Name___Clement Runtung________________ ECO 140

GSI____Caroline Le Pennec _______________ Fall

2017

Homework 7

1. This problem set examines the viewing of violence in movies as a determinant of crime. Some
argue there is a connection between viewing violent movies and the incidence of violent crimes.
On the other hand, movies get people off the streets and into theaters at times when most crimes
occur, the nighttime. And they substitute for other activities that lead to a violent behavior (drugs
and alcohol). To analyze this type of behavior you will work with a data set Movies.###. A
description of the variables is given in a file called movies_description.pdf. The data set
includes weekend U.S. attendance for strongly violent movies (attend_v) (such as Hannibal),
mildly violent movies (attend_m) (such as Spider Man) and non-violent (attend_n) (such as
Moana).

(a) Compute the logarithm of assaults (ln_assaults) and regress it on month and year dummy
indicators using January as a base. Is there evidence of seasonality in this type of violent
crime, i.e., do there tend to be more assaults in some months than others? Is there a trend in
assaults over the sample years?

Yes, there is evidence of seasonality. We can test this formally by using F-test that states
that all months coefficient = 0; the alternative being at least one of them is not equal to 0.
From this, we found that our F-stat is 76.26 with p-value of 0. Thus, we can reject the null
that there is no seasonality and say that there is evidence of seasonality from the data.

We can do similar thing to find a trend in assaults over the sample years by using F-test that
states that all years coefficient = 0; the alternative being at least one of them is not equal to
0. From this, we found that our F-stat is 6440.5. Thus, we can reject the null that there is no
trend over the sample years and say that there is a trend over the years regarding the assault.
(b) Generate the total attendance of all three categories of movies (attend = attend_v + attend_m
+ attend_n) and regress it on year and month indicators. Do you see seasonality in movie
attendance? Is there a trend in movie going over time?

Yes, there seems to be seasonality in movie attendance. We can test this by conducting
another F-test that have a null that says that there is no seasonality (month coefficient = 0).
Conducting an F-test on STATA gives us a value of 33.72 and p-value of 0. This means that
we can reject the null that there is no seasonality and say that there is evidence of seasonality
in the data.

Yes, there is also a trend in movie going over time. We can test this by conducting another F-
test in which the null states that there is no trend over time (year coefficient = 0). Conducting
an F-test on STATA gives us a value of 9.24 and p-value of 0. This means that we can reject
the null that there is no trend over time and say that there is evidence of a trend over time in
the data.

(c) Now regress ln_assaults on the attendance in each of the three types of movies, along with
month and year dummy indicators, and the weather and holiday control variables available in
the data set. Does viewing a strongly violent movie increase or decrease assaults, by how
much? Is it statistically significant?

For every 1 million people viewing a strongly violent movie is predicted to decrease assaults
by -0.31689%, ceteris paribus. This number is significant as the p-value is close to 0. Thus,
this number is statistically significant even at a 1% significance level.
(d) Test whether the effects of attendance at strongly violent and mildly violent movies are the
same. Do the same for the effects of attendance at strongly violent and non-violent movies.
State the Null hypothesis and specify your restricted regression.

1. Strongly Violent Movie attendance and Mildly Violent movies attendance

Change the normal form into:
Unrestricted Regression:
ln() = + 1 + 2 ( + ) + 3
+ , , +
where 1 = 1 2
Assume Null is true:
Ho: 1 = 0
H1: 1 0

Since we only have 1 restriction, we can look at the t-stat or the p-value of the attend_v
coefficient.
After regressing the Unrestricted regression above, it was found that the p-value of the
coeffieicnt attend_v is 0.973 which is not statistically significant at 5% level.

Conclusion: As the p-value is > 0.05, we conclude that we fail to reject the null hypothesis
that there is no effect of attendance at strongly violent movies and mildly violent movies and
say that there may be an effect between attending strongly violent movies and mildly violent
movies.

2. Strongly violent movie attendance and non-violent movies.

Unrestricted Regression:
ln() = + 1 + 2 ( + ) + 3
+ , , +

Assume Null is true:

Ho: 1 = 0
H1: 1 0

Similarly, since we only have 1 restriction, we can simply look at the t-stat or the p-vale
of the attend_v coefficient.

After regression the unrestricted regression above, it was found that the p-value of the
coefficient attend_v is 0.270 which is not statistically significant at 5% level.

Conclusion: As the p-value is >0.05, we conclude that we fail to reject the null hypothesis
that there is no effect of attendance at strongly violent movies and non-violent movies and
say that there may be an effect between attending strongly violent movies and non-violent
movies.
2. Movie attendance and the incidence of violent crimes can be determined by certain common
factors. Weather is a good example: when bad weather causes people to stay at home, they do not
go to theaters and they are also not the target of assault. Instrumental variable regression might
correct for the endogeneity of the attendance variables. The dataset contains three potential
instruments which provide estimates of the predicted attendance of strongly violent, mildly violent
and non-violent movies, pr_attend_v, pr_attend_m and pr_attend_n, respectively. These
predictions are based on knowledge of typical patterns for attendance of movies during their
theatrical distribution.

(a) Estimate the first stage of the 2SLS procedure and test whether the three variables are
strong instruments.

The first stage is to regress the attend_v, attend_m, and attend_n on the instrumental variables
and all the exogenous variables.

Start by regressing attend_v on all the instrumental variables and all the exogenous variables
(month, year, holiday, and weather). After regression, we found that the F-statistics is 75.25 and
p-value of 0. Thus, we conclude that the variable is statistically significant.

Second, we regress attend_m on on all the instrumental variables and all the exogenous variables
(month, year, holiday, and weather). After regression, we found that the F-statistics is 199.87 and
p-value of 0. Thus, we conclude that the variable is statistically significant.

Third, we regress attend_n on all the instrumental variables and all the exogenous variables
(month, year, holiday, and weather). After regression, we found that the F-statistics is 123.4 and
p-value of 0. Thus, we conclude that the variable is statistically significant

Thus, we conclude that the instrumental variable passes the relevance test and that all
instrumental variables is a good instrumental variable.
(b) Run two stage least squares regression using these three variables as instruments for the three
endogenous regressors. Evaluate the coefficient on attendance of violent movies and test
whether it is different from the coefficient on mildly violent movies, and then again for
nonviolent movies.

Coefficient: -.0038738, which means that for every 1 million attendance of strongly violent
movies, the predicted number of assault decreases by 0.387%, ceteris paribus.

To find whether the coefficient on attendance of violent movies and mildly violent movies,
we use identical technique from question 1(d) and by regressing it using ivreg in STATA, it
was found that the p-value is 0.983; meaning that we fail to reject the null that the effect of
attending violent movies is not different from attending mildly violent movies and say that
there may be an effect between attending violent movies or mildly violent movies.

Similarly, to find whether the coefficient on attendance of violent movies and non-violent
movies, we use similar technique from question 1(d) and by regressing it using ivreg in
STATA, it was found that the p-value is 0.228; meaning that we fail to reject the null that the
effect of attending violent movies is not different from attending non-violent movies and say
that there may be an effect between attending violent movies or non-violent movies.

(c) Comment on R2 from regression in part (b)

In this case, R2 is now unreliable as variability in the dependent variable (ln_assault) is due to
change in both the independent variables and also the error term.

(d) A different collection of instruments included in the dataset are the attendance of each of the
three categories of movies in the previous and subsequent weeks, e.g, attend_v_b and
attend_v_f are attendance at violent movies the week before and the week following,
respectively. Perform 2SLS using these six variables as instruments for the three endogenous
regressors. As before, evaluate the coefficient on attendance of violent movies and test whether
it is different from the coefficient on mildly violent movies, and then again for non-violent
movies.
 Coefficient: -.0031738; meaning that for every 1 million attendance of strongly violent
movies, the predicted number of assault decreases by 0.317%, ceteris paribus.
To find whether the coefficient on attendance of violent movies and mildly violent movies,
we use identical technique from question 1(d) and by regressing it using ivreg in STATA, it
was found that the p-value is 0.522; meaning that we fail to reject the null that the effect of
attending violent movies is not different from attending mildly violent movies and say that
there may be an effect between attending violent movies or mildly violent movies.

Similarly, to find whether the coefficient on attendance of violent movies and non-violent
movies, we use similar technique from question 1(d) and by regressing it using ivreg in
STATA, it was found that the p-value is 0.664; meaning that we fail to reject the null that the
effect of attending violent movies is not different from attending non-violent movies and say
that there may be an effect between attending violent movies or non-violent movies.

(e) Perform a test for overidentification. What do you conclude about the instruments?

estat overid

Tests of overidentifying restrictions:

Sargan (score) chi2(3) = 8.58313 (p = 0.0354)

Basmann chi2(3) = 8.03479 (p = 0.0453)

We found that both the p-value is statistically significant at 5%; thus, we reject the null that our
instrumental variable is not correlated with the error term and say that at least 1 instrument is not
exogenousmeaning that one of our instruments is invalid as we want Cov(Z,u) = 0