Environmental Awareness & Carbon Footprint Analysis

STEM for Development — Introduction to Statistics & Data Science
Capstone Project | Group #2 | 2025

Members: Khadija Khatoon, Lia Tamara Cedeño Zambrano, Maryam Hampaei, Ofei Harriet Koryo
Instructors: Charles Opondo, Ruchir Raman, Habibah Tijani

---

Table of Contents

1. Introduction & Research Question
2. Data Summary and Exploratory Analysis
   • 2.1 Dataset Summary
   • 2.2 Distributions
   • 2.3 Other Visualizations and Interpretations
3. Inferential Testing
   • 3.1 Parametric Tests
   • 3.2 Non-parametric Tests
4. Regression Models & Interpretation
   • 4.1 Simple Linear Regression
   • 4.2 Multiple Linear Regression
   • 4.3 Logistic Regression
5. Insights
6. Recommendations

---

1. Introduction & Research Question

Climate change is one of the defining challenges of the 21st century. A major driver is the accumulation of greenhouse gases in the atmosphere. While industrial activity and government policy are often cited as the primary culprits, household-level behavior is a significant and frequently overlooked contributor.

Electricity and gas usage, transportation choices, waste generation, and consumption patterns all add meaningfully to the global carbon footprint. Everyday decisions — appliance use, plastic consumption, recycling habits, travel preferences — can either increase or decrease a household's environmental impact. Yet most households remain unaware of this.

This study examines how household lifestyle and environmental practices affect carbon footprint levels, using a dataset of 2,000 observations covering household energy use, waste management, transportation, and lifestyle choices. Findings are intended to help identify which behaviors matter most, and to inform policymakers and environmental organizations working on awareness and sustainable living initiatives.

Research question: To what extent do household lifestyle and environmental practices influence carbon footprint levels?

---

2. Data Summary and Exploratory Analysis

2.1 Dataset Summary

The dataset captures household-level environmental behaviors and carbon emissions across 2,000 observations and 12 variables — 5 categorical and 7 numerical.

Categorical variables:

• Household_ID
• Income_Level
• Waste_Recycling
• High_Emissions
• Environmental_Awareness_Score (ordinal)

Numerical variables:

• Household_Size
• Electricity_Usage
• Gas_Usage
• Plastic_Consumption
• Public_Transport_Usage
• Distance_to_Work
• Carbon_Footprint

---

2.2 Distributions

The following variables are approximately normally distributed (with noted skew):

Variable	Distribution
Electricity_Usage	Slightly skewed right
Gas_Usage	Slightly skewed right
Plastic_Consumption	Heavily skewed right
Public_Transport_Usage	Lightly skewed left
Distance_to_Work	Heavily skewed right

The following variables do not follow a normal distribution, confirmed via QQ plots:

1. Environmental_Awareness_Score — ordinal variable; discrete step pattern visible in QQ plot

2. Carbon_Footprint — deviation at tails confirmed by QQ plot

4. Household_Size — discrete values produce a stepped QQ plot

---

2.3 Other Visualizations and Interpretations

2.3.1 Distance to Work vs. Public Transport Usage (Scatter Plot)

Trend: Negative correlation — as public transport usage increases, commuting distance tends to decrease.

Interpretation: Households that rely more on public transport tend to live closer to their workplaces.

---

2.3.2 Carbon Footprint by Income Level (Box Plot)

Insight: Median carbon footprint is fairly consistent across income levels. Outliers are more prominent in the medium-income group.

Interpretation: Income alone is not a strong predictor of carbon emissions. The higher variance in medium-income households likely reflects external or behavioral factors not captured in this dataset.

---

2.3.3 Public Transport Usage by Income Level (Bar Chart)

Insight: Medium-income households use public transport the most; high-income households use it the least.

Interpretation: Public transport is predominantly used by medium- and low-income households. Lower uptake among high-income groups likely reflects higher rates of personal vehicle ownership and potentially longer or less transit-accessible commutes.

---

3. Inferential Testing

3.1 Parametric Tests

Independence and large-sample-size assumptions are considered met. Normality was assessed in the previous section; variables that are not normally distributed are excluded from parametric tests. Homoscedasticity is validated where relevant.

3.1.1 Two-Sample t-Test — Gas Usage & Waste Recycling

Research question: Is there a significant difference in gas usage (LPG kg) between households that recycle and those that don't?

A variance test was run first (p = 0.1421), confirming equal variance between groups. The box plot also supports this — both groups show a similar spread.

• H0: Gas usage is not significantly different between recyclers and non-recyclers.
• H1: Gas usage is significantly different between recyclers and non-recyclers.

Result: p = 0.02418 → reject H0. Gas usage differs significantly between the two groups. The bar chart indicates that households that recycle waste have lower average gas usage than those that do not.

---

3.1.2 Correlation Test — Electricity Usage & Gas Usage

Research question: Is there a significant correlation between electricity usage (kWh) and gas usage (LPG kg)?

• H0: No significant correlation between electricity usage and gas usage.
• H1: Significant correlation exists between electricity usage and gas usage.

Result: p < 2.2e-16 → reject H0. There is a strong positive correlation between the two variables, with a correlation coefficient of r = 0.95. As electricity usage increases, gas usage increases proportionally.

---

3.2 Non-parametric Tests

3.2.1 Chi-Square Test — Waste Recycling & Income Level

Research question: Is there a relationship between waste recycling behavior and income level?

• H0: Recycling rates do not differ significantly across income levels.
• H1: Recycling rates differ significantly across income levels.

Result: p = 2.2e-16 → reject H0.

	High	Low	Medium
Does not recycle	192	507	591
Recycles	208	144	358

Higher-income households are more likely to recycle; lower-income households are less likely. A Chi-Square test on Environmental_Awareness_Score vs. Income_Level yields the same result (p = 2.2e-16), reinforcing the presence of a socioeconomic factor.

---

4. Regression Models & Interpretation

4.1 Simple Linear Regression (Gas Usage ~ Electricity Usage)

Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 0.521428 0.142680 3.655 0.000264 *** Electricity_Usage 0.046183 0.000328 140.797 < 2e-16 *** Residual standard error: 2.621 on 1998 degrees of freedom Multiple R-squared: 0.9084 Adjusted R-squared: 0.9084 F-statistic: 1.982e+04 on 1 and 1998 DF, p-value: < 2.2e-16

Model formula:

$$y = 0.046183x + 0.521428$$

Interpretation:

• The coefficient for Electricity_Usage is 0.046183 (p < 2.2e-16) — statistically significant.
• For every additional kWh consumed per month, gas usage increases by approximately 0.046 LPG kg.
• The intercept (0.5214) is not practically meaningful — a household with zero electricity usage is not realistic.
• R² = 0.9084: electricity usage explains 90.84% of the variance in gas usage.
• Correlation coefficient R = 0.9531 — very strong positive relationship.

---

4.2 Multiple Linear Regression (Gas Usage ~ Electricity Usage + Carbon Footprint)

Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 0.4176353 0.1455540 2.869 0.004157 ** Electricity_Usage 0.0456221 0.0003666 124.462 < 2e-16 *** Carbon_Footprint 0.7233755 0.2130971 3.395 0.000701 *** Residual standard error: 2.614 on 1997 degrees of freedom Multiple R-squared: 0.909 Adjusted R-squared: 0.9089 F-statistic: 9970 on 2 and 1997 DF, p-value: < 2.2e-16

Model formula:

$$y = 0.0456221x + 0.7233755z + 0.4176353$$

where x = Electricity Usage (kWh), z = Carbon Footprint (tons/year)

Interpretation:

• Both predictors are statistically significant (p < 2.2e-16).
• For every additional kWh/month, gas usage increases by ~0.046 LPG kg.
• For every additional ton of CO₂/year, monthly gas usage increases by ~0.723 LPG kg.
• R² = 0.909: a marginal improvement over the simple model (0.9084), suggesting Carbon_Footprint adds minimal new explanatory power beyond what Electricity_Usage already captures.
• Correlation coefficient R = 0.9534.

---

4.3 Logistic Regression (High Emissions ~ Household Size)

Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) -3.48535 0.16825 -20.72 <2e-16 *** Household_Size 0.55737 0.03496 15.94 <2e-16 ***

Exponentiated coefficients:

$$e^{0.55737} = 1.746074 \quad \text{(Household Size)}$$ $$e^{-3.48535} = 0.03064303 \quad \text{(Intercept)}$$

Interpretation:

• Household size is a statistically significant predictor of high emissions (p < 2e-16).
• For each additional occupant, the odds of being a high emitter increase by 1.74× (74% increase per person).
• The intercept is not meaningful — a household with zero occupants is not a household.
• The model correctly captures the trend visible in the predicted probability plot: probability of high emissions rises steeply as household size increases.

---

5. Insights — What Factors Affect Carbon Footprint the Most?

• Energy usage is the dominant driver. Gas usage has a near-perfect positive correlation with electricity usage (r = 0.95). Households consuming more electricity also consume more gas, compounding their total carbon footprint.

• Household size significantly increases emissions risk. Each additional occupant raises the odds of a household being classified as a high emitter by 74%, likely driven by higher aggregate consumption of energy, food, and goods.

• Income shapes recycling behavior. Higher-income households are significantly more likely to recycle waste (p = 2.2e-16). Lower-income households show the inverse pattern — suggesting access, infrastructure, and education all play a role alongside motivation.

• Public transport usage is income-stratified. Medium-income households use it the most; high-income households use it the least, which likely reflects higher rates of private vehicle ownership in that group.

• Income is not a clean predictor of carbon footprint level. Median emissions are similar across income brackets, but medium-income households show more high-end outliers — possibly due to lifestyle variability that income alone doesn't capture.

---

6. Recommendations

Reducing household emissions

Transportation: Government investment in public transport quality and accessibility would shift uptake across income groups. For private transport, awareness campaigns focused on trip consolidation and carpooling can reduce per-household emissions without requiring large behavioral shifts.

Large households: Energy-efficient and smart appliances are a practical first step given the strong relationship between household size and emissions. Where feasible, investment in residential renewables (e.g. solar) offers longer-term reduction.

Recycling: Broad government-backed recycling initiatives — particularly those targeting lower-income communities — are needed. Deposit refund systems (such as Germany's Pfand system) provide a concrete, proven model that creates financial incentive to recycle.

---

Directions for further research

Richer socioeconomic modeling: The finding that higher-income households emit less and recycle more warrants deeper examination. Do high-income households offset this through high-carbon behaviors like frequent air travel? Incorporating flight data and other consumption variables would test whether the current picture holds.

Non-linear model exploration: Some variable pairs rejected H0 in correlation tests but produced poor linear fits. As additional modeling techniques become available, exploring non-linear relationships (e.g. polynomial, tree-based) would likely yield a more accurate picture of what drives household emissions.