credit: Sven-Erik Arndt/Newscom
credit: Sven-Erik Arndt/Newscom
Climate change and its effects are accelerating at a rapid pace, almost to the point of irreversibility. While many studies have analyzed the sources for this ongoing climate change and work is being done into how to fix it, in this post, we want to explore the structural characteristics of countries associated with climate change, and in particular, \(CO_2\) emissions. A paper from the Fraser Institute, a non-partisan Canadian public policy think tank, suggests that there may be a relationship between economic freedom and CO2 emissions, in that higher economic freedom scores are associated with lower carbon dioxide emissions. In this exploratory post we attempt to determine the inner workings driving this effect that the Fraser Institute finds, and in particular what parts of economic freedom actually may hurt carbon dioxide emissions.
We used the Gapminder dataset’s information on each country’s \(CO_2\) emissions in conjunction with economic freedom data from the Fraser Institute to explore this relationship. The Gapminder carbon dioxide data was collected and maintained by the US Department of Energy and the Lawrence Berkeley National Laboratory, and records \(CO_2\) emissions in terms of tonnes per person (per country per year). The Economic Freedom dataset contains 42 economic variables ranked on a scale from 1 to 10 by the Fraser Institute which uses various third-party sources to do so. The 42 variables are also grouped into 5 summary categories, and each of these 5 categories have their own 1 to 10 rating.
The five summary categories are Size of Government, Legal and Property Rights, Sound Money, Freedom to Trade Internationally, and Regulation. Each of these scores is broken down even further.
Image courtesy of TIME magazine
Jacundá National Forest fire in Brazil’s Amazon
First, from the economic freedom dataset, we wanted to determine if we could categorize countries in a meaningful way. To get a sense of economic categories of countries generally (i.e. throughout all of the data), we used k-means clustering, and constructed a graph to show the amount of variation per each possible number of clusters.
Left: Elbow graph depicting variation by number of clusters, K
## Warning: did not converge in 10 iterations
We settled on 3 clusters as an optimal balance of minimizing variation and complexity, shown by the elbow point in the above graph. We then ran the clustering and visualized it using a) a matrix of plots for the group distributions of each variable, and b) a radar chart.
For each of 4 of the 5 variables, the blue group had the highest values, the green group had moderate values, and the red group had the lowest values, a simple conception of grouping. The one exception is size is government, for which the positions of the blue and green groups are swapped. The blue group seems to consist of highly developed and stable countries like the United States, while the red and green groups represent various stages of developing or less stable countries, like Iran and Ukraine for the red group (presumably the least developed or stable group) and Mexico and Indonesia for the green group (possibly the moderately developed/stable group). We see from the radar chart characteristics of our three clusters: the blue group clearly is the high scoring group, the green is the mid score group, and the red is the low scoring group. This is with the exception of size of government, for which all clusters have similar values. Given this clustering, we tried exploring the effects of each of the 5 variable scores on carbon dioxide emissions, controlling for cluster fixed effects. In some sense, we were trying to get a sense of what variables were important even controlling for whether a country had good scores already in all categories.
After categorizing our countries, we turn to CO2 emissions. Within country, we attempted to discover what factors were associated with higher levels of CO2 emissions, with an eye towards business regulation.
A linear regression to examine the effect of each of the 5 variables on CO2 emissions, controlling for cluster fixed effects resulted in the following:
## Estimate Std. Error t value
## `1 Size of Government` -0.3578108 0.3944547 -0.9071024
## `2 Legal System & Property Rights` 1.0918687 0.3984242 2.7404678
## `3 Sound Money` 0.5702078 0.3770767 1.5121801
## `4 Freedom to trade internationally` 0.9400612 0.3815151 2.4640212
## `5 Regulation` 1.6767521 0.4000648 4.1912012
## Pr(>|t|)
## `1 Size of Government` 3.644492e-01
## `2 Legal System & Property Rights` 6.183345e-03
## `3 Sound Money` 1.306276e-01
## `4 Freedom to trade internationally` 1.381262e-02
## `5 Regulation` 2.881386e-05Given that the Fraser Institute paper that motivated our study saw a 9% decrease in long term carbon dioxide emissions given a 1 point increase in economic freedom index, we wanted to know what sub-categories of the index actually increased carbon dioxide emissions. We noticed that Regulation scores in particular had a significant positive effect on tons of CO2 emitted per person, in addition to property rights. So we decided to explore this further. First, we selected variables from our economic freedom dataset relating to economic business regulation, things like property rights, tariffs, and capital controls. If we take a look at heatmap, we can initially see the correlations between tons of CO2 produced per person and our other variables:
In the heatmap, looking at the row for tonnes_person, we see that we have a few variables strongly positively correlated with CO2 emissions in the dark orange coloring: Property rights and business regulations being the lightest two bars. Looking at a scatterplot, we see somewhat of a linear relationship between property rights and CO2 emissions, and a similar looking scatterplot for business regulations and CO2 emissions.
From these scatterplots, which are colored by country, if see a few countries that seem to be skewing this positive relationship to be more positive, year after year. These have high property rights scores as well as business regulations scores in addition to their carbon dioxide per capita emissions.Qatar, Kuwait, Singapore, and UAE are examples of this. These generally are countries that heavily depend on oil still and have not turned to renewable energy for the most part. Given our set of variables and our initial hypothesis, we ran a linear regression of the following specification, to examine within country variation in CO2 emissions due to business related scores.
\[ CO_2 = \beta_0 + \beta_1*PR_t + \beta_2*BR_t + \beta_3*X + \theta + \epsilon\]
where CO2 is tons of CO2 emitted per person, \(PR\) is property rights score, \(BR\) is the overall score for business regulation, X is the rest of our business related covariates, and \(\theta\) is fixed effects for year t. The relevant results for our regression are shown below, for property rights in the first row and business regulation in the second.
## Estimate Std. Error t value Pr(>|t|)
## 1.383940e+00 1.564044e-01 8.848477e+00 4.785337e-18## Estimate Std. Error t value Pr(>|t|)
## 0.8570525398 0.2556849591 3.3519865340 0.0008365017We find that both are significant at the \(\alpha = 0.01\) level, rejecting the null hypothesis that \(\beta_1 = 0\) and \(\beta_2 = 0\). From our results, we see that property rights scores have a significant positive relationship with CO2 emissions, and so do business regulation scores.
So why are these particular indicators significant for CO2 emissions? Actually, after searching through some papers, better property rights seem to imply better CO2 mitigation policy. It is likely that the property rights score is likely positively correlated with wealth of a country, which should be positively correlated with CO2 emissions, as wealthier countries generally do have developed industry producing these emissions. This could confound the results, and thus that result should be taken with a grain of salt. Better business regulation scores is positively correlated with tons of CO2 per person, which makes sense considering CO2 emissions would be curbed based on worse business regulation.
credit: University of Aberdeen
Lastly, considering these factors, we also have not taken into account how countries produce, export, and import fuels that create these carbon dioxide emissions. We want to know which countries are driving carbon dioxide emissions in the world by producing and trading crude oil. Considering these things, we present one last visualization: using data from the Observatory of Economic Complexity, below you can find a network representation of Crude Oil trade in OECD countries + a few more big carbon emission countries. The edge widths represent the total volume traded (imported + exported) between the two countries (each country is a node). Clicking on nodes reveals their country name, and the graph can be dragged and zoomed. The visualization below is for the year 2016.
By far, the most volume of trade of crude oil in 2016 is between Canada and the US, and secondly the UAE and Japan.
The nature of our study depends heavily on the constructed indices for economic freedom. This was helpful in narrowing down our predictors of interest, but to fully understand how country structure affects climate change, we would like more data into business regulation and property rights to conduct another analysis of its effects. Furthermore, we studied only between country variation and not within country variation, which could also be interesting. For example, one interesting study could find the relationship between state business policy and environmental laws with carbon dioxide emissions. In that study we would want to focus on whether these policies actually are effective in reducing emissions. Additional context into cap and trade programs and other measures of environmental regulation could deepen our analysis further.
Tidyverse
Pairwise plots: GGally
Interactive scatterplots: scatterD3
Radar chart: radarchart
Interactive heatmap: heatmaply (or d3heatmap)
Force directed network: networkD3