1. Causal Inference and Bias 1a) Causal effect of education on earnings Suppose we took survey data that included people's earnings and the number of years of education they have, and we estimated this model: Earnings; = Bo + B. Education; + U And our estimate for B. is very high. Think of one unobservable variable that is absorbed in U; that correlates with both Earnings and Education. By omitting that variable, do we expect B will be biased? What would an ideal experiment look like to find the causal effect of education on earnings? 1b) Causal effect of health insurance on health Suppose we took survey data that included measures of people's health and whether they have health insurance or not, and we estimated this model: Health = Be + B, Health Insurance; +U And our estimate for B, is large and negative. Think of one unobservable variable that is absorbed in U; that correlates with both Health and Health insurance. By omitting that variable, do we expect , will be biased? What would an ideal experiment look like to find the causal effect of health insurance on health? 1c) Causal effect of Migration on Earnings Suppose we took survey data that included people's earnings and whether they have recently moved to a new city or not, and we estimated this model: Earnings; = Bo + B. Migration, + Ui And our estimate for B. is large and positive. Think of one unobservable variable that is absorbed in Wy that correlates with both Earnings and Migration. By omitting that variable, do we expect Bi will be biased? What would an ideal experiment look like to find the causal effect of migration on earnings? 1d) Causal effect of friends of the opposite sex on a high school student's GPA Suppose we took survey data that included high school students' GPAs and the number of friends of the opposite sex they have, and we estimated this model: GPA; = Bo + B, Opposite Sex Friends, + U And our estimate for B. is large and negative. Think of one unobservable variable that is absorbed in w; that correlates with both GPA and Opposite Sex Friends. By omitting that variable, do we expect Bi will be biased? What would an ideal experiment look like to find the causal effect of Opposite Sex Friends on GPA? 1. Causal Inference and Bias 1a) Causal effect of education on earnings Suppose we took survey data that included people's earnings and the number of years of education they have, and we estimated this model: Earnings; = Bo + B. Education; + U And our estimate for B. is very high. Think of one unobservable variable that is absorbed in U; that correlates with both Earnings and Education. By omitting that variable, do we expect B will be biased? What would an ideal experiment look like to find the causal effect of education on earnings? 1b) Causal effect of health insurance on health Suppose we took survey data that included measures of people's health and whether they have health insurance or not, and we estimated this model: Health = Be + B, Health Insurance; +U And our estimate for B, is large and negative. Think of one unobservable variable that is absorbed in U; that correlates with both Health and Health insurance. By omitting that variable, do we expect , will be biased? What would an ideal experiment look like to find the causal effect of health insurance on health? 1c) Causal effect of Migration on Earnings Suppose we took survey data that included people's earnings and whether they have recently moved to a new city or not, and we estimated this model: Earnings; = Bo + B. Migration, + Ui And our estimate for B. is large and positive. Think of one unobservable variable that is absorbed in Wy that correlates with both Earnings and Migration. By omitting that variable, do we expect Bi will be biased? What would an ideal experiment look like to find the causal effect of migration on earnings? 1d) Causal effect of friends of the opposite sex on a high school student's GPA Suppose we took survey data that included high school students' GPAs and the number of friends of the opposite sex they have, and we estimated this model: GPA; = Bo + B, Opposite Sex Friends, + U And our estimate for B. is large and negative. Think of one unobservable variable that is absorbed in w; that correlates with both GPA and Opposite Sex Friends. By omitting that variable, do we expect Bi will be biased? What would an ideal experiment look like to find the causal effect of Opposite Sex Friends on GPA
