Urbanization and Carbon Emission Levels

Updated: May 28

Because Google Adsense deemed this website to have too little content for ads, I'll be posting my top 25 undergraduate papers. This was the term paper for required political inquiry class for political science majors. Even though I was also an economics major, this was the only time in which I was able to do my own regression analysis. It really makes you feel like you're a real academic. Together with my assigned partner Christian Alfonso Alcantara, we tried to determine the correlation between urbanization and carbon levels. My theory was that countries with higher urbanization would have lower carbon emissions because cities are more efficient than rural areas.

Dalhousie University

POLI 3492

April 21st, 2020

I.Introduction and Research Hypothesis

In a time where desperate solutions to climate change are being searched for, the search for an environmentally friendly future has been a demanding pursuit. The idea being that cities have less emissions per capita because of lower commute times, the existence of public transit, greater housing density, etc. While it is common to think that highly urbanized cities are a major cause for high carbon emissions with their densely populated communities and technological reliance for their day-to-day lives there have been numerous studies that show evidence to the contrary. The goal of this paper is to analyze using statistical data along with multiple researches how highly urbanized cities produce less carbon emissions and hopefully provide insight as to how modernizing cities ought to move forward to a sustainable tomorrow.

II.Review of Resources

We turn to a BBC news article by Paul Swinney which points out using statistics from the European Commision and official government statistics to show how areas defined as cities produce less carbon emission on average per person in three different sectors of the community (Industry, Domestic, and Transport) as opposed to areas away from densely populated areas. According to the article, this is due to various reasons. First, we look at the Industry section of the presented chart. In the UK, chemical and steelwork factories have moved away from urban areas due to environmental and health concerns which are at a greater risk in cities. Furthermore, jobs in the cities are more likely to be office-based, as such, a larger number of people are inside one building producing little carbon emissions from their desks limited to only heating or air conditioning. Second is transport, the reasoning behind this is that people who live outside urban areas tend to use their automobiles to reach their destination whether it be for work, groceries, or leisure as opposed to urban cities where public transport accounts for the mass transportation methods of the residents.(1)

How the article provides statistics and explanation into the three sectors of an urban area is relevant to our analysis as it provides context into the key factors to look into when arguing how urban cities produce less carbon emissions with one of our Independent Variables being a composition of GDP, which is Industry. At its conclusion, it also briefly details how cities in the UK are turning its direction into clean energy, reducing carbon emissions and the health risks that come along with it.

Stating that urban cities produce less carbon emission is too simplistic. There are complexities when it comes to city planning and distribution of population as evidenced in a research analysis by Carl Gaigne. Focusing on the Industrial and Domestic aspects of an urban area, he goes in-depth using a series of algebraic and statistical formulas to get an estimate as to how increasing population density will affect prices, wages, and land rents. He recognizes the increased usage of automobile transportation among polycentric cities as opposed to monocentric cities but are outweighed by the consequences in social welfare. In conclusion, Gaigne argues the importance of city structure with emphasis on the effects of transportation such as the labour market and social welfare. While his work is largely theoretical with other domestic factors such as the cost of housing sectors were not included the objective was to introduce the balance between planning cities vertically and horizontally. Monocentric megacities prove to be detrimental in the goal of reducing carbon emission but a spread-out polycentric city with limited height on taller buildings can prove to be effective.(2)

Using the ARDA Cross-National and Socio-Economic dataset, the goal of this analysis deals with the effects among multiple countries. Gaigne’s research tells us that urbanization should not be the only factor we tunnel our sights into but that there are further details to consider that goes beyond what was provided to us in the dataset. Indeed, while his research does challenge the goal of this data analysis paper it does give us careful consideration when interpreting the results of the data.

City structure is one of the important determinants of whether an urban area produces more or less of carbon emissions and as previously seen in Gaigne’s research the outer formation of an urban area can make a sizable difference in CO2 levels. However, the structure within those urban areas is another half of the big picture and in Daniel Hoorweg’s analysis of greenhouse gas emission in multiple Canadian cities, this proves to be the case. Using statistics from official government sources that measure in Tonnes of CO2 equivalent (tCO2e), which is a value that gauges other greenhouse gas emissions equivalent to one unit of CO2, he compares data from multiple Canadian cities in how much waste material these cities produce segregated into household, citywide, and metropolitan emissions. The results are, on average, household emissions are producing less amounts of tCO2e by as much as 40%. He explains the reason for this difference is due to the amount of greenspace more prevalent in household areas than metropolitan ones. This is also dependent on lifestyle choices of the population living in these areas as the constant daily production is lower among household areas. CO2 and tCO2e emissions per capita are also highly dependent on the type of energy sources these countries run. For example, Ontario, British Columbia, and Quebec which powers most of their cities using hydro-electricity produces less GHG emissions per capita than that of Alberta which runs largely on geothermal power and oil. The Economic industry that is specific to each province also plays a role in emissions, as stated before Alberta is a major producer of oil but the environmental consequences of that production contributes to its per capita emissions.(3)

The significance of Hoorweg’s research in connection with our data analysis is plain to see. His statistical comparisons of greenhouse gas emissions among multiple cities shows that it is not simply a matter of GDP nor levels of urbanization that are a result of high or low CO2 emissions but also the geographical and industrial levels that are unique with each city. In his research, it shows that low to middle income cities produce less waste material than those of richer cities which may challenge our independent variable for urbanization but strengthen the variable for fossil fuel energy.

To further drive the point of urbanization and city structure we take a look at Yasudo Makido’s research on urban form and its relations to carbon emission production in 50 Japanese cities. Using landscape metrics and multiple linear regression followed by a coefficients’ correlation test he concluded that less fragmented and compact cities produce lesser amounts of CO2 emissions. However, he stated that this relationship between urban form and levels of CO2 emissions were only significant with residential areas as such it is only limited to one sector of an urban city yet the relevance still doesn’t change. When it comes to urban planning and policy making taking into account the resulting form of a city can lead to a drastic change with levels of CO2 production even if in residential areas, from our previous sources we have to take into account factors such as greenspace, transportation etc. To find the right balance for an environmentally-stable city structure is increasingly becoming high demand as we move into the future.(4)

III.Statistical Data and Analysis

This multivariate regression will be done in SPSS Statistics using the Cross-National Socio-Economic and Religion Dataset from 2011. The dependent variable in question is “carbon dioxide emissions per capita in 2006, in tones” and the primary independent variable of interest is “percentage of the total population living in urban areas, 2010.” The only potential issue for measurement and sampling issues could be the sample size, which will be revealed in the correlations table. It would be helpful if there existed a definition for “urban areas” anywhere online for this ARDA dataset, but after a search it couldn’t be found. All datasets were measured in a fairly close window of time. For this multivariate regression there will be three other independent variables being used as control variables. The first is “gross domestic product per capita in US dollars, 2008” the second is “composition of GDP: industrial sector” and the third is “percentage of total primary energy supply derived from fossil fuels, 2007.” The logic behind using these variables as control variables for CO2 emissions per capita is that we would expect richer countries to use more electricity and thus have higher emissions per capita. In terms of styles of economies (agricultural, manufacturing, services) using conventional wisdom we’d expect countries with more manufacturing to have higher CO2 emissions per capita. And finally we would expect a countries energy supply (fossil fuel versus renewable) to be a large, if not the largest explainer of the variation of CO2 emissions per capita. There are five tables of interest that SPSS generates for the regression. The first is the correlations table.

Looking at the correlations table, we can see the individual relationships between each variable. The largest bivariate relationships are those of CO2 emissions per capita and GDP per capita, as well the relationship of CO2 emissions per capita and percentage of primary energy supply derived from fossil fuels. The smallest relationship is that of CO2 emissions per capita and composition of GDP (industrial sector) which is a positive but weak correlation. The relationship between CO2 emissions per capita and percentage of the population living in urban areas is sandwiched in between with a positive but mediocre correlation. The correlations table also demonstrates the sample sizes for the variables. The same size is 123 countries for each variable. Given that the sample is above 100 countries that should avoid any sampling errors.

The model summary table reveals the R Squared value. This is the figure that informs of how much the independent variables explain the variation in the dependent variable. An R squared value of 0.646 tells us that our independent variables as a whole explain about 65% of the variation in CO2 emissions per capita.

As far as the ANOVA table, this allows us to accept or reject the null hypothesis. Our null hypothesis is that none of our independent variables has a statistically significant relationship with our dependent variable. To reject the null hypothesis one needs a significance value of 0.05 or greater because our confidence intervals are 95 percent. The significance value for the model as a whole is lower than 0.000 therefore we can reject the null hypothesis and conclude that there is a statistically significant relationship. However, this ANOVA table does not inform us of which individual independent variables have statistically significant relationships with the dependent variable holding all other independent variables constant. For that information, we must turn to the coefficients table. 

The coefficients table is by far the most important and revealing. First of all, the unstandardized coefficients beta values show a numerical impact on the dependent variable. Meaning that a one unit (in this case a one percent) change in the percentage of the total population living in urban areas responds to a -0.01 change in CO2 emissions per capita, holding constant all other independent variables. Does this mean that we can conclude that more urbanization causes lower CO2 emissions per capita? Not exactly. The reason is that, as demonstrated in the other independent variable values for the unstandardized coefficients, their absolute values are much larger than the value for urbanization. The value for GDP per capita only appears tiny because it should be converted into thousands for this particular comparison. To help illustrate this, we look towards the significance level portion of the table. In order for there to be a statistically significant relationship between the individual independent variable and the dependent variable, its sig. value must be lower than 0.05 because we have a 95 percent confidence interval. The sig. value for all independent variables besides urbanization is below 0.05. The sig. value for urbanization is 0.618 which is far above the value we need to declare a statistically significant relationship between urbanization and CO2 emissions per capita when all other independent variables are held constant. 

To further emphasize that point, some statisticians compare standardized coefficient values as a way of measuring the relative impacts of each independent variable. For our model, the lowest impact in absolute terms is the percentage of population living in urban areas at 0.036. The next is the composition of GDP (industrial sector) at 0.242, then the percentage of total energy supply derived from fossil fuels at 0.346. The independent variable that is the most impactful by far is GDP per capita with a standardized coefficient beta value of 0.626. Which is why the conclusion of the regression is that while urbanization has a negative impact on CO2 emissions per capita when the other independent variables are held constant, its impacts are not statistically significant. The other independent variables are far more important factors in deciding a country's CO2 emissions per capita, especially the percentage of energy supply derived from fossil fuels and the countries GDP per capita. It is also worth noting that there is no issue of multicollinearity in this model because all VIF values are well below (5).

This model is not without its outliers however. The Casewise diagnostics table demonstrates 4 outliers. Two of which are Norway and Kuwait. The model says that Norway has much lower CO2 emissions per capita than we would expect given its independent variable figures. At the same time, the model says that Kuwait has a much higher CO2 emissions per capita figure than we would expect given the values from its independent variables. One possible explanation for this could be the fact that Norway is a country that has a relatively high level of offshore drilling(5) and Kuwait is a country with a high level of oil production on its soil. 

IV.Policy Implications and Agenda for Further Research

The immediate policy implication from this regression is that governments should focus more on transitioning away from fossil fuels towards renewable energy more than they should focus on incentivizing and encouraging urban living with respect to fighting climate change. That is an easy conclusion to draw given how much larger the standardized coefficient beta value is of the fossil fuel variable compared to the urbanization variable. The discussion about fossil fuel usage is a perfect transition into an agenda for further research. It is unfortunate that the ARDA dataset does not have any variables with respect to transportation. A great control variable to add to this regression would be the percentage of the population that drives a car. That would be a great variable to add because not all cities are alike. Imagine two cities that are the exact same in every way. The only difference between them is one city has an immaculate public transportation system and the other has no public transportation. Clearly they are both almost identical cities but they are going to be quite different in terms of CO2 emissions per capita. And that is a difference in which this ARDA dataset cannot capture. So given that public transportation is so often seen as an essential tool to fight climate change, car usage among the population would have been a great control variable to add.


Dunn, Katherine. “Norway Is Set To Drill More Than Ever Before.” Fortune. Fortune, October 21, 2019. https://fortune.com/2019/10/18/norway-drilling-climate-oil-and-gas/.

Gaigne, Carl, Stephane Riou, Jacques-François Thisse, and Gate Working Paper Series. “Are Compact Cities Environmentally Friendly?” SSRN Electronic Journal, 2010.


Hoornweg, Daniel, Lorraine Sugar, and Claudia Lorena Trejos Gomez. “Cities and greenhouse gas emissions: moving forward” Environment and Urbanization, 2011


Mashido, Yasuko, Shobhakar Dhal, and Yoshiki Yamagita. “Relationship between urban and form and CO2 emissions: Evidence from fifty Japanese cities” Urban Cities, 2012, pp. 55-67.


Swinney, Paul. “Are Cities Bad for the Environment?” BBC News. BBC, December 16, 2019. https://www.bbc.com/news/science-environment-49639003.

1 - Swinney, Paul. “Are Cities Bad for the Environment?” BBC News.”

2 - Gaigne, Carl, Stephane Riou, Jacques-François Thisse, and Gate Working Paper Series. “Are Compact Cities Environmentally Friendly?”

3 - Gaigne, Carl, Stephane Riou, Jacques-François Thisse, and Gate Working Paper Series. “Are Compact Cities Environmentally Friendly?”

4 - Gaigne, Carl, Stephane Riou, Jacques-François Thisse, and Gate Working Paper Series. “Are Compact Cities Environmentally Friendly?”

5 - Dunn, Katherine. “Norway Is Set To Drill More Than Ever Before.” Fortune.