1. We are interested in whether being contacted (that is, mobilized) by a campaign prior to the 2020 election made survey respondents more likely to vote. Our dependent variable is whether the individual voted (1=yes, 0=no). (Note: there are more advanced statistical approaches for dealing with dichotomous dependent variables. However, they rarely lead to meaningfully different results. In this case, think about predicted values from the regression as the predicted probability that an individual turns out to vote.)

    We estimate the predicted probability of voting based on whether the individual reports being contacted by either the Democratic or Republican campaign (1=yes, 0=no). Doing so yields:
Effect of Contact from Either Campaign on Turnout
Predictor B SE t p
Constant 0.662 0.012 53.38 <0.001
Contact from Either Campaign 0.214 0.018 11.87 <0.001


Two other factors that political scientists believe might affect turnout are how strongly people feel about the political parties (because more committed partisans are more likely to turn out because they feel more strongly about the outcome) and education (because more educated people are more likely to be knowledgeable about political matters). We estimate a new model adding strength of party affiliation and education to the bivariate “contact” model from question 2.

Strength of party affiliation is measured as; 0=pure independent; 1=leaner; 2=weak identifier; 3=strong identifier.

Education is measured as: 1=no high school; 2=some HS, no diploma; 3=HS diploma; 4=some college; 5=associate’s degree; 6=bachelor’s degree; 7 advanced degree

Doing so yields:

Effect of Contact from Either Campaign on Turnout
Predictor B SE t p
Constant 0.233 0.027 8.63 <0.001
Contact from Either Campaign 0.154 0.017 9.03 <0.001
Strength of Party Affiliation 0.117 0.008 14.23 <0.001
Education 0.061 0.005 11.14 <0.001
  1. Interpret the coefficient for the constant in the multivariate regression model. [1 point]

  2. Interpret the coefficient for the “Education” independent variable in the multivariate regression model. [2 points]

  3. Discuss what happened to the coefficient for the “contact” variable (in the bivariate regression) when you added the two new variables. What might explain this change? [3 points]

  1. The relationship between education and turnout may not be linear. As we discussed in class, we can assess this possibility by measuring education using a series of indicator (or dichotomous) variables. We rerun the analysis from question #2 replacing the education measure with these indicator variables. We use “bachelor’s degree” as the reference category using the following linear equation:

\[Turnout = \beta_{contact} + \beta_{noHS} + \beta_{noHSdip} + \beta_{HSdip} + \beta_{somecollege} + \beta_{associates} + \beta_{advanced}\]

This gives us the following [SEE NEXT PAGE]:

Effect of Contact from Either Campaign on Turnout
Predictor B SE t p
Constant 0.587 0.028 20.90 <0.001
Contact from Either Campaign 0.154 0.017 9.05 <0.001
Strength of Party Affiliation 0.117 0.008 14.24 <0.001
Education: No HS -0.304 0.047 -6.46 <0.001
Education: Some HS, no diploma -0.305 0.034 -8.84 <0.001
Education: HS Diploma -0.162 0.026 -6.22 <0.001
Education: Some College -0.058 0.029 -2.05 0.041
Education: Associate’s Degree -0.056 0.032 -1.75 0.081
Education: Advanced Degree 0.031 0.039 0.81 0.419
  1. Interpret the coefficient for the constant. [1 point]

  2. Interpret the coefficient for the “Education: Advanced degree” variable. [2 points]

  3. Calculate the predicted value (probability of voting) for a weak identifier with a high school diploma who has been contacted by either campaign. (Hint: write out the equation.) [1 point]

  4. How would you test whether individuals with “some college” were more likely to vote than those with an “associate’s degree?” [2 points]

    To do this we would need to rerun the model after swapping bachelor’s degree and the new excluded category of some college. This would give us the following linear model:

    \[Turnout = \beta_{contact} + \beta_{noHS} + \beta_{noHSdip} + \beta_{HSdip} + \beta_{associates} + \beta_{bachelors} + \beta_{advanced}\] This new model gives us the following output:
    Effect of Contact from Either Campaign on Turnout
    Predictor B SE t p
    Constant 0.528 0.025 21.35 <0.001
    Contact from Either Campaign 0.154 0.017 9.05 <0.001
    Strength of Party Affiliation 0.117 0.008 14.24 <0.001
    Education: No HS -0.245 0.046 -5.35 <0.001
    Education: Some HS, no diploma -0.246 0.033 -7.53 <0.001
    Education: HS Diploma -0.104 0.024 -4.35 <0.001
    Education: Associate’s Degree 0.002 0.031 0.07 0.946
    Education: Bachelor’s Degree 0.058 0.029 2.05 0.041
    Education: Advanced Degree 0.090 0.038 2.39 0.017


    As you can see, in each case the constant changes to reflect the new reference category.