).
Positive Relationship
When an increase in one variable predicts an increase in the other. For example, when exposure to lead increases the probability of neurological difficulties. Risk ratios, rate ratios, and odds ratios above 1 indicate a positive relationship; risk differences above 0 indicate a positive relationship. For dichotomous we say that the exposure causes the outcome (rather than prevents it).
Random Sampling Error
Non-exchangeability that arises due to chance.
Response Type
An individual’s set of potential outcomes for a particular exposure and disease. For binary exposures and outcomes, each individual can be categorized into 1 of 4 possible response types:
Type 1, doomed (gets disease with or without the exposure)
Type 2, causal (gets the disease if exposed by not if unexposed)
Type 3, protective (does not get the disease if exposed, does get the disease if unexposed). Also referred to as preventative types.
Type 4, immune (does not get the disease if exposed and does not get the disease if unexposed)
