ECON 0150 | Economic Data Analysis

The economist’s data analysis skillset.

Part 2.3 | Numerical by Numerical by Category

Economic Realignment

Has American politics become more geographically divided along economic lines?

• Do wealthy and poor counties vote differently now than twenty years ago?
• This is the kind of question you might explore in a final project.
• This question requires combining data from multiple sources.

Question: Has the relationship between income and voting changed?

The Data

Two datasets, each answering different questions

Dataset 1: Income by County

• Median household income for each U.S. county
• From the Census

Dataset 2: Elections by County

• Democratic vote share for each county
• Multiple years: 2000, 2004, …, 2024
• Includes total votes cast (proxy for county size)

Question 1: Do Counties Vote Differently Today?

Compare county distributions between 2000 and 2024.

This is a Part 2.2 (or 1.4) question: a numerical variable across categories.

# Use a box+stripplot
sns.boxplot(data_long, x='Year', y='dem_share', whis=(0,100))
sns.stripplot(data_long.sample(400), x='Year', y='dem_share')

Democratic Vote Share by Year

Counties shifted Republican on average

> median dropped from 41% to 29% — but why were elections still close?

Question 2: Do Large Counties Vote Differently?

Lets try to explaining the paradox.

• Counties aren’t equal in population.
• Small rural counties outnumber large urban ones.

Question: Do larger counties lean Democratic?

This is a Part 2.1 question — a relationship between two numerical variables.

# Use log votes
data['log_votes_2024'] = np.log(data['total_votes_2024'])

# Plot using lmplot (scatterplot is good too)
sns.lmplot(data, x='log_votes_2024', y='dem_share_2024')

Do Large Counties Vote Differently?

Large counties vote more Democratic

> Democratic-leaning counties are larger (correlation: 0.49)

Question 3: Is This Related to Income?

Large counties tend to be richer. Is income related to voting?

Question: Is the Democratic lean related to income?

This is another Part 2.1 question.

# Use log income
data['log_income'] = np.log(data['income'])

# Use lmplot for a best fit line
sns.lmplot(data, x='log_income', y='dem_share_2024')

Income and Democratic Voting (2024)

Richer counties vote more Democratic

> correlation: 0.25 — but was it always this way?

Question 4: Has the Relationship Changed?

Have higher income counties always leaned Democratic?

Prior tools fall short:

• Part 2.1 (scatter): Shows one year at a time
• Part 2.2 (boxplot): Ignores the income relationship entirely

> neither tool shows how the relationship between two numerical variables varies by category

New Tool: Scatter by Category

Color points by year to see how the relationship differs

We need a scatter plot that shows bivariate relationships separately by category.

# Log income in the long format dataset
data_long['log_income'] = np.log(data_long['income'])

# Use lmplot with hue for the category
sns.lmplot(data_long, x='log_income', y='dem_share', hue='Year')

Income and Democratic Voting: 2000 vs 2024

The relationship has flipped

> in 2000, richer counties leaned Republican; by 2024, they lean Democratic

The Insight

Each question built on the previous

• Q1: Counties shifted Republican on average
• Q2: But large counties vote Democratic (explains why elections are close)
• Q3: Richer counties vote Democratic
• Q4: This relationship completely flipped since 2000

Merging Datasets

Combining income and election data

To answer Q2-Q4, we needed data from multiple sources.

data = pd.merge(income, elections, on='county_fips')

> the merge matches 3,106 of ~3,200 counties — 97% success

The Workflow

A template for final projects

1. Q1 (Part 2.2): Boxplot by year — counties shifted Republican
2. Q2 (Part 2.1): Scatter of size vs. voting — large counties vote Democratic
3. Q3 (Part 2.1): Scatter of income vs. voting — richer counties vote Democratic
4. Q4 (Part 2.3): Scatter by category — the relationship flipped

Part 2.3 | Summary

• Part 2.1 tools (scatter) show relationships between two numerical variables
• Part 2.2 tools (boxplot) show how a numerical variable differs by category
• Part 2.3 tools (scatter by category) show how a relationship differs by category
• Merging combines datasets when your question requires variables from both

Building Blocks

What this unit adds to your toolkit

Block	Part 2.3
Variables	Numerical + Categorical
Structures	Cross-section, Panel
Operations	Merge, Reshape (melt), Add hue
Visualizations	Scatter by category (lmplot with hue)