Data Investigation · United States · 2014–2023

Does Poverty
Drive Gun Violence
in America?

Every year, tens of thousands of Americans are killed or injured by gun violence. The causes are complex — but two factors are consistently suspected: poverty and population. This investigation sets out to find the truth.

By Cynthia Mutua & MS Data Science & Analytics, Grand Valley State University
Read the Full Report ↗ View Code on GitHub
Scroll to explore
"High-poverty areas have gun violence rates that are statistically significantly higher — confirmed across permutation tests, t-tests, and 10,000 bootstrap samples. Population alone tells us nothing."
— Key finding from this analysis
100K+
Incidents Analyzed
Gun Violence Archive, 2014–2023
p<.05
Poverty Significance
Permutation test result
10
Years of Data
Across all 50 U.S. states
10K
Bootstrap Samples
For confidence interval estimation
Chapter One

The Question That Motivated This Study

Gun violence in the United States is not random. It clusters — by geography, by season, by circumstance. Some states have far higher rates than others. Some cities see violence spike in summer months. Some communities are hit harder than others, year after year.

The conventional wisdom points to two suspects: poverty and population. Larger cities, the thinking goes, have more violence simply because they have more people. And poorer communities, the research suggests, face structural conditions that make violence more likely.

But which one actually matters — statistically? Is it just a numbers game, or is something deeper at work? This study was designed to find out.

We combined two rich datasets — the U.S. Census American Community Survey (26 socioeconomic variables per state, 2008–2023) and the Gun Violence Archive (every recorded incident, 2014–2023) — to build a unified picture of where gun violence happens and why.

Rather than relying on simple correlation, we used rigorous statistical methods — permutation tests, bootstrap analysis, Welch's t-test — that don't require distributional assumptions and are robust to the skewed, real-world nature of violence data.

✦ ✦ ✦
Chapter Two

How We Built the Analysis

The study followed a careful five-step pipeline — each stage building evidence toward the central question.

1

Merging Two Worlds of Data

We joined gun violence incident counts with census economic data by year and state, then standardized everything into a gun violence rate per 100,000 population. This matters — without this step, states like California and Wyoming can't be fairly compared.

2

Testing Population First

We ran 1,000 permutations shuffling population values to ask: does population size predict gun violence rate? The answer was clear — observed correlation r = −0.052, p = 0.259. Population alone is not a significant driver.

3

Testing Poverty Next

The same permutation test applied to poverty rate told a very different story — p < 0.05. Poverty rate is a statistically significant predictor of gun violence rate, confirmed independently of population size.

4

High vs. Low Poverty Comparison

We split states into High and Low Poverty groups and compared their gun violence rates using Welch's t-test. The variability in High Poverty states (SD = 210.82) was dramatically higher than Low Poverty states (SD = 38.55) — a sign of deep instability in high-poverty communities.

5

Quantifying Uncertainty with Bootstrap

10,000 bootstrap samples gave us a 95% confidence interval of [11.66, 14.19] incidents per 100,000 for high-poverty states — a stable, robust estimate that doesn't rely on normality assumptions.

The Findings

What the data revealed

Six key findings emerged from this investigation — each telling part of the same story.

p < .05
Poverty Drives Gun Violence
Permutation test confirms poverty rate is a statistically significant predictor of gun violence rate across U.S. states.
p = .259
Population Does Not
Population size falls well within the null distribution — it is not a significant predictor on its own. Size doesn't explain the pattern.
[11.66–14.19]
Bootstrap Confidence Interval
95% CI for median gun violence rate in high-poverty states. A stable, robust estimate across 10,000 resamples.
5.36%
Armed Robbery Share
Of all filtered incidents, 5.36% involve armed robbery with injury — concentrated heavily in southern states like Texas, Florida, and Georgia.
Summer
When Violence Peaks
In California, gun violence deaths spike on summer weekends and holidays — a clear temporal pattern visible in the calendar heatmap.
SD 210 vs 38
Inequality in Variability
High-poverty states don't just have higher rates — they have far more unpredictable rates. The variance gap tells a story of instability.
The Methods

Statistical rigour behind the story

Every finding in this study is backed by a specific statistical test chosen for its robustness to real-world data conditions — including skewed distributions and unequal variances.

Method What it tested Result Significant?
Permutation Test
1,000 iterations
Population size → Gun violence rate r = −0.052, p = 0.259 Not Significant
Permutation Test
1,000 iterations
Poverty rate → Gun violence rate p < 0.05 Significant ✓
Welch's T-Test
Unequal variances
High Poverty vs Low Poverty gun violence rates SD: 210.82 vs 38.55 Significant ✓
Bootstrap CI
10,000 samples
Median rate in high-poverty states 95% CI: [11.66, 14.19] Stable Estimate ✓
The Visualizations

Seeing the data come alive

This project produced 12 figures — from interactive plotly charts to a calendar heatmap showing daily deaths across an entire year. Here are the highlights.

Figure 6 · usmap Package · 2023
US choropleth map of gun violence by state 2023
Gun Violence by State — 2023

Texas and California light up yellow — the highest incident counts nationally. The South and East Coast show the next concentration. Geography is not random — it follows the poverty map.

Figure 10 · ggplot2 · Poverty Groups
Density plot of gun violence rates by poverty group
High vs Low Poverty — The Gap

High-poverty states (red) have a wider, fatter tail — more variability, more extremes. Low-poverty states cluster near zero. The difference is stark.

Figure 12 · Bootstrap · 10,000 Samples
Bootstrap distribution of median gun violence rate
Bootstrap CI — [11.66, 14.19]

10,000 resamples confirm the estimate is stable. The blue lines mark the 95% confidence interval — the truth lies reliably between 11.66 and 14.19 per 100,000.

Michigan median income vs poverty rate over time
📉 Figure 4 · Michigan Case Study
Rising Income, Persistent Poverty

Michigan's median income has grown steadily since 2012 — but poverty rates haven't fallen at the same pace. Economic growth alone doesn't close the gap that drives violence.

See All 12 Visualizations in the Full Report ↗
Built With

Tools & Technologies

This analysis was built entirely in R using the Quarto publishing system, with a rich ecosystem of visualization and statistical packages.

R 4.3+ Quarto tidyverse ggplot2 plotly gganimate calendR usmap dplyr lubridate flextable gt + gtExtras broom janitor magick GitHub Pages

The answer is yes — and it matters.

Poverty is a statistically significant predictor of gun violence rates. Population is not. High-poverty states don't just have higher rates — they have more volatile, unpredictable rates. The 95% bootstrap confidence interval confirms a median of 11.66 to 14.19 incidents per 100,000.

This isn't just a statistical finding. It's a call for targeted, equity-focused intervention in communities that data consistently identifies as most at risk.

Read the Full Analysis ↗
← Back to Portfolio