Description

Long answer:

5) Read the 2014 Gentile et al. article “Mediators and Moderators of Long-term Effects of Violent Video Games on Aggressive Behavior” (available alongside this exam and also here: http://archpedi.jamanetwork.com/article.aspx?articleid=1850198).

a) The authors posit that violent video games increased aggressive behavior. Write down a model specification that corresponds to the basic OLS regression model testing whether exposure to video games increases violent behavior as shown in the paper. Explain how you would interpret the coefficient on video games.

b) Explain how the conditional mean zero assumption may be violated by the researchers research design. What might you expect this would do to the expected value of the coefficient? Why does this matter?

c) Write down an alternate model specification that makes reference to and quantifies an additional social factor that might be correlated with aggressive behavior, and explain how including that variable in the model might potentially change the researcher’s results.

d) Propose a research design - experimental, natural, or otherwise - that would better allow you to determine whether video games lead to aggressive behavior. Make sure to pay attention to needs for properly identifying your parameters (eg through randomization), addressing inference concerns, and ensuring a representative sample.

Empirical work:

 6) Note: YOU MUST DO ALL YOUR WORK IN A DO FILE including loading the data sets from a folder on your computer and any data cleaning / preparation. You should make use of globals to refer to file locations and have a well-formatted do file following the lecture 7 sample do file on Canvas under Files>Assignments.

a) First, go to the United Nations’ National Accounts Main Aggregates Database website (https://unstats.un.org/unsd/snaama/basic). Download two data series, ‘GDP at constant 2010 prices in US Dollars’, and ‘Population’ as Excel or csv files for all (250) countries and all (48) years, and save both in an easily accessible file location referred to by a global. In your do file, import both excel files to Stata and save them as Stata format files with appropriately named variables, using the import or insheet and save commands and reference to globals associated with file directories. Merge the two datasets by country and year, report how many country-year observations were missing from both data sets. For each country-year pair, create a new GDP per capita (GDPpc) variable by dividing GDP by population. Then, calculate the average value of GDPpc for each country using by and egen and create three indicator variables indicating whether a country’s average annual GDPpc was below $3000 2010USD, above $3000 but below $10,000, or above $10,000 (i.e., lower, middle, and upper income). Be sure that each country is consistently assigned to only one group across observations. In your write up, create a summary statistics table showing the mean, standard deviation, min and max values of GDP, population, and GDP per capita for each of the three groups, and include it in your answers.

b) Choose 2 countries, one each from the low, and high income groups. Using the twoway and graph export commands, plot a separate line chart for each country showing the value of GDP per capita on the vertical axis and time (the years 1970–Present) on the horizontal. Label both axes and put a title or legend so the graph is a standalone figure. Submit a copy with your write up, and interpret the two graphs. Which country grew fastest over all? Are there any major events (e.g., wars, periods of rapid growth) that you can find evidence of in either country’s time series?

c) Generate two new variables for the whole sample: ln(GDP) and ln(population), i.e., the natural log of GDP, and the natural log of population. Run the cross sectional regression of lnGDP on lnPop for all countries in the year 2000 (i.e., one observation per country where year == 2000). What does this unconditional relationship suggest? Either copy and paste your regression output (making sure it is legible) or output it using outreg2 or estout, and then interpret your model, making reference to the coefficient on your independent variable, the constant, the standard errors and test statistic / confidence intervals, and the R2.

d) Now run the regression with all observations (i.e., not just year 2000) as in part (c), but including year as an additional x variable; i.e., regress lnGDP on lnPopulation and year. Insert your output into your write up as a table again and interpret your results. What happened to the relationship between lnGDP and lnPopulation compared to part (d)?

e) Now replicate the same regression (lnGDP on lnPop and year) for the low and high income subsamples separately, including the table in your results. How does the relationship in each case differ compared to the overall relationship?

f) "Partial out" lnGDP and lnPopulation for the whole sample by first regressing each separately on year, predicting residuals from those regressions (using the predict varname, resid command), and then running the regression of residual Ys on residual Xs. How does this regression relate to the regression in part e?

g) Use the xtset command to define your data as a panel of countries with yearly observations, and then use the l1. time series operator to calculate per capita growth, or the % change in GDP per capita for each country in a given year (i.e., the difference between the current and previous year divided by the previous year). Run the regression of GDP growth against both population and time for the whole sample, and for the two low and high income subsamples. Interpret your results to explain how the relationship between population, economic growth, and time varies within the sample.

h) Run two separate regressions of lnGDP per capita against lnPopulation, one for the year 1980 sample, and one for the year 2010. Interpret your output for both, and then plot a graph for each subsample showing both the scatter plot of lnGDPpc against lnPopulation with a linear fit curve, to be submitted with your write up. You will need to use the stata graph commands scatter, lfit, and twoway (to combine the two graphs into one). What can you conclude about the relationship between GDP per capita and population in 1980 vs in 2010?