A medical clinic that surveys patients with a history of heavy drinking is likely to find that patients who have started to experience serious liver problems are more likely to cut back their drinking than others with a history of heavy drinking. An enthusiastic doctor uses data from this population and regresses current liver function, L, on current alcohol consumption, A. The regression specification is L = a + bA. The doctor is likely to get a positive estimate for b: Within the population of people with a history of heavy drinking, those who drink more currently tend to have healthier livers. What is wrong with this specification? (Hint: Think about factors that affect the demand for alcohol in this population.)