Exercise 7 (10 Credit Points) H. Allcott and M. Gentzkow analyzed the role of social media and fake news in the US presidential election of 2016. They wanted to find out who believed in fake news by constructing a variable Cia, which they used as dependent variable. Cia takes the value 1 if the respondent i correctly identifies whether article a is true or false Cia takes the value 0.5 if the respondenti is "not sure whether article a is true or false Cia take the value 0 if otherwise. Table 1 What Predicts Correct Beliefs about News Headlines? (2) (3) (4) (5) Democrat 0.029 (0.020) -0.024 (0.024) -0.004 (0.023) 0.040 (0.027) Republican In (Daily media time) Social media most important Use social media 0.028 (0.019) -0.037* (0.020) -0.002 (0.007) -0.066*** (0.025) 0.014 (0.030) -0.027 (0.036) 0.014*** (0.004) -0.011 -0.010 (0.021) 0.021 (0.023) 0.042*** (0.008) 0.065*** (0.024) -0.023 (0.038) 0.028 (0.046) 0.004 (0.004) 0.006 (0.022) 0.000 (0.001) 6,040 0.207 0.015 (0.013) -0.018 (0.014) 0.013*** (0.004) -0.023 (0.016) 0.002 (0.019) -0.008 (0.024) 0.011*** (0.003) -0.005 (0.013) 0.002*** (0.000) 18,120 0.035 Social media ideological segregation Education Undecided (0.017) Age 0.002*** (0.000) 12,080 0.004 12,080 0.029 6,040 0.124 N p-value (Democrat = Republican) Articles in sample False True False True All Note: This table presents estimates of a regression of a dependent variable measuring correct beliefs about headlines on individual characteristics. Columns 1 and 3 include only false headlines, columns 2 and 4 contain only true headlines, and column 5 contains all headlines. All columns include additional demographic controls: income, race, and gender. Social media most important" means social media were the respondent's most important sources of election news. "Social media ideological segregation" is the self-reported share (from 0 to 1) of social media friends that preferred the same presidential candidate. "Undecided" is an indicator variable for whether the respondent decided which candidate to vote for less than three months before the election. Observations are weighted for national representativeness. Standard errors are robust and clustered by survey respondent. * **, *** indicate statistically significantly different from zero with 90, 95, and 99 percent confidence, respectively. How do you interpret the above results presented in the regression model? What about their quality? Exercise 4 (13 Credit Points) In a recent article in the Journal of Economic Perspecitves (Winter 2009) B. Edelman analyzed the question "Who buys Online Adult Entertainment?" in the US. The dependent variable is "the share of subscribers per 1,000 households with broadband access". The specification of one of his models is as follows: Coefficient Standard Errors 0.0057 -1.79x105 0.001 5.58x106 0.259 -0.095 0.104 0.068 Household Income In thousands In thousands squared Proportion aged 15-25 65+ Education of College or more Graduate degree Population density Urban (1,000-5,000 / squre mile) Married Regularly attending religious services 0.195 -0.524 0.594 0.138 0.361 -0.606 -0.002 0.004 0.072 0.008 a. Please "translate" the findings based on this data analysis to one of your friends, who has no idea about econometrics. (4 CP)
