PS6

2 Part II: Schooling and interpersonal trust We are interested in understanding whether schooling affects how much people trust one another. We will use data from the World Values Survey, a nationally representative survey conducted in a large number of countries. It captures a wide array of attitudes, beliefs, individual characteristics, etc. The dataset includes the following variables: country - Country of residence of the respondent urban - Indicator for urban area (vs. rural area) trust - Indicator for whether the respondent thinks most people can be trusted (1=yes) male - Indicator for the respondent ’ s sex being male age - Respondent ’ s age educ - Respondent ’ s highest educational level achieved GDPpercap - GDP per capita in respondent ’ s country of residence 1. Create an indicator variable called college for whether the respondent has a college degree or not. 2. Regress the indicator for trust on the indicator for college and interpret your result. 3. Do you think the above regression gives you the causal relationship between graduating college and trust? What are the factors that you think might cause problem in this case? Why? 4. Rerun your regression adding some controls: male , age , urban , and GDPpercap . How does the value and significance of the coefficient on college change? Why did it change? 5. Do you think it would be important to control for the respondent ’ s country of residence? Why, and why isn ’ t it enough to control for GDPpercap ? 6. Rerun your regression adding country fixed effects. (You should use the <areg> command with the absorb( … ) option. Type help areg for details.) How does the value and significance of the coefficient on college change? Interpret the coefficient given the new set of controls. Write down the equation that you are estimating, and label all the variables. 7. Which variable can be omitted from the regression above without changing the result? Why? 8. Do you think we can claim this is an unbiased estimate of the effect of graduating college on trust? Can you think of an omitted variable that could be introducing bias, even after controlling for country fixed effects? Many correct possible answers here – we will evaluate based on the explanation. o.io sa this result is statistically significant at p value it means being college educated in associate with a 102 increase in trust controlling for an e is I don't think it does Somefactors that may cause problems in this case could potentially be ours an example of this isfamily background personality traitsor cultural differences that could affect tru and college Oops It decreases slightly bit remainssignificant It changes because mom omittedvariables were added making result more adjusted yes Ithink it i r important because there are many factors that ooo s It remains significant but decreases again it means Being conge educated is assoc w a t.gg increase in trust controlling forall other factors trust 0.19130.584 college to 0133 male to ooo age 0.0028 urba GDP per capitacanbe omitted from the regression because it is coninear we already account forit with the other variables Yes wecanclaim it's a nonbiased estimate an omitted variable that could introduce bias are specific personality traits Fixed effect only help addressunobserved country level factors whereas individual level factors such asthese differing personality traits could leave there tobe a biasedestimate of trust