Social and Political Data Science: Introduction

Karl Ho

University of Texas at Dallas

Polarization of World Politics and Impact  on Voting  Behavior

Prepared for presentation at the Soochow University, Taipei, Taiwan, December 14, 2023

Speaker bio.

The Puzzle

  • Are Taiwanese voters getting tired of the two major parties or generally politics only driven by the two colors?

The Puzzle

  • Are Taiwanese voters getting tired of the two major parties or generally politics only driven by the two colors?

What would prompt you to ask your future son or daughter in law what party he or she supports?

Motivation

  • Affective polarization refers to the animosity between the parties or supporters of parties, according to recent reports and studies noting the increasing strong negative sentiments such as distrust and dislikes toward members of the other party in America (Iyengar, Sood, and Lelkes 2012, Iyengar et al. 2019).
  • We take a close look into Taiwanese voters' affect structure and how it shapes the partisan dealignment.
  • This study focuses on "affect" or generally the sentiments in decision making.
  • Specifically, we analyze the affective polarization in Taiwan.

Affective Polarization: Literature

  • Affective polarization refers to the animosity between the parties or supporters of parties, according to recent reports and studies noting the increasing strong negative sentiments such as distrust and dislikes toward members of the other party in America (Iyengar, Sood, and Lelkes 2012, Iyengar et al. 2019).  
  • Increasing political sectarianism, as termed in some recent studies, plagues American voters the most compared to the most polarized nations like Canada, New Zealand and Switzerland as out-party hate grows the fastest in the United States, according to the comparative study by Boxell, Gentzkow and Shapiro (2022).

Affective Polarization: Literature

Affective Polarization: Literature

Comparing 12 nations, Boxell et al. (2022) found the United States suffers the most in the two parties' affective distance, i.e. Democrats and GOP supporters hate each other more and more in recent years.

Affective Polarization: Taiwan

  • Like the American voters, the green and blue camp supporters drive themselves further away from the moderate center, with growing out-party animus against each other.
  • Outcries of deserting the two major parties invite third parties claiming providing third choice for voters supporting neither camp.

Approach

  • Questions to ask:
    • Do affects matter?  
    • How affects can be compared to other factors in determining partisanship (what party a person supports)?
  • Exploratory analysis

Method

  • Structural Equation Model: Affect structure
  • Machine learning approach in comparing issue and affect models
    • Generalized Linear Models (GLM)
    • Deep Learning (Neural Networks)
    • Automated Machine Learning
      • Affect vs. Issues

Data

  • Taiwan Institute for Governance and Communication Research (TIGCR) (https://tigcr.nccu.edu.tw/en/)
  • Surveys on media usage and political attitudes and political behavior since 2018.
    • Face-to-face, internet, telephone interviews
    • Longitudinal design and generally low rate of attritions (40%)
    • Five-wave panel (2018, 2019, 2020, 2021, 2022), 4,507 cases each year (total N=22,530)

Sentiments (sympathy, antipathy)

  • Point 0 represents 'no positive impression', and point 10 represents 'very positive impression'. From point 0-10, how many points will you grade KMT/DPP?

  • Point 0 represents 'no negative impression', and point 10 represents 'very negative impression'. From point 0-10, how many points will you grade KMT/DPP?

Sentiments (threat)

  • Point 0 represents 'not threatening to the normal development of the country', and point 10 represents 'very threatening to the normal development of the country'. From point 0-10, how many points will you grade KMT/DPP?

Sentiments (marriage)

  • How would you feel if you had a son or daughter who married a KMT/DPP's supporter?

Sentiments (antagonism, anxiety)

  • Which political party ever made you feel angry?
  • Which political party ever made you worried?

Sentiments (optimism, relief)

  • Which political party ever made you feel hopeful?
  • Which political party ever made you feel relieved?

Issues

  • Transformative justice
  • New southbound policy
  • Non-nuclear policy
  • Coal-fired substitution
  • Mandatory day off
  • Same-sex marriage
  • Long-term care
  • Pension reform
  • Cross-strait political negotiations
  • Peace agreement
  • Unification vs. Independence vs. Status Quo

Structure of the Negative Partisan Sentiments and Exclusivity: Correlation Heatmap (Ho et al 2022)

  • hatekmt/hatedpp:
    • Which of the following ethnic groups do you hate the most?
  • angrykmt/angrykpp:
    • Which political party ever made you feel angry?
  • threatkmt/threatdpp:
    • From point 0-10, how many points will you grade KMT/DPP as threatening to the normal development of the country?
  • negkmt/negdpp:
    • From point 0-10, how many points will you grade KMT/DPP from "no negative impression" to "very negative impression"?
  • marriedkmt/marriedpp:
    • How would you feel if you had a son or daughter who married a KMT/DPP's supporter? Very disappointed/upset to not disappointed/upset at all?

KMT Affect

DPP Affect

DPP Affect

KMT Affect

Most important variables: GLM

KMT

DPP

Deep learning model (NN)

KMT

DPP

MSE:  0.05650798
RMSE:  0.2377141
LogLoss:  0.190111
Mean Per-Class Error:  0.07756352
AUC:  0.9773588
AUCPR:  0.9645369
Gini:  0.9547177

MSE:  0.05053361
RMSE:  0.2247968
LogLoss:  0.1659664
Mean Per-Class Error:  0.08273317 AUC:  0.9781768
AUCPR:  0.9399842
Gini:  0.9563536

Automated Machine Learning: KMT

Variable importance heatmap shows variable importance across multiple models. Some models return variable importance for one-hot (binary indicator) encoded versions of categorical columns (e.g. Deep Learning, XGBoost). In order for the variable importance of categorical columns to be compared across all model types we compute a summarization of the the variable importance across all one-hot encoded features and return a single variable importance for the original categorical feature. By default, the models and variables are ordered by their similarity.

Automated Machine Learning: KMT

Variable importance heatmap shows variable importance across multiple models. Some models return variable importance for one-hot (binary indicator) encoded versions of categorical columns (e.g. Deep Learning, XGBoost). In order for the variable importance of categorical columns to be compared across all model types we compute a summarization of the the variable importance across all one-hot encoded features and return a single variable importance for the original categorical feature. By default, the models and variables are ordered by their similarity.

This plot shows the correlation between the predictions of the models. For classification, frequency of identical predictions is used. By default, models are ordered by their similarity (as computed by hierarchical clustering).

Automated Machine Learning: KMT

SHAP summary plot shows the contribution of the features for each instance (row of data). The sum of the feature contributions and the bias term is equal to the raw prediction of the model, i.e., prediction before applying inverse link function

Automated Machine Learning: KMT

Partial dependence plot (PDP) gives a graphical depiction of the marginal effect of a variable on the response. The effect of a variable is measured in change in the mean response. PDP assumes independence between the feature for which is the PDP computed and the rest.

Automated Machine Learning: KMT

Partial dependence plot (PDP) gives a graphical depiction of the marginal effect of a variable on the response. The effect of a variable is measured in change in the mean response. PDP assumes independence between the feature for which is the PDP computed and the rest.

Automated Machine Learning: KMT

Partial dependence plot (PDP) gives a graphical depiction of the marginal effect of a variable on the response. The effect of a variable is measured in change in the mean response. PDP assumes independence between the feature for which is the PDP computed and the rest.

Automated Machine Learning: KMT

Partial dependence plot (PDP) gives a graphical depiction of the marginal effect of a variable on the response. The effect of a variable is measured in change in the mean response. PDP assumes independence between the feature for which is the PDP computed and the rest.

Automated Machine Learning: KMT

Partial dependence plot (PDP) gives a graphical depiction of the marginal effect of a variable on the response. The effect of a variable is measured in change in the mean response. PDP assumes independence between the feature for which is the PDP computed and the rest.

Automated Machine Learning: KMT

Partial dependence plot (PDP) gives a graphical depiction of the marginal effect of a variable on the response. The effect of a variable is measured in change in the mean response. PDP assumes independence between the feature for which is the PDP computed and the rest.

Automated Machine Learning: KMT

Automated Machine Learning: DPP

Automated Machine Learning: DPP

Automated Machine Learning: DPP

Automated Machine Learning: DPP

Automated Machine Learning: DPP

Automated Machine Learning: DPP

Automated Machine Learning: DPP

Automated Machine Learning: DPP

Automated Machine Learning: DPP

The Big Takeaway

  • Emotions vs. Serious Issues

  • Highly correlated 

  • Affects prevail 

  • One more thing .......

The Big Takeaway

Which political party ever made you feel hopeful (讓您覺得台灣有希望)?

Which political party ever made you feel relieved (最放心)?

The Big Takeaway

Which political party ever made you feel angry (請問哪一個政黨的作風或作法最讓您生氣)?

The Big Takeaway

Which political party ever made you worried (哪一個政黨的作風或作法最讓您擔心)?

The Big Takeaway

Political Polarization: Three Party Structure

Polarization, media and hyperpartisanship

Polarization, media and hyperpartisanship

Hyperpartisanship and Talkshow:

Panel Vector Autoregression

VAR Granger Causality Wald test: Talkshow Granger-causes Hyperpartisanship

Eigenvalue Stability and Impulse-Response Plots for the Panel Vector Autoregression

Eigenvalue Stability and Impulse-Response Plots for the Panel Vector Autoregression

Eigenvalue Stability and Impulse-Response Plots for the Panel Vector Autoregression

In a nutshell

  • Different Structures of Party Support

  • More vague and unstable among KMT supporters

  • Critical election 2020: Game changing

  • The China factor

Conclusion and limitations

  • Preliminary findings on Affective Polarization

  • Finkel et al. explain the splits:

    1. Partisan sorting

    2. Rise of new media

    3. Political elites and politicians

  • Limitations:

    1. Vote functions

    2. Next hypotheses

    3. More data

Implications

  • Drivers of affect driven partisanship?

  • Rational vs. Emotional voting models

  • Party system in fluidity

  • Future of Democracy

Thank you!

Questions or suggestions will be most welcome!

Thank you!