Module 7: Statistical Significance

Lab 7: Effects on Political Polarization on Health

Click here to run the code for this lab yourself on RStudio.Cloud!

Background and Instructions

Does political polarization affect our health? Since the 2016 election, pundits, scholars, and ordinary citizens have been talking about a growing feeling of unease and anxiety among the public about politics. The culprit, some argue, is political polarization. Some communities are deeply polarized, meaning some residents feel their political views are very different from their peers. In these communities, they argue, residents might be less likely to go outside, check in on older or vulnerable neighbors, family, and friends, and in the long run, we may see decreases in both our physical and mental well-being. Recently, scholars at Northeastern University surveyed respondents across the country and found that more political polarization in a community did hurt health! But, residents with stronger social capital - social ties between friends, family, and neighbors - saw better health, even in polarized communities. These findings relied on a survey sample - what about in the population at large?

Fortunately, the University of Wisconsin’s 2020 County Health Rankings dataset ranks communities based on the Behavioral Risk Factor Surveillance System (BRFSS) survey, a massive survey of samples of residents in every US county. We can use its measures to measure residents’ self-reported physical and mental health outcomes, as well as social capital. Further, the MIT Election Lab has collected 2016 Presidential Election outcomes in every county, allowing us to estimate poltical polarization in each community.

The Lab: This lab investigates why some US counties experience worse health outcomes than others. You will conduct several experiments on county data using the t_test() function in the infer package in R. Please import and explore the dataset, with instructions in each of the tabs below. You have been commissioned to test two main hypotheses:

  1. Is political polarization linked to worse health outcomes?

  2. In communities that are polarized, is greater social capital linked to better health?


Data & Variables

Import Data

Load in the data in R using the code below.

# Load key packages
library(tidyverse)
library(infer)
library(viridis)

# Load in our dataset
counties <- read_csv("polarization_and_health.csv")

View Data

View 3 counties randomly sampled from the full population of US counties.

counties %>%
  sample_n(size = 3)
county fips state region gap_2016 polarization social_associations social_capital poor_fair_health days_poor_mental_health days_poor_physical_health frequent_phys_distress diabetes obesity physical_inactivity exercise_access life_expectancy pop_black poc urban party wealth democrat_2016 republican_2016
Morgan County 13211 GA South Atlantic 41.1 Low 10.86 Low 0.15 3.69 3.21 0.10 0.137 0.347 0.235 0.34 77.6 0.24 Low Rural Area Republican Above Median 28.1 69.2
Scott County 20171 KS West North Central 74.0 High 16.13 High 0.15 3.37 3.19 0.10 0.100 0.390 0.300 0.81 80.1 0.00 Low Rural Area Republican Above Median 10.7 84.7
Randolph County 29175 MO West North Central 50.9 High 12.03 High 0.20 4.40 4.56 0.14 0.153 0.404 0.328 0.53 75.7 0.06 Low Rural Area Republican Below Median 22.1 73.0

Codebook

Explore variable definitions below. Click on the dropdown menu to select variables by group.

Group of Variables

Health
  • poor_fair_health: percentage of surveyed residents who rated their health as either poor or fair on a 5 point scale (1 = poor, 2 = fair, 3 = good, 4 = very good, 5 = excellent). (Note: higher = worse health implications.)

  • days_poor_physical_health: average number of physically unhealthy days that residents reported in past 30 days (age-adjusted).

  • days_poor_mental_health: average number of mentally unhealthy days that residents reported in past 30 days (age-adjusted).

  • frequent_phys_distress: percentage of adult respondents reporting 14 or more days of poor physical health per month.

Polarization
  • gap_2016: the difference in the percentage of votes for Democrats vs. Republicans in a county in the 2016 presidential election. 0 indicates an even split, meaning low polarization, because no one group is particularly marginalized. 50 indicates one party is much greater than the other (75%-25% split), meaning strong polarization, because one group is marginalized. 90 indicates one party is almost completely dominant (90%-10%), meaning extreme polarization, because one group is a very tiny minority.

  • polarization: dichotomous variable showing whether there is a 50% or greater difference in votes for Democrats vs. Republicans in a given county (“High”), or less than a 50% difference in votes (“Low”).

Social Capital
  • social_associations: number of membership associations in a county, per 10,000 residents. This includes civic organizations, bowling centers, golf clubs, fitness centers, sports organizations, religious organizations, political organizations, labor organizations, business organizations, and professional organizations.

  • social_capital: dichotomous variable showing whether the rate of social associations in a county is above the national median, building high social capital (“High”) or below the national median, building lower social capital (“Low”).

Task 1: Difference of Means

Select 2 health outcomes from the 4 listed above, and produce the following 4 t-tests, using the instructions below.

  • Using the t_test() function in infer in R, please test the effect of polarization on your first health outcome, then test the effect of social capital on your first health outcome. [2 t-tests]

  • Then, repeat these two tests for your second health outcome. [2 t-tests]

Please report your results in a table in your word processor, following the format of the one below. There are 4 cells to fill in, 1 for each of your 4 tests. Please also appropriately label the table based on what health outcomes you chose.

Table 1: Effect of Treatment on Mean Health Outcomes

Explanatory
Variable
Meaning of
Treatment
% Poor or Fair Health
(percentage)
[My Health Outcome 2]
(units here)
Polarization ‘High’ = Polarization > 50% 0.25
(p = 0.02)
Estimate here
(p-value here)
Social Capital ‘High’ = Rate of Associations
above median
fill me in fill me in

Task 2:

For this next step, you will examine how social capital shapes the effect of polarization on health, using 4 t-tests.

  • Using the filter() function, zoom into just counties with high levels of social capital (where social_capital equals "High"). Use t_test() to test the effect of polarization on your 2 health outcomes in communities with high social capital. [2 t-tests]

  • Next, using the filter() function, zoom into just counties with low levels of social capital (where social_capital equals "Low"). Use t_test() to test the effect of polarization on your 2 health outcomes in communities with low social capital. [2 t-tests]

Please record your t-tests in a table in your word processor, using the following format: You have 4 cells to fill in, 1 for each of your 4 tests. Please also appropriately label the table based on what health outcomes you chose.

Table 2: Effects of Polarization on Health
depending on level of Social Capital

Explanatory
Variable
Meaning of
Treatment
Type of Counties
Examined
% Poor or Fair Health
(percent)
[Health Outcome 2 Here]
(unit)
Polarization ‘High’ = Polarization > 50% High
Social Capital
0.25
(p = 0.02)
1.13
(p = 0.50)
Polarization ‘High’ = Polarization > 50% Low
Social Capital
fill me in fill me in


Task 3: Interpretation

Analyze your results using the following questions.

  • Was high political polarization associated with improvements or reductions in health? Note that our health outcomes are reverse coded, where higher values mean worse health (eg. days of poor health). How likely is it that this difference was just due to chance?

  • Was social capital associated with improvements or reductions in health? How likely is it that this difference was just due to chance?

  • How does the effect of polarization on health vary depending on the level of social capital? Are they consistent, or varied? How so? Does it grow, shrink, change directions, etc?

  • Does the measure of health you chose matter? Were some measures of health more affected by polarization than others?

Hint: This is a very good way to report your results.



Task 4: Visualizations

Make 2 visualizations to accompany this analysis.

  • For your first visual, please use the counties dataset to visualize the difference in county health outcomes by polarization, on average, for 1 of the 2 health outcomes you examined. You may do this for any of the 5 comparisons we made related to polarization above. Please emphasize the difference between counties with high vs. low polarization, using the polarization variable, for communities with high vs. low social capital. You could use geom_jitter(), geom_boxplot(), geom_col(), or others. Please add meaningful labels, themes, and colors.

  • For your second visual, zoom into just 1 health outcome, and select all 3 t-tests where you tested the effect of polarization on health. Please manually collect your 3 difference of means into a data.frame. There should be 3 rows, 1 per test. Then, use this data.frame to make a bar plot that compares the size of the effects of polarization on health, depending on the level of social capital. Please add meaningful labels, themes, or colors.

(Extra Credit: For 0.5 points of extra credit, add a second panel to your visual that visualizes the other 3 t-tests for your second outcome. For another 0.5 points of extra credit, add confidence intervals using geom_linerange() with the lower_ci and upper_ci columns from your t_test() output. The final visual should depict 6 t-tests results.)



Submit!

Finally, please write up your results in a short 2-page report. Length does not include your tables or visualizations. Please be sure to structure your report using research questions and hypothesis, and fully describe any tests, tables, or visuals you did in your analysis.

Congrats! You’re done!