Analysis of Political Data

Large N & Leeuwenhoek (70700173)

Yue Hu

Overview

Instructor introduction
Course introduction
Decision instrument

Instructor

Who am I

(And what you should [not] learn with me)

Experience

PhD. of political science
- Informatics
Computational Social Science Platform

Research：Microfoundations of Modernization

W. Economics & CSS
- Lab & survey experiment
- Latent variable
- Text analysis, visualization

Area

W. Psychology
- Micro-foundation
- Traits → behavior
W. Economy
- Perceptive inequality
W. Linguistics
- Political influence of language proficiency

Course

What a class

Not (that) methodologistic
- Equations, though
- Statistics, though
- Operations, though

Very basic
- (1 + 2 + 3 + 4) / 4
- OLS

Considerable load
Unmuddlable tasks (assignments, project, interactions)

OLS, that’s it

Association modeling > Generalized linear model > OLS

Morucci, Marco, and Arthur Spirling. 2024. “Model Complexity for Supervised Learning: Why Simple Models Almost Always Work Best, And Why It Matters for Applied Research.”

we start with a “full size” model, and then move through eight smaller ones for dimensions d = 50, 100, 250, 500, 1000, 2000, 3000, 5000. As we do so, we record several performance statistics (Accuracy, Precision, Recall, AUC) on the test set for each intrinsic dimension.

the horizontal black solid line records the “90%” solution—i.e. when we obtain 90% of the best possible accuracy/precision/recall and AUC achieved under any number of parameters for this problem.

But this pattern is not nearly so starkly the case for Montgomery, Goldstone, Beck et al and Colaresi data. We do see some improvements as intrinsic dimension increases, but the lines are basically flat after an (log) intrinsic dimension number of 8 or so.

Map of the Class

The Leeuwenhoek

\[ \begin{aligned} \sigma^2 =& \sum(X - \mu)^2p(x),\\ =& \sum(X^2 -2X\mu + \mu^2)p(x),\\ =& \sum X^2p(x) - 2\mu\sum Xp(x) + \mu^2\sum p(x),\\ =& \sum X^2p(x) - \mu^2. \end{aligned} \]

Meaning of equations
Traps in concepts, terms, and process
- What does randomization actually do?
- Do you need to worry about multicollinearity?
- How do you explain the interactive term in a conditional-effect model?

Structure

Lecture > 2/3

R Lab < 1/3

Gateway training (first 8 weeks)
- R basics
- Munging
- Basic Visualization
- Lecture related skills

Student-oriented Learning
- Project
- Map
- Regular expression
- Rstudio addin
- Version control
- Academic writing
- Slides
- Poster

Assignments

Assignments
- Three required
- Four to five questions
- Final project
  - Original paper
  - Replication (recommended)

Replication, replication

The only way to understand and evaluate an empirical analysis fully is to know the exact process by which the data were generated and the analysis produced. —King, Gary. 1995. “Replication, Replication.” PS: Political Science & Politics.