POL 51: Scientific Study of Politics
Professor Jones
Fall 2019
Problem sets 2 and 3: Inferential statistics, visualization of data, and regression modeling.
Problem set 2 involves questions 1-3 and problem set 3 involves questions 4-7. Problem set 2 is
due November 26. Problem set 3 is due December 5.
Overview: This assignment will utilize a novel data set I recently collected. The survey itself
involves an experiment, and the experimental component will be analyzed in this homework
set. You will use R to compute some very straightforward inferential statistics, plots, and a
multiple regression model. You will have ample time to do this homework assignment;
however, do not wait until the last minute to complete it. The goal of the assignment is to
increase your familiarity with R, testing hypotheses, and interpreting political science/political
psychology data.
Data overview: In late September and early October of 2019, a national sample of white (non-
Latino) Americans was undertaken. In all, about 3,100 individuals completed the survey. The
survey was designed to be regionally representative of the white population from the nine
Census regions, as well as representative of the age distribution of the American white
population. The survey imposed quotas such that we oversampled partisan identifiers
(Republican and Democrat). About 40 percent of the identifiers are Republican and about 40
percent are Democratic. The remaining 20 identify as independent. Quotas were imposed such
that about 50 percent of the sample identified as male and 50 percent identified as female.
Details on these quotas will be discussed in class.
Experimental component: In studies of communication and mass media, scholars are often
interested in “framing.” Framing entails the concept of how an issue is presented—or framed—
to the public by politicians or the news media. In this study, respondents were randomly
exposed to one of six experimental conditions. These conditions are in the codebook and in
class, I will discuss the conditions. Of interest here is whether or not attributions of blame for
migrant deaths are sensitive to framing. In other words, is it possible to “push” individuals up
or down on the blame attribution scale.
Theory: This survey is a component to my research agenda examining attitudes about the
immigration issue, writ large. In recent years, I’ve come to be interested in the concept of
“blame attribution;” that is, the reasons people think about for why bad things happen. There
are two ways to think about blame: dispositional and situational. “Dispositional blamers” are
more prone to blame the individual for his/her plight/problems; “situational blamers” are more
prone to blame systemic causes for why negative outcomes occur. This concept was explored
deeply in homework 1.
In this study, I am interested in how beliefs about the legitimacy of the U.S. political system is
elated to the willingness to endorse or oppose dispositional attributions of blame. This theory
emanates from social psychology and the work of Jost and his associates. As a precursor to
eginning this assignment, you should all read Jost (2019; posted on Canvas) to better
understand the theory. I STRONGLY recommend reading this.
In a nutshell, SJT says that individuals who endorse the legitimacy of the political system are
more likely to accept or tolerate outcomes that may be non-democratic, unfair, or lead to
heightened inequality. Further, high system justifiers seem to be more willing to support public
policies that may actually have a negative impact on themselves. In my survey, I ask a battery
of questions that measures the concept of system justification. These items are then combined
in the form of an additive scale such that higher scores reflect greater endorsement of system
justifying beliefs and lower scores means lower endorsement of such beliefs.
One question you will consider is the relationship between SJ beliefs and blame attribution for
migrant deaths on the U.S.-Mexico border. To measure blame attribution, I relied on a series of
pairwise, forced-choice comparison questions where respondents were asked to state which
“reason” was most to blame. For each pair, the respondent could choose an item that was
“systemic” or “dispositional.” I scaled together these questions to form a measure such that
higher scores on the scale implies greater endorsement of dispositional blame and lower scores
imply lower endorsement of dispositional blame (i.e. higher endorsement of situational blame).
In addition to these two scales, I asked about a respondent’s partisan affiliation, gender, their
support for different kinds of immigration policies, and an assessment of the size of the
Latina/o population that is undocumented, all variables that were used to some extent in the
first problem set.
In addition to these factors, I’m also interested in how framing influences assessment of blame
attribution.
R Tasks
For this assignment, you will need to do the following tasks in R. Before you begin, create a
dummy variable for party affiliation coded 1 if a respondent scores a 5 or higher on the variable
named “pid7” and coded 0 if a respondent scores a 3 or lower on the variable “pid7.” I will
explain in class what this coding means (spoiler alert: this is a different coding than used in
problem set 1).
1. This set of questions looks at the system justification scale, the blame attribution scale, and
the undocumented population estimate variable and asks whether or not there are differences
in them due to party affiliation and gender identification. Test the following hypotheses using a
t-test:
a. Republican identifiers will be more likely to endorse stronger system justifying beliefs than
Democratic identifiers. In your write up, formally state the null and alternative hypotheses.
What does your test show and, given the results, do you have sufficient evidence to reject the
null?
. Males will be more likely to endorse stronger system justifying beliefs than females. In your
write up, formally state the null and alternative hypotheses. What does your test show and,
given the results, do you have sufficient evidence to reject the null?
c. Republican identifiers will be more likely to endorse dispositional attributions of blame than
Democratic identifiers. In your write up, formally state the null and alternative hypotheses.
What does your test show and, given the results, do you have sufficient evidence to reject the
null?
d. Males will be more likely to endorse dispositional attributions of blame than females. In your
write up, formally state the null and alternative hypotheses. What does your test show and,
given the results, do you have sufficient evidence to reject the null?
e. Republican identifiers will be more likely to overstate the size of the Hispanic undocumented
population than Democratic identifiers. In your write up, formally state the null and alternative
hypotheses. What does your test show and, given the results, do you have sufficient evidence
to reject the null?
f. Males and females will be exhibit significant differences in their estimates of the size of the
Hispanic undocumented population. In your write up, formally state the null and alternative
hypotheses. What does your test show and, given the results, do you have sufficient evidence
to reject the null?
g. Create a dot chart of undocumented population estimates for males and females. How does
this chart related to 1f?
2. This set of questions asks you to examine the experimental conditions and how they relate to
lame attribution. Please do the following tasks.
a. Do a side-by-side box plot of the blame attribution scale for the six experimental conditions
for:
i. all respondents
ii. Republicans
iii. Democrats
What do the plots show? Are there important party differences that you see over experimental
conditions. In what ways do Democrats and Republicans look differently?
. Conduct a t-test testing the following hypotheses:
i. Republicans exposed to the climate na
ative will be significantly less likely to endorse
dispositional blame compared to Republicans exposed to the high-crime na
ative. In
your write up, formally state the null and alternative hypotheses. What does your test
show and, given the results, do you have sufficient evidence to reject the null?
ii. Democrats exposed to the climate na
ative will be significantly less likely to endorse
dispositional blame compared to Democrats exposed to the high-crime na
ative. In
your write up, formally state the null and alternative hypotheses. What does your test
show and, given the results, do you have sufficient evidence to reject the null?
iii. Republicans exposed to the visa na
ative will be significantly less likely to endorse
dispositional blame compared to Republicans exposed to the high-crime na
ative. In
your write up, formally state the null and alternative hypotheses. What does your test
show and, given the results, do you have sufficient evidence to reject the null?
iv. Democrats exposed to the visa na
ative will be significantly less likely to endorse
dispositional blame compared to Democrats exposed to the high-crime na
ative. In
your write up, formally state the null and alternative hypotheses. What does your test
show and, given the results, do you have sufficient evidence to reject the null?
3. Estimate a regression model treating blame attribution as a function of the experimental
conditions. From this model, provide the predicted value for blame attribution for each
condition. Do any of the experimental conditions seem to produce different prediction and if
so, in what direction?
4. Estimate a regression model treating blame attribution as a function of a dummy variable