36315 Final Project - Energy Consumption Across the World

Introduction

Energy is the crucial foundation of society since the beginning of times and today there is a vast amount of different types of energy that are both renewable and non-renewable. In recent years there has been a push for more renewable energy sources as well energy sources to reduce the amount of greenhouse emissions. In this project we hope to be able to analyze what the trends across the World are for both renewable and non-renewable energy sources to determine global trends in energy type usage.

Dataset:

The dataset contains information about a vast array of energy information across the World. Its rows are individual data observation and its columns contain different countries across the world, the year of the observation, the greenhouse_gas_emissions, along with inforamtion about the annual percentage change in consumption (_cons_change_pct), annual change in consumption, measured in terawatt-hours (_cons_change_twh), per capita primary energy consumption, measured in kilowatt-hours (_cons_per_capita), primary energy consumption measured in terawatt-hours (_cons_per_capita), primary energy consumption measured in terawatt-hours (_consumption), per capita electricity generation measured in kilowatt-hours (_elec_per_capita), electricity generation measured in terawatt-hours (_electricity), share of electricity generation (_share_elec), share of primary energy consumption that (_share_energy) for each of the following types of energy: Biofuels, Coal, Fossil Fuels (Coal, Oil, Gas), Gas, Hydropower, Low-Carbon (Renewables and Nuclear), Nuclear, Oil, Other Renewables (Including Biofuels, Excluding Biofuels), Renewables, Solar, Wind.

Research Questions

Research Question 1: What Variables are related to Greenhouse Gas Emissions?

The first question we want to answer is what variables are related greenhouse gas emissions, and which of these variables are most important? For this question, there is a need to narrow down which variables we suspect have a significant relationship with greenhouse gas emissions. As such, we have decided to conduct our investigation on the values of electricity production in regards to each renewable and non-renewable resource.

Graph 1.1: Histograms of Energy Type Electricity Production in the Year 2021

In order to get some intuition on the distribution of the production of electricity for each energy type, we created a facetted-histogram for both renewable and non-renewable resources for the year 2021.

Before moving on, however, it is important to note that every single histogram is extremely right-skewed. Some countries’ electricity production is league’s higher above the rest - the United States, for example - and if these countries were included in the histogram, we would only be able to see a histogram with around 3 bins that actually contain observations. As such, we have excluded countries who produce over 300 terrawat-hours of electricity using a specific energy type.

Looking at the histograms for above regarding electricity production for non-renewable resources, we can once again see that, even when eliminating outliers, the histogram is extremely right-skewed. We can see that, for all energy types, there are many countries that produce almost no electricity.

However, that is not to say that countries are not producing electricity from these energy types, or do not intend to in the future. When looking into the dataset, if a country has no association with a certain resource type at all, the data is marked with an “NA”. This means that even if a country currently does not produce electricity using a certain resource type, they have some other association with that resource type.

Moving on, it is important to see how many countries are actually producing more than single-digit values per resource type. We can see that the nuclear and biofuel resources are not being used to create more than 10 terawatt-hours of electricity in 2021. In comparison, there are countries using coal, fossil fuels, gas, and oil to produce more than 10 terawatt-hours of electricity in 2021, leading us to believe that these four resource types may have a strong relationship with greenhouse gas emissions.

Looking at the histograms above regarding electricty production from renewable resources, we can see that the histograms are very similar to the non-renewable histograms, as they are both heavily right-skewed.

As for the nature of the left-most bar in the histogram, it has the same interpretation as the non-renewable histogram. It is important to note that there are countries that are producing at least some electricity - and those who do not have some association with the energy type.

Looking at the histograms for the hydro resource type, we can see that its left-most bar is significantly smaller than the other resource types, and that its right tails are more visible as well. This means that there are more countries using the hydro resource types to create a larger amount of electricity in the year 2021. With this insight in mind, this leads us to suspect that the hydro is going to be closely related with greenhouse gas emissions.

Graph 1.2: PCA Bi-plot of All Energy Type Comsumptions Colored by Greenhouse Emissions

For the following plot preprocessing was done to remove all non-countries (ie. continents) that were present in the country column. Additionally the greenhouse_gas_emissions column was bucketed into four categories: low, medium-low, medium-high, high.

The PCA biplot indicates that there is a distinct separation between energy sources associated with high greenhouse gas emissions and those with low emissions. Fossil fuel-based electricity sources such as coal, oil, and gas—align strongly with the “High” emissions category, suggesting a correlation between the consumption of these energy types and higher greenhouse gas emission levels. In contrast, renewable and low-carbon energy sources—such as wind, solar, and hydroelectricity—tend to cluster with the “Low” emissions category, indicating their role in a cleaner energy mix. It can be observed that ellipse for high greenhouse gas emissions lies in the direction of principle component 1 thereby suggesting that high greenhouse emission energy sources account for a large proportion (63.6%) of the total variance in the data. The other three categories of greenhouse emissions mostly lie in the direction of PC2.

Graph 1.3: Loess Regression Analysis of PC2 vs PC1 Colored by Greenhouse Emissions

The Loess regression plot of PC2 against PC1 reveals a non-linear relationship between the first two principal components. The curve dips sharply in the middle, indicating an inverse relationship between the scores on PC1 and PC2 within this region. As the color intensity represents greenhouse gas emissions, we observe that the highest emissions are associated with large values of PC1 and near 0 values for PC2 once again suggesting that high greenhouse gas emissions lie almost entirely along PC1. This could suggest that countries with a moderate reliance on certain energy types, possibly those transitioning between high-carbon and low-carbon sources, have the highest emissions. The scatter of points with lower emissions is more spread out across PC1 and PC2, which may imply a more diverse energy mix or efficient energy use.

Research Question 2: How are population and GDP of countries related to greenhouse gas emissions and the rest of the dataset?

Next, we wanted to explore how the population and GDP (gross domestic product) of countries relates to the rest of the dataset. In particular, if we categorize the countries into relatively equal groups of low, medium, and high population or GDP, would we see any interesting trends?

Graph 2.1: Heat Map of MDS Coordinates

We once again filtered out the non-countries and cleaned up the NA rows before continuing with further analysis. Then we performed MDS on the scaled quantitative variables of interest and plotted the first two MDS coordinates against each other on a heat map.

As indicated by the red portion on the graph, there is a bin centered at around (0, 0) with over 1000 observations, indicating a concentration of similar points there. Other than that, any other given been seems to have very few observations relatively.

Graph 2.2: Contour Map of MDS Coordinates by Population Group

For this part, we created three relatively equal groups of low, medium, and high populations for countries by first examining the histogram of population values and then doing some educated guessing and checking until the three categories had similar counts, making sure the cutoffs made sense. Finally, we plotted the first two MDS coordinates against each other, by population group.

Based on this plot with the primary two coordinates from performing MDS, we can see that a vast majority of the countries are centered around (0, 0) and thus have similar characteristics regardless of population group. However, there is a clear set of points characterized by higher population that have high values of both MDS coordinates, indicating a specific set of countries that might have different trends.

To further examine these trends, we just plotted the first MDS coordinate against greenhouse gas emissions, a variable we have taken particular interest in, to try and get some more meaningful results.

From this, we can see that the first MDS coordinate has a generally positive association with greenhouse gas emissions; that is, points with more positive values of this coordinate are associated with more greenhouse gas emissions. Thus, we can draw the conclusion that a certain few high population countries are responsible for contributing to large amounts of greenhouse gas emissions, which makes sense.

Graph 2.3: Heat Map of MDS Coordinates by GDP Group

Next, we did the same thing for the other variable in the dataset that described countries: GDP. Similarly to before, we created groups of low, medium, and high GDP and used the same MDS coordinates, with the same type of contour map.

We get similar trends when compared to the population groups; there is a large concentration of points around (0, 0), and there is a select few observations characterized by high GDPs that have higher positive values of both MDS coordinates.

Plotting the coordinates against greenhouse gas emissions, we ultimately see similar trends as the previous part; the graphs were omitted to reduce repetitiveness. One meaningful conclusion we can draw is that countries with a higher gross domestic product emit more greenhouse gasses, which of course, makes sense.

Research Question 3: How does the United States compare to the rest of the World in regards to Energy Use?

Finally, we wanted to explore how the United States’ energy generation methods and greenhouse gas emissions compare to the rest of the world and explore some reasons why there may be differences.

Graph 3.1: Time Series of Greenhouse Gasses

For a initial comparison of greenhouse gas emissions between the United States and the World, we created a time series displaying both of the yearly greenhouse gas emissions in the 21th century.

From this graph, we can see two interesting results.

Firstly, we can see the opposite trends between the United States and the World. While the United States is decreasing their greenhouse gas emissions by about 25%, the total greenhouse gas emissions in the World have increased about 75% over the same time period. Looking at the line for the United States, we can see the consistent decrease in emissions after 2008. This is likely due to legislation in the US limiting the use of fossil fuels, lowering the legal quantities of carbon emissions, and incentives to adopt greener and renewable sources of energy production. On the other hand, the greenhouse gas emissions worldwide have rapidly increased in the recent decades. This is likely accounted for increased production in less developed country. One notable example would be China, which has massively increased its development and manufacturing, fueling mainly by large amount of fossil fuels.

Secondly, the percentage of greenhouse gas emissions accounted for by the United States has massively decreased over this time period. Of course, the previously mentioned factors also apply here, but another contributing factor would be the offloading of manufacturing to less developed countries. Instead of US companies meaningfully decreasing their emissions, they are instead increasing emissions of other countries.

We will further explore the move from fossil fuels to renewable sources in the United States as well as the general trends of energy sources in the next graph.

Graph 3.2: Bar Plot of Energy Sources

To continue exploring the reasons for the increased greenhouse gas emissions worldwide and the consistent decrease in the United States, we will compare energy production types in between them.

Generally, we can see the overall growth in renewable energy sources both worldwide and in the United States. Therefore, we need to look deeper within each bar plot why greenhouse gas emissions worldwide and in US are having such different trends.

Within the United States bar plot, we can see the increase in renewable energy sources, mostly in wind and solar. At the same time, fossil fuel generated electricity slightly decreases, which is almost completely attributed to the decrease in coal.

On the World bar plot, we can see a massive increase in both renewables and fossil fuels. There are similar increases in solar and wind, but an opposite trend in coal.

Given that energy usage is generally increasing and that renewables have even grown more in the World bar plot, we can assume that the United States is likely using the energy generation from other countries to lower their greenhouse gas emissions.

## # A tibble: 2 × 3
##    year country       net_elec_imports
##   <dbl> <chr>                    <dbl>
## 1  2000 United States             33.8
## 2  2021 United States             39.3

We can see here that the net electricity imports have gone up by about 5.5 TWh from 2000 to 2021, but this likely does not account for non-domestic energy usage. As a very consumerist society, many of the products used in the US are likely produced in countries will much higher greenhouse gas emissions.

To continuing examining the relationships of fossil fuels and renewable sources of electricity, we ran a linear regression model of greenhouse gas emissions with fossil fuels and renewables as predictors.

## 
## Call:
## lm(formula = greenhouse_gas_emissions ~ fossil_electricity + 
##     renewables_electricity, data = .)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -53.176 -16.428   3.562  18.960  37.625 
## 
## Coefficients:
##                         Estimate Std. Error t value Pr(>|t|)    
## (Intercept)            445.26657  211.56681   2.105   0.0489 *  
## fossil_electricity       0.65857    0.06935   9.496 1.20e-08 ***
## renewables_electricity  -0.65256    0.05430 -12.017 2.53e-10 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 27.96 on 19 degrees of freedom
## Multiple R-squared:  0.9815, Adjusted R-squared:  0.9796 
## F-statistic: 504.2 on 2 and 19 DF,  p-value: < 2.2e-16

As you can see, for the United States in the 21th century, both fossil fuels and renewables have played a large part in the trend of greenhouse gas emissions, and are therefore both significant predictors of it (with an alpha value of 0.05).

Next, we can run the same linear regression model on the World.

## 
## Call:
## lm(formula = greenhouse_gas_emissions ~ fossil_electricity + 
##     renewables_electricity, data = .)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -45.837 -17.683  -3.432   7.286  72.009 
## 
## Coefficients:
##                          Estimate Std. Error t value Pr(>|t|)    
## (Intercept)            169.928407  59.779372   2.843   0.0104 *  
## fossil_electricity       0.712101   0.007583  93.911   <2e-16 ***
## renewables_electricity  -0.023755   0.011404  -2.083   0.0510 .  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 32.19 on 19 degrees of freedom
## Multiple R-squared:  0.9997, Adjusted R-squared:  0.9996 
## F-statistic: 2.914e+04 on 2 and 19 DF,  p-value: < 2.2e-16

For the World, fossil fuels becomes an even more significant predictor of greenhouse gas emissions, while renewables are no longer as significant, becoming larger than our alpha value of 0.05.

These linear models in combination with the previous graphs gives us an obvious picture that while renewables and fossil fuels have contributed largely to the decrease in greenhouse gas emissions in the US. On the other hand, we can also see that the large increase in renewable energies in the World have not significantly affected the greenhouse gas emissions. Therefore, we can confidently say that the increase in energy demand is much higher than the increase in renewable energies.

Overall, we can see that compared to the rest of the World, the United States has embraced renewable sources of energy and significantly reduced their greenhouse gas emissions. Although these strides towards a more sustainable future are good, we should be aware of the caveats that the data shows us. Many US companies have continuously used more overseas manufacturing and net electricity imports have significantly increased in the last 2 decades.

Conclusion

Overall, we found some meaningful trends that aligned with our interests of exploring this dataset. Through initial EDA, reducing dimensions with PCA and MDS, making general comparison graphs, and more, we were able to provide insights into our research questions of interest. However, there were drawbacks with some of the choices we made; while we made important general observations, due to the sheer size and nature of our data, we could not make absolutely precise graphs that narrowed down too many specific energy sources, countries, or even statistical numbers. For example, the MDS plots when exploring population and GDP couldn’t display proper contour lines or even many variables from the dataset. We do maintain that exploring some of these further would be within the scope of a larger project.

However, we ultimately made well-supported analyses that not only built on each other, but demonstrated that we can draw relevant conclusions from complex data. The graphs we produced directly answered our questions about greenhouse gas emissions, categorization of countries, and the role of the U.S. specifically in comparison to the world. We aimed to provide complements and supplements to individual graphs such that our conclusions were as supported and accurate as possible. As a result of this, we were able to infer and learn a lot about the energy of the world - the right skew of all energy sources, the most efficient countries emitting the least greenhouse gasses, the relationship of population and GDP in the context of energy and emissions, and how our country plays a role in transitioning to cleaner energy use.

Future Directions

We have identified several points of future work that are listed in this section. First, a more granular analysis of energy consumption patterns at a regional level within countries could reveal localized trends and barriers to adopting renewable energy sources. This would require more detailed data on regional infrastructure and policies, which were beyond the scope of the dataset currently. Second, we intend to apply more nuanced statistical techniques, such as machine learning models, to predict future energy consumption trends based on historical data and policy changes. This approach was not utilized in the current project due to the complexity of these models and time constraints. Lastly, examining the impact of international trade on national energy consumption and emissions could provide insights into the global dynamics of energy use. These avenues would enhance our understanding of global energy consumption patterns and their implications for sustainability.