Knowledge Goals:
• Investigate American opinions towards select policies and politicians.
• Examine how the partisan divide in America reflects or exaggerates opinion differences.
• Become more familiar with SDA statistical analysis program.
Statistical Goals:
• Learn how to complete & interpret frequency tables and cross-tabulations using SDA.
• Learn to identify different levels of measurement
• Communicate ‘typical’ opinions by applying some common descriptive statistics.
• Understand the implications of using (or not using) weights when interpreting survey analyses.
• Learn to test whether categorical variables (nominal and ordinal with few categories) are associated.
• Apply Chi-square tests of statistical significance and measures of association to bivariate analyses.
Part I
A. Go to http://sda.berkeley.edu/archive.htm unless you prefer to work with a software package like R, SPSS or STATA. Then go to http://electionstudies.org/ to download the data.
B. Choose “American National Election Study (ANES) 2016 – released 4/3/17 (SDA 4.0)” or, for the older version, click on (SDA 3.5) immediately to the right. These instructions are for SDA.
C. In the bottom-left, select the folder “Pre-Election Early Voting”
D. Then, click on variable v161021a: PRE: For which candidate did R vote in the primary?
E. Click the button that says “row” in the top-left. You should see v161021a on the right in the field box next to where it says “required” in red (you can also type, or cut and paste v161021a from this document into that field).
F. For weight, choose, “pre-election weight, full sample.”
G. For “output options,” selected column percentages, weighted N, summary statistics and deselect “color coding” (unless you prefer otherwise).
H. For “chart options,” select “no chart” (you can select a chart, but it will not make a difference to your analysis).
- What percent of the sample voted for Ohio Governor John Kasich for President in the primaries?
a. 4.4%
b. 5.9%
c. 21.4%
d. 22.6% - True (a) or False (b): More respondents voted in the primaries for Bernie Sanders than voted for Donald Trump in the primaries.
- At what level of measurement is this variable?
a. Nominal
b. Ordinal
c. Interval or Ratio - The average (mean) response is 3.10 (close to “3-Another Democrat”), the median is 2 (Bernie Sanders). Given the level of measurement of this variable, does it make sense to report either of these statistics as a measure of how respondents generally behaved?
a. No, the variable is nominal, so the mean (Another Democrat) or the median (Bernie Sanders) makes no sense as a measure of how respondents generally behave; only the mode (1 – Hillary Clinton) means anything.
b. Yes; the median (2- Bernie Sanders or the mode, 1- Hillary Clinton) can be reported as a measure of how respondents generally behaved, but not the mean.
c. Yes; the mean or the median can be reported as being a measure of how respondents generally behaved.
d. Yes; the mean (close to “3-Another Democrat”) can be reported as a measure of how respondents generally behaved, but not the median.
I. With all the talk lately about President Trump’s frequent disregard for the truth, look at two variables found under the category “Candidate Traits”: v161162 and v161167. The former measures how well survey respondents consider the Democratic candidate, Hillary Clinton, to be honest, and the latter measures how well it describes the then-Republican nominee, Donald Trump. Run frequency tables for both variables, making sure “summary statistics” is chosen to answer the following questions and the pre-election weight is included: - What is the modal response (the most common response) to both questions?
a. “Honest” describes the candidates as extremely well (1).
b. “Honest” describes the candidates as very well (2).
c. “Honest” describes the candidates as moderately well (3).
d. “Honest” describes the candidates as slightly well (4).
e. “Honest” describes the candidates as not well at all (5). - On average, which candidate was seen as less honest? Remember lower numbers on the scale mean more honest.
a. Trump (the Republican)
b. Clinton (the Democrat)
c. Both were seen as being equally [dis-] honest.
J. With v161162 in the row field, click on the category “Party Identification” (located immediately above “Candidate Traits”) and add variable v161158x – “PRE:SUMMARY – Party ID” to “Column” by clicking the button in the top left marked “COL” or manually entering v161158x into the column field (below row) on the right. This will create a bivariate cross-tabulation between the question that asked whether Secretary Clinton was honest and party identification. We can use this cross-tabulation to see whether Republicans are more or less likely than Democrats or Independents to view Secretary Clinton as honest. Make sure only column percentages are selected under “output options.” - What percentage of Strong Republicans think that “being honest” describes Secretary Clinton “not well at all”?
a. 7.7%
b. 15.2%
c. 49%
d. 89% - What percentage of Strong Democrats think “being honest” describes Secretary Clinton “not well at all”?
a. 7.7%
b. 15.2%
c. 49%
d. 89%
K. Chi-square is a basic significance test that checks to see whether the distribution of the observations are significantly different across the columns. The actual observations are compared to what would be observed if the observations fit the pattern predicted by the null hypothesis. - True (a) or False (b): According to the Chi-squared statistic, can we be sure that the differences we observe between the categories of party identification are significant.
L. Measures of association, like correlation (for two continuous variables), Cramer’s V (for one nominal and one ordinal variable, or two nominal variables) or Kendall’s Tau-B (for two ordinal variables) go a step further to see to how well one can predict the values of the dependent variable if you know the values of the independent variable - What does a Kendall’s Tau-C of 0.50 tell us about the relationship between Republican party identification and thinking that “being honest” describes Secretary Clinton?
a. A Kendall’s Tau-C of 0.50 tells us that (generally) the more of a Republican the respondent is on a seven-point party identification scale, the more likely the respondent will think that Secretary Clinton is not honest at all.
b. A Kendall’s Tau-C of 0.50 tells us that (generally) a respondent’s party identification has little bearing on how likely it is that the respondent will think that Secretary Clinton is not honest at all.
c. A Kendall’s Tau-C of 0.50 tells us that (generally) the less of a Republican the respondent is on a seven-point party identification scale, the more likely the respondent will think that Secretary Clinton is not honest at all.
M. With v161167 in the rows and v161158x in the columns, answer these questions: - What percentage of Strong Republicans think that “being honest” describes Mr. Trump “not well at all”?
a. 7.1%
b. 10.9%
c. 43.7%
d. 77.8% - What percentage of Strong Democrats think that “being honest” describes Mr. Trump “not well at all”?
a. 7.1%
b. 10.9%
c. 43.7%
d. 77.8% - What is the Kendall’s Tau-c?
a. -0.44
b. -0.56
c. 0.00
d. 0.50 - True (a) or False (b): There is an inverse relationship between party identification and thinking that Mr. Trump is not at all honest.
- What sentence best describes the relationship between party identification and perceptions of candidate honesty in 2016?
a. Partisans and non-partisans alike thought both candidates were dishonest in 2016.
b. Partisans thought their candidate was honest but that their opponent was dishonest in 2016.
c. Partisans and non-partisans thought Donald Trump was a liar, but not Hillary Clinton.
d. Partisans thought that the candidates were relatively honest, but independents thought both candidates were dishonest.
Do men and women perceive the candidates’ honesty similarly? Please take a look at the mean ratings of honesty for both Clinton and Trump by first clicking on the tab on the right that will take you away from the options for frequency tables or cross-tabulations to do a comparison of means. To do that, click on the tab next to “Tables” called “Means” (see arrow points to tab in picture below).
N. First, put the rating of whether Clinton is honest, v161162, in the field for “Dependent [Variable].” We will look at differences between men and women, so we will need to put variable v161342 in the rows. Because there are a small number of “other” gendered responses in the dataset that we will ignore as part of this analysis (simply to keep the analysis looking at two, large sub-population groups), we will need to tell SDA to only look at men and women (value categories 1 and 2, respectively) . To do that, in the Row field on the right, we will type: v161342(r:1 “Male”;2 “Female”).
O. In the bottom right, we will need to select the correct output options. Reveal “output options” (see picture below). Make sure Complex Std Errs (Standard Errors) is NOT checked. Check Std Dev (Standard Deviation) and the box next to P-Value (see picture) and “ANOVA Stats.”
P. In the output, instead of percentages, you will see the mean ratings of how honest men and women perceived Clinton to be. Below that table you should see a table that will look like this:
Q. Look at the value under “P” at the far right. If that number is less than 0.05, we can be better than 95% confident that the differences in means observed are not due to chance. In this example, .000 indicates that the differences in between men and women regarding Clinton’s honesty, are statististically significant.
Remember that high values means that the candidate is perceived as dishonest.
- True or False: Compared to women, men, on average, perceived Clinton as more dishonest.
- True or False: The difference between how men and women, on average, perceived Clinton’s honesty was less than one point on a five point scale.
R. Finally, re-run the comparison on whether Trump is seen as honest using v161167 as the dependent variable.
- True or False: Compared to women, men, on average, perceived Trump as more dishonest.
- True or False: The difference in between how men and women, on average, perceived Trump’s honesty was less than a ½ point on a five point scale.
- True or False: The differences in perceptions of Trump’s honesty (by gender) is statistically significant (p is less than 0.05).
Well done! Submit your answers on Blackboard and then click back on the “Tables” tab to continue to Part II ->
Part II:
Now it is your turn to examine differences between Republicans and Democrats. To do so, put v161158x in the columns while running cross-tabulations between that variable and two dependent variables that should be either a) some behavior that respondents have engaged in, like voting or watching Fox news, b) some opinion about policy or candidates. For each of the two cross-tabulations, answer the following:
i. What do you expect to find when you run the cross-tabulation? Why? (2 points)
ii. How would you characterize the distribution of opinion measured by the dependent variable you chose? What is the mean, median and/or mode of the dependent variable (only report what is relevant, given the level of measurement of the variable)? (2 points)
iii. What, if anything, strikes you when you observe the table between party identification (v161158x) and the variable you chose? How would you characterize the relationship between party identification and the dependent variable? How strong is the relationship between party identification (v161158x) and the dependent variable? (5 points)
iv. Did what you find surprise you, or were your findings consistent with your expectations? Explain. (3 points)
After running the two crosstabulations (remember that both should have party identification v161158x in the columns), consider the results from both of your tables and answer:
v. What do you think is the cause of any observed differences between Republicans and Democrats (or lack thereof)? Does party identification drive opinion or behavior, or does opinion/behavior amplify differences between partisans?
Sample Solution
