(#2) college_gpa.dta contains data on first semester college GPA for a class of 4,137 college freshmen along
with information on student demographics, SAT scores and class rank in high school, and other potential
determinant of a student’s success in college.
a) Estimate the following equation and report the results. Remember to add the vee (robust) option
to Stata’s regress command.
colgpa = Bo + Pi hsize + Pa hspctile + Bs sat + Pafemale + Ps athlete +u
where colgpa is college grade point average; hsize is the size of the student’s high school
graduating class (in hundreds); hspctile is academic percentile in graduating class; sat is combined
SAT score; female is a binary (dummy or indicator) gender variable =1 if female and 0 if male; and
athlete is a dummy variable equal to one for student-athletes.
) What is the estimated GPA differential between athletes and non-athletes? Is it statistically
significant?
¢) Suppose you were to estimate this model with dummy variables male (=1 if male and 0 if female)
and nonathlete (=1 if not an athlete and 0 if an athlete). Without estimating the model in Stata, can
you determine what estimates you would get for fo, f3,84and fs? Check your answers using
Stata.
d) Return to the model in part (a), but now drop sat from the model and re-estimate the equation.
Now, what is the estimated effect of being an athlete? Discuss why the estimate is different than
that obtained in part (a).
e) Now modify your model from part (a) to allow the effect of being an athlete to differ by gender.
(You will need to create a new variable.) Does the effect of being an athlete differ significantly by
gender? If so, how?
f) Next, modify your model in part (a) so that you can test whether the effect of sat on colgpa differs
y gender. (You will need to create a new variable.) Does the effect of sat differ significantly by
gender? If so, how?
g) Finally, modify your model in part (a) to include the interaction of sat with Aspctile. (You will need
to create a new variable.) Is the interaction term statistically significant? Write two interpretations
of the results (refer to the signs of the main effects and interaction term; you do not need to
interpret the magnitudes).
(#3) ratings.dta contains historical ratings data on 88 made-for-TV movies that aired in 1992 on one of three
networks: ABN, CBC, BBS. (This was a long, long, time ago, before cable, Netflix, Amazon, and DVRs...)
a) Construct dummy variables for two of the three networks (ABN, BBS and CBC) using the categorical
string variable network. First, tabulate the network variables to see what the string values are.
. tab network /* so we can see what the string values are *
To construct a dummy variable for each value, use the generate command as follows:
. gen x = expression
This will create a dummy variable that is equal to one if the expression is true and 0 if it is false.
For example:
. gen abc = network=="ABN" /* enclose the string value in quotes *
creates a variable that is equal to one for observations where the variable network takes the value “ABN”
(note that string values must be enclosed in quotes) and zero otherwise. Note also the use of the logical
operator == (double equals sign).
) Regress ratings on two of the three dummies, using CBC as the omitted category. Why do we only
include two of the three dummies in the regression?
¢) Use the regression coefficients to calculate the average ratings for each network. Which network has the
highest average ratings? Which has the lowest average ratings?
d) For which network pairs is there a statistically significant difference in average ratings? [Hint: Perform
additional hypothesis test(s) as needed following the regress command. Use a 5% significance level.]