Question

1.) Your uncle sends you a message over Twitter with a link to a medical study. The study compares older adults who report drinking red wine in moderation to older adults who reported abstaining from alcohol. Those who drink red wine in moderation are less likely to have hypertension. Your uncle concludes that he is better off drinking red wine for the health benefits.

(a) What is the causal relationship your uncle is trying to uncover and what is the counterfactual of interest?

(b) What are the types of omitted variables we might worry about in interpreting the study he cites? Recall the two criteria for an omitted variable to be problematic: that it is correlated with the “treatment” (red wine) and has an effect on the outcome (health).

(c) Suppose that wine drinkers are more likely to survive to old age than non-drinkers. Would this create a selection bias? How would this affect your interpretation of the evidence?

Answer 1

A) He is trying to draw the conclusion that wine drinking in moderation has health benefits such as lower hypertension. The counterfactual hypothesis of interest is health/hypertension in non-wine drinkers.

B) Variables such as age, gender, Body Mass Index(BMI), past history in family of hypertension; could have been used as well to explain this relationship better. It might be so that this is true only as one attains a particular age and not suitable for teenagers or a particular gender might be more or less prone to hypertension(genetically) which should be controlled for in the regression equation. Also, the present health of the individual(BMI) and any past history in family of hypertension decide whether one will suffer from hypertension or not and should also be controlled for. If the current health is good, there are lower chances of having it and same is the case if there is no or little family history regarding hypertension.

C) This may create a selection bias if we only choose a sub-sample of the population which is over say, 50 years. Now, studying only a sub-sample of the population may give bias results due to lower sample size. Also, if this proposition is true, then those who have survived old age will have higher number of people who have wine and therefore, our sample will contain higher number of people who have wine than those who don't creating a bias i.e. if inherently the sample has more wine drinkers then there will certainly be a sample selection bias.

The evidence is not entirely reliable since there are omitted variables and it suffers from possible sample selection bias problem.