Climate Change Adaptation as Disaster Risk Reduction: A global study of the relationship between exposure to natural hazards and climate change adaptation

Climate Change Adaptation as Disaster Risk Reduction

A global study of the relationship between exposure to natural hazards and climate change adaptation

Moa Christoffersson

Bachelor's Thesis in Political Science, 15 credits, December 2020 Department of Government, Uppsala University

Supervisor: Jacob Hileman Word Count: 13 216

Pages: 39

Abstract

In this thesis, I conduct a global event-data study investigating the relationship between exposure to natural hazards and climate change adaptation. Exposure to natural hazards has previously been linked to actions aimed at reducing risks related to future natural hazards. With climate change, and predicted increase in hazard frequency and intensity, a feasible approach to risk mitigation is climate change adaptation, which can thus be considered a disaster risk reduction strategy. I investigate the effects of disaster frequency and severity on the amount of climate change adaptation actions taken on a subnational level of government, using disaster data from the International Disaster Database (EM-DAT) and data on adaptation actions from CDP. Disaster severity is operationalised in three separate ways to distinguish between different kinds of disaster impacts: in terms of (1) economic damage, (2) how many are affected, and (3) fatalities.

I hypothesise that all independent variables are positively related to climate change

adaptation, and test the hypotheses using OLS regression. The result depicts a positive

correlation between the number of disasters experienced and adaptation actions. I do not

see a positive relationship between climate change adaptation actions and the two

impact variables total affected and total fatalities. The relationship between economic

damages and adaptation actions indicates that economic damages could have different

impacts depending on the level of economic development in a country. This study

contributes to the integration of the two research fields climate change adaptation and

disaster risk reduction by studying climate change adaptation as a form of disaster risk

reduction, and deepening the knowledge of what can drive adaptation. Finally, this

study contributes by showing that the level of economic development could be an

important aspect of the exposure-adaptation relationship.

Table of contents

1. Introduction and purpose ... 1

2. Previous research and theoretical framing ... 5

2.1 Disaster & climate change adaptation ...5

2.2 Previous research on CCA and DRR ...6

2.3 Theoretical framing ...8

2.4 Hypotheses ...9

3. Research design and method ... 10

3.1 Research design ...10

3.2 Separation into income groups ...11

3.3 Dependent variable: climate change adaptation actions ...11

3.4 Independent variables ...14

3.4.1 Disaster frequency ...15

3.4.2 Disaster severity ...16

3.5 Control variables ...16

3.5.1 GDP/Capita ...16

3.5.2 Population size ...18

3.6 OLS regression ...19

3.7 Critical reflection ...20

4. Results ... 22

Table 1. Low income countries ...22

Table 2. Lower middle income countries ...23

Table 3. Upper middle income countries ...24

Table 4. High income countries ...25

Table 5. All countries ...26

5. Discussion ... 28

5.1 Number of disasters (H1) ...28

5.2 Economic damages (H2) ...29

5.3 Total affected (H3) ...31

5.4 Total fatalities (H4) ...32

5.5 Implications and critical reflection ...33

6. Conclusions ... 36

7. References ... 37

7.1 Data sources ...37

7.2 Literature ...37

8. Appendices ... 40

8.1 Appendix 1: Causal diagram ...40

8.2 Appendix 2: Dataset ...41

1. Introduction and purpose

Disasters are not in any way a new phenomenon and their effects can constitute substantial impacts on human life as well as ecological systems and property. The purpose of this thesis is to investigate the relationship between exposure to natural hazards and climate change adaptation.

Exposure is increasing, both due to expected rise in hazard frequency and increasing vulnerability (Aitsi-Selmi & Murray 2016, 3). Natural hazards such as floods, extreme temperatures and storms can have immense consequences. With climate change, an increase in both disaster frequency and severity is expected to follow. The risk landscape is changing and so is our understanding of risk as a concept. In the report ”The human cost of disasters: An overview of the last 20 years (2000-2019)” from The Centre for Research on the Epidemiology of Disasters (CRED), the following is written:

Disasters have never waited their turn, and increasingly risk is interconnected. Risk drivers and consequences are multiplying and cascading, colliding in unanticipated ways. We must have a commensurate systemic response with national and local strategies for disaster risk reduction fit for purpose. Political commitment, strategies and scenario planning have never been more important for disaster risk management (CRED 2019, 3).

Disaster risk reduction (DRR) has become globally recognised as a research field and is now an

important part of the work to try to achieve sustainable development (UNDRR 2020a). In 1999,

the UN Office for Disaster Risk Reduction (UNDRR) was created (CRED 2019, 5). The UNDRR

coordinate and support efforts for mitigation of disaster impacts as well as work to prevent future

risks from emerging (UNDRR 2020b). As more people are becoming affected by natural hazards,

the focus on mitigating risk and building resilience is receiving more recognition (UNDRR

2020a). The UNDRR is also responsible for aiding in the implementation of the Sendai

Framework for Disaster Risk Reduction 2015-2030 (UNDRR 2020b). The Sendai Framework

lists areas of priority, and states that while the actor foremost responsible for risk mitigation is the

state other stakeholders should also be taking part, e.g. subnational governments (CRED 2019,

11). The Sendai Framework consists of seven targets, four of which with a "focus on substantial

reductions in (a) disaster mortality, (b) number of affected people, (c) direct economic losses and

(d) reducing damage to critical infrastructure and disruption of basic services", the latter three in

regards of seeking "a substantial increase in (e) national and local disaster risk reduction

strategies by 2020, (f) enhanced cooperation to developing countries, and (g) a substantial

increase in multi-hazard early warning systems, disaster risk information and assessments" (CRED 2019, 11). The Sendai framework can be said to constitute a link between the sustainable development goals of Agenda 2030 and the Paris Agreement (UNDRR 2019, iii).

The UNDRR also produces recurring global assessment reports on disaster risk reduction, which in 2019 is largely focused on climate change. In the report, it is stated that our time stands out regarding the amount of both known and unknown risks that we are facing. With climate change, new risks are coming into existence and it is happening faster than anticipated (UNDRR 2019, iii). We have reached a point where adaptation is necessary regardless of future greenhouse gas emissions and many believe that climate change adaptation needs to occur at a higher rate than today. For example, Alistair Hunt and Paul Watkiss write in the article ”Climate change impacts and adaptation in cities: a review of the literature” that it is becoming widely recognised that mitigation of greenhouse gas emissions is not enough and that we must ”address and adapt to the unavoidable consequences of climate change” (2011, 39). For these reasons, understanding what can motivate climate change adaptation is of great importance.

Of course, climate change adaptation actions can be taken for numerous reasons and this subject can be approached from several angles. In this thesis, I intend to study climate change adaptation as a possible response to being exposed to natural hazards and as a means for risk reduction. One reason for doing this is the relationship between climate change and disastrous natural hazards.

Although there is still some uncertainty regarding this connection, ample research supports the existence of such a relationship. With a changing climate, natural hazards have been predicted to become more frequent and more severe (see e.g. Birkmann et al. 2010, 653; Islam et al. 2019, 255; van Valkengoed & Steg 2019, 158). Furthermore, exposure to natural hazards is linked to the implementation of risk-mitigating policies. As noted by T. A. Birkland in Lessons of Disaster:

Policy Change after Catastrophic Events:

A disaster can often do in an instant what years of interest group activity, policy entrepreneurship, advocacy, lobbying, and research may not be able to do: elevate an issue on the agenda to a place where it is taken seriously in one or more policy domains (Birkland 2006, 5).

Disasters may thus provide an opportunity for policy change, possibly reducing risks related to

future disasters. However, as risk is increasingly understood as complex and interrelated, taking

actions to focus on reducing risk directly linked to a specific natural hazard is becoming recognised as inefficient (see UNDRR 2019, iv). Attention has been brought to the fact that current risk management institutions and approaches, while appropriate for handling individual hazards, are not fit for handling systemic risk on the scale indicated by increasing climate-related disaster events (CRED 2019, 7). Instead, disaster risk reduction increasingly focuses on building resilience and improving adaptive capacity (Ibid 2019, 11). In Assessment of Vulnerability to Natural Hazards: A European Perspective, Birkmann et al. write that building resilience continues to be a challenge for both developed and developing countries (2014, 2). Adger et al.

(2013) write that climate change induced extreme weather events could potentially cause crises in planning (2013, 330). In order to be viable under the more uncertain conditions that will follow climate change, responses should involve increasing resilience and enable flexibility (Hunt &

Watkiss 2011, 13). With a changing climate, and the predicted increase in hazard frequency and intensity, a feasible approach to risk mitigation is thus adapting to climate change. In this sense, climate change adaptation can be considered a disaster risk reduction strategy. Logically, exposure to natural hazards that are predicted to increase in frequency and intensity due to global warming, ought to encourage actions that reduce the effects of climate change. This is why, in this thesis, I will conduct a global study investigating the relationship between exposure to natural hazards and climate change adaptation. Climate change adaptation is thus seen as a form of disaster risk reduction.

The purpose of this study is to examine the relationship between exposure to natural hazards and

climate change adaptation. More precisely, this thesis will investigate the effects of disaster

frequency and severity on the amount of climate change adaptation actions taken. Disaster

severity will be operationalised in three different ways to distinguish between different kinds of

disaster impacts: in terms of (1) economic damage, (2) how many are affected, and lastly, (3)

fatalities. Studying this particular set of impact variables is motivated by the central role the

mitigation of these impacts play in the Sendai Framework (see Targets a - c). Using OLS

regression, I will perform both bivariate and multivariate regressions in an attempt to determine

the significance of the different severity measures for triggering climate change adaptation

actions on a global scale.

The following research question will guide the study: In what way does exposure to natural hazards influence the amount of climate change adaptation actions being taken?

To answer the research question, I will be conducting a study on the global scale studying natural hazards 2013 - 2017 and adaptation actions taken by cities and municipalities in 2018 - 2019.

Araos et al. write that ”[c]ities globally face significant risks from climate change, and are taking

an increasingly active role in formulating and implementing climate change adaptation

policy” (2016: 375). They thus constitute important arenas for climate change adaptation. By

examining the relationship between exposure to natural hazards and adaptation actions, this study

can hopefully contribute to the existing literature of disaster risk reduction and climate change

adaptation as well as further the integration of the two research fields. The aim of this study is

also to contribute to the literature on policy change and to the knowledge of what motivates

climate change adaptation.

2. Previous research and theoretical framing

In this section, I will first define two central concepts: (1) disaster and (2) climate change adaptation. Next, I will review previous research about disaster risk reduction and climate change adaptation in order to situate the study within the broader scientific discourse. The theoretical framing will then be described. Lastly, I will summarise the theoretical framing with four hypotheses.

2.1 Disaster & climate change adaptation

In this section are definitions of two concepts that are central to this thesis, disaster and climate change adaptation. In Lessons of Disaster: Policy Change after Catastrophic Events, T. A.

Birkland (2006) writes that a disaster should be understood as a social construct, meaning that a natural hazard only becomes a disaster when humans are affected. Any hazard is thus a potential disaster but not a disaster in itself (Birkland 2006, 104). All natural hazards mentioned in this thesis qualify as disasters. Therefore, I will henceforth use the terms natural hazard and disaster synonymously.

As previously stated, this thesis will be examining climate change adaptation as a form of disaster risk reduction. Climate change adaptation can be understood as:

[A]n adjustment in ecological, social or economic systems in response to observed or expected changes in climatic stimuli and their effects and impacts in order to alleviate adverse impacts of change or take advantage of new opportunities (Adger et al. 2005, 78).

Furthermore, there are two dimensions of adaptation: (1) building adaptive capacity and (2)

turning that capacity into actions, both of which can be used both in preparation for future events

related to climate change and as a response to events having already occurred (Adger et al. 2005,

78). In this study, climate change adaptation actions will be understood in a wide sense,

incorporating both dimensions.

2.2 Previous research on CCA and DRR

In the article ”Integrating disaster risk reduction and climate change adaptation: a systematic literature review” Shafiqul Islam, Cordia Chub, James C. R. Smart and Leong Liew highlight the necessity of integrating the two research fields climate change adaptation (CCA) and disaster risk reduction (DRR) (Islam et al. 2019, 255). Whereas Lisa Schipper writes in the article ”Meeting at the crossroads? Exploring the linkages between climate change adaptation and disaster risk reduction” that CCA and DDR share the same long-term purpose, which is ”to reduce vulnerability and create sustainable and flexible long- term strategies to reduce the risk of adverse impacts” (2019, 16). Integration of research fields who have dealt with the socio-economic effects of natural hazards, including the sharing of experiences and understandings, is necessary for reducing vulnerability to future environmental risks (Thomalla et al. 2006, 39). Moreover, Anne M. van Valkengoed and Linda Steg write in the article ”Meta-Analyses of Factors Motivating Climate Change Adaptation Behaviour” that adaptation is vital to mitigate risks related to the impacts of a changing climate (Valkengoed et al. 2019, 158). They underline that there is a lack of comprehensive, large scale analysis of factors that are relevant for motivating adaptive actions (Ibid 2019, 158). Additionally, other studies have highlighted that there is an absence of global assessments of climate change adaptation that is carried out across cities (see e.g. Araos et al. 2016, 375). Moreover, Birkmann et al. propose that future research focus on longitudinal studies for the purpose of deepening the knowledge about changes that have occurred after disasters (2010, 652). The authors further distinguish between disaster impact and change as an effect of disaster as two dimensions of DRR, claiming that the latter has often been neglected (Ibid 2010, 653). This thesis has been outlined with these findings in mind.

Birkland writes that the larger a disaster is concerning the number of fatalities, the affected area,

and the property that is destroyed, the greater the prospective impact on policy (Birkland 2006,

4f). Based on this, it is clear that disaster impacts in general have the potential to change agendas

and by extension possibly result in policy change. However, it is not entirely clear whether a

certain kind of impact matters more than the others when it comes to triggering adaptation

actions. Given what has been mentioned above, there are good reasons to presume that disasters

can provide opportunities for policy change. However, the literature is ambiguous. For example,

Birkland finds that the immediate period after a disaster, which ought to be a good time to

advocate for change and thus reduce risks of future disasters, instead can be a time where risk

mitigation is given very little attention due to the focus on disaster relief and rebuilding (2006, 105). Conversely, there are other reasons to presume that change may follow a disaster. One being that disasters can lead to an inflow of resources that may enable development that could otherwise not have occurred due to lack of capacity (Birkmann et al. 2010, 652). This will not be explicitly investigated in this essay, but it could be a possible mechanism explaining why exposure to natural hazards may lead to adaptation actions and is thus of interest.

In the last three decades, economic losses from disasters have risen in absolute terms but have decreased as a proportion of global GDP (Pielke 2019, 2). Georgeson et al. have studied climate change adaptation in ten global megacities, with adaptation proxied by the cities’ economic responses to climate change. In none of the investigated cities do adaptation spendings exceed 0.33% of the city’s GDP (2016, 584). Besides showing that the adaptation economy constitutes a minor proportion of the overall economy, Georgeson et al. also find that there are large differences in adaptation spending between the studied cities in developing, emerging and developed countries (Ibid 2016, 584). Their findings show that cities in developed countries spend twice as much as those in developing countries measured as a proportion of GDP, and furthermore, that economic damages from disasters in absolute terms are larger in developed countries. At the same time, damage as a proportion of GDP is larger in developing countries, while developing countries also see higher fatality rates. Their findings imply that adaptation efforts might be conducted to protect capital rather than people. Lastly, the authors also note that poverty, informal settlements, insufficient infrastructure together with insufficient spending on climate change adaptation make developing countries even more vulnerable in the face of climate change (Ibid 2016, 584).

The above-mentioned findings of Georgeson et al. (2016) are of interest for several reasons. First,

because it shows that adaptation spending is still quite small in scale and therefore further

investigations regarding what triggers adaptation may be useful. In that regard, studying policy

change in the area of climate change adaptation could be useful for researchers to better

understand policy changes in general and what may motivate climate change adaptation in

particular. Second, the findings are of interest because they suggest that protecting economic

property might be of greater significance than protecting human safety. With this in mind, it is

fruitful to investigate if this also implies that economic impacts of natural hazards motivate

climate change adaptation policy actions in a significant way, as well as whether it does so to a

greater extent than other kinds of disaster impacts that are more direct human costs, i.e. how many are affected and the number of fatalities.

2.3 Theoretical framing

There is an extensive body of research on policy change, including policy change in the environmental domain (e.g. Birkland 2006; Birkmann et al. 2010). Adger, Arnell & Tompkins argue that climate change adaptation is mainly reactive, meaning that it is triggered by events in the past or by presently ongoing events (2005, 77). One commonly studied dimension in the literature is the theory of focusing events. Focusing event theory became widespread after being used by John Kingdon (1984) in his book Agendas, Alternatives, and Public Policies and has since been an important concept for research on public policy (Birkland 2006, 1). Meanwhile, Birkland defines a focusing event as an event which is:

[S]udden, relatively rare, can be reasonably defined as harmful or revealing the possibility of greater potential future harms, inflicts harms or suggests potential harms that are or could be concentrated on a definable geographical or community of interest, and that is known to policymakers and the public virtually simultaneously (Birkland 2006, 2).

Several case studies have shown that natural hazards can act as focusing events triggering policy

change. For example, Birkmann et al. write about so-called ”mega-disasters” such as the tsunami

that hit the Indian Ocean in 2004. They conclude that disasters may provide a ”window of

opportunity” for societal change, including in the environmental domain (Birkmann et al. 2010,

650). However, focusing event theory may not in itself be sufficient in this study. Numerous

disasters in the International Disaster Database (EM-DAT) cannot be considered sudden or

relatively rare. It is thus questionable if all disasters that are of interest would qualify as focusing

events. Several of the disasters studied in this thesis are smaller in scale, although still large

enough to be considered disasters, and sometimes reoccurring. Some are so-called slow-onset

disasters, e.g. droughts, that in that sense do not constitute spectacular events. However, they

might nonetheless be important for policy changes since they directly, and often to a large extent,

have an effect on human life and wellbeing as well as on economic values. Therefore, the

theoretical framework is not focusing event theory itself since this study looks at aggregated data

rather than specific events. Instead, the theoretical framework draws on the logic behind focusing

event theory and the notion that the more severe disasters are, the more likely it is that policy

change will follow. This is the main motivation for studying the three impact severity variables in this thesis (see hypotheses 2 - 4).

An alternative theory, which the theoretical framework of this study also draws upon, is Birkland’s theory of cumulative learning. He discusses what he calls event-related learning, which in some ways is related to focusing event theory. According to the idea of event-related learning, change may occur due to pressure to “do something” after a disaster. It is not possible to say that there is a direct relationship between policy change and exposure to an individual natural hazard. Not all events will lead to change, but lessons learned ”may contribute to a base of experience that may promote learning from subsequent events as knowledge accumulates” (2006, 21). This brings us to the theory of cumulative learning:

Instead, the accumulation of experience influences policy change; learning is cumulative. Individual natural disasters do not appear to have much power to change policy because few groups mobilize in response to an event and explicitly lobby for policy change at the federal level (Birkland 2006, 153).

To put it differently, it might also be worth investigating if countries who experience repeated natural hazards are more likely to have taken more adaptive actions. Thus, more disasters experienced ought to be positively related to more adaptation actions. This part of the theoretical framework is the main motivation behind studying the frequency of disasters as a separate variable (see hypothesis 1).

2.4 Hypotheses

To summarise the theoretical framework, four hypotheses have been formulated:

H1: The number of disasters experienced significantly and positively affect the number of adaptive actions.

H2: Economic damages from disasters significantly and positively affect the number of adaptive actions.

H3: The total number of people affected by disasters significantly and positively affect the number of adaptive actions.

H4: The total number of fatalities from disasters significantly and positively affect the

number of adaptive actions.

3. Research design and method

In this section, I will first outline the research design. Next, I will describe the data used in the analysis, including what variables are used and why. I will then describe the research method (OLS regression analysis), followed by a critical reflection on the research design and data.

3.1 Research design

Within the social sciences, finding evidence for causality is a substantial challenge (Teorell &

Svensson 2007, 159). To have a satisfactory causal explanation, as well as when discussing effects, we must be able to account for contrafactual difference, time order (i.e. x must precede y), as well as isolation. There must also be a mechanism which explains why the independent variable has an effect on the dependent variable (Teorell & Svensson 2007, 71). The main purpose of conducting a regression analysis is to be able to give evidence for the criteria of contrafactual difference (Ibid 2007, 166). By controlling for underlying variables we can also start to isolate the relationship that we are actually interested in and thus come closer to a causal explanation (Ibid 2007, 219). Complete isolation, however, is not possible since we cannot ever state that there are no exogenous explanations which could have been included in the models.

What we can do is to control for factors which we have reasons to presume are of importance.

However, as noted by Djurfeldt at al. (2018), causal theory within the social sciences is not deterministic and neither is the multivariate regression model. In the case of multivariate regression analysis, it means that the residuals need not be 0, i.e. we do not expect our models to be able to account for all the variance in the dependent variable. In principle, there could be an infinite amount of causes for a social event, all of which cannot be included in our model. The other two criteria, time order and mechanism, cannot be met using regression analysis alone.

Mechanisms are oftentimes not observable. Instead, they are postulates of the theory that is the

basis for our models (Djurfeldt at al. 2018, 356f). The criteria of establishing time order is met in

this thesis by the design of the study; I am studying natural hazards that have occurred in the

years 2013 - 2017 and adaptive actions that have been taken in the years 2018 - 2019. Thus, the

dependent variable cannot have occurred prior to the independent variables investigated. With

this in mind, the study is designed with the intention of enabling claims about causality. The aim

is to be able to provide insight into the relationship between disaster frequency and severity and

climate change adaptation policy in general, not only the studied time period.

In order to test the hypotheses, I will perform several bivariate and multivariate regressions. I will first separate the data into four groups based on the countries’ economic statuses and run separate regressions on all groups. Next, I will perform regressions with data from all countries of the world together. The reason for this design is a suspected reporting bias in the data. In short, the suspected bias likely leads the global regression to favour high income countries. This design is also intended to make it possible to discern whether or not the model fits regardless of the level of economic development in a country, or if there are major differences due to economic status. I will be controlling for population size in all regression tables since it likely is an underlying variable influencing the independent variables. In the global regression, I will also be controlling for GDP/Capita. Next, I will describe my variables and the data used for the analysis. The dataset used can be found in Appendix 2.

3.2 Separation into income groups

To deal with suspected bias and to enable analysis regarding whether the model fits regardless of level of economic development, I use a classification obtained from the World Bank Group (2020c), and separate the material into four groups: low income; lower middle income; upper middle income; and high income countries. Thus, countries with different degrees of economic development will be studied separately. The data used for the categorisation is for the calendar year of 2018. Four units of analysis (Réunion, Saint Barthélemy, Martinique, Anguilla) had no GDP figures and were therefore not included in the categorisation from the World Bank Group.

To be able to include those units of analysis in the regressions, these territories have been included in the nations to which they are linked. Réunion, Saint Barthélemy, and Martinique have been added to France, and Anguilla has been added to the United Kingdom.

3.3 Dependent variable: climate change adaptation actions

The dependent variable in this thesis is climate change adaptation. There is, to my knowledge, no dataset in existence that could accurately include all adaptive actions taken by different actors.

Thus, the dependent variable must be operationalised. For the purpose of this study, this will be

done using data on adaptation actions taken on a subnational level of government, i.e. by cities

and municipalities. As previously mentioned, cities and municipalities are important actors for

deciding and implementing climate change adaptation actions.

For the dependent variable, I will rely on data from CDP. CDP describe themselves as ”a not-for- profit charity that runs the global disclosure system for investors, companies, cities, states and regions to manage their environmental impacts” (CDP 2020). The dataset I am using was collected by CDP in collaboration with Local Governments for Sustainability (ICLEI) and contains all subnational climate adaptation actions that were publicly disclosed in 2018 and 2019 (CDP Worldwide 2020). The data for these years is the first available complete dataset of its kind provided by CDP. Since the data used for natural hazards is national, the data on adaptation actions have been aggregated to national level data for the purpose of this study.

The data used in this thesis relies on self-reporting, which likely introduces a systemic bias. Self- reporting logically ought to occur to a greater extent where there is larger capacity, here understood in the sense that there are more resources available. Therefore, the suspected bias could influence the results of the global regression in a way that gives more weight to high income countries. In the worst case, this would mean that a global regression does not show the relationship between the variables as it is in the world, but rather as it is in high income countries.

This is problematic as we are interested in the relationship between disaster frequency and severity and climate change adaptation in general. However, it is also possible that those who might lack the capacity to report actions also lack the capacity to perform actions. It is not possible to know whether the lack of reported actions actually means that no actions were taken or that an unknown number of actions were taken but not reported. Since there is no way to estimate the amount of unreported actions taking place, no weights have been attached to the data used. If we were to alter the data due to this suspected bias, we might distort the results in ways we do not understand and cannot make reasonable assumptions about. Of course, the best option would be to have undeniably unbiased data. Since that is not possible, the next best option is understanding the bias and designing a study to work despite this flaw. By being aware of the possible bias, it can be taken into account. In this thesis, it is dealt with by performing separate regression on countries split into four income groups, as well as on all countries together. Seeing as the suspected bias likely is due to varying capacity, proxied by level of economic development, this design is intended to make it possible to work with biased data by enabling a comparison between the regressions on the separate income groups and the possibly biased global regression.

Moreover, the data used is in this sense imperfect, but still constitutes the best available guess of

the actual amount of climate change adaptation actions being taken worldwide. Thus the

operationalisation of the dependent variable is the reported amount of actions taken. The self- reported adaptation actions in the CDP dataset will in this thesis serve as a proxy for adaptation actions taken.

Another possible operationalisation of climate change adaptation is the one used by Georgeson et al. They study megacities and use spending in sectors related to adaptation as a proxy for climate change adaptation (2016: 584). This is a viable approach for intensive in- depth studies of a few cities, but since this thesis is a global study and no aggregated data on adaptation spending exists, at least that I am aware of, that was not an option. To be able to study climate change adaptation on a global scale, adaptation will instead be operationalised by, as previously mentioned, self- reported adaptation actions. Although adaptation reporting as a proxy for climate change adaptation has its problems, it is sometimes necessary to have a less than perfect proxy to be able to study the phenomenon at all. Adaptation reporting has also been used as a proxy for adaptation in previous research, e.g. by Araos et al. with the motivation that ”[d]ata and knowledge on adaptation are difficult to find, and we have to date relied on the reporting of adaptation as the only option currently available for systematic analysis” (2016, 376).

This study is limited to the adaptation actions taken by subnational governance units. Other actors

may also engage in adaptation actions, but there is at present no realistic way to account for such

actions. Adaptation actions by other actors are not the focus of this study and no data on such

actors is included in the analysis. Moreover, seeing as cities and municipalities constitute some of

the most important arenas for climate change adaptation, this is not a major limitation. The

dataset consists of a wide array of different kinds of adaptation actions, e.g. creating action plans

for various risk scenarios, building community resilience, creating green roofs etc. The possibility

to focus on certain types of adaptation action was considered but dismissed, partly because not all

actions in the dataset are described, but mostly because by removing certain actions, I might

introduce bias based on my knowledge and understanding. There is no reason to remove any of

the actions in the dataset considering the way adaptation has been defined in this thesis. By

having an inclusive definition of adaptation, I am able to fully utilise the information in the

dataset. Thus, any action will qualify as an adaptation action in this thesis as long as it is included

in the dataset provided by CDP.

3.4 Independent variables

The independent variable in this thesis are disaster frequency and severity. All disaster data in this thesis (see CRED/UCLouvain 2020) come from the International Disaster Database (EM-DAT) which is operated by CRED. Since the focus of this thesis is climate change adaptation as a form of disaster risk reduction, and given the relationship between climate change and natural hazard frequency and severity, only disasters with a strong connection to climate change will be included in this study.

CRED has sorted all natural hazards in their database into different categories. First, CRED classifies hazards as either natural or technological (CRED 2020b). In this thesis, only the ones classified as natural will be investigated. This category is further divided into the following disaster subgroups: geophysical, e.g. earthquakes; meteorological, e.g. extreme temperatures;

hydrological, e.g. floods; climatological, e.g. droughts; biological, e.g. animal accidents; or extra- terrestrial, e.g. impacts from asteroids, (CRED 2020b). In this thesis I will use the following disaster subgroups: climatological, hydrological, and meteorological disasters. For details on these disaster subgroups, see CRED 2020b. For the purpose of this study, the subgroups biological, geophysical, and extra-terrestrial disasters have been removed from the dataset since they lack an apparent connection to climate change. To explain why, I will exemplify using geophysical disasters. Geophysical disasters include the following disaster subtypes: earthquake, volcanic activity, and mass movement (dry). These are either not at all, or not in an apparent way related to climate change. Therefore, the entire disaster subgroup, i.e. geophysical disasters, has been excluded from my dataset. It could be argued that biological disasters are related to climate change, for example, the subtype insect infestation, but the link is less obvious than that of those disasters included and there are biological disasters which are completely unrelated to climate change (e.g. the subtype ”animal accident”). It is deemed that the inclusion of, for the purpose of this thesis, irrelevant disasters might distort the results more than the exclusions of a few disasters that might be climate related. Therefore, no biological hazards are included in this study. When there is some ambivalence regarding the possible connection between climate change and certain disaster subtypes within a disaster subgroup, I have considered the disaster subgroup as a whole.

This is not only due to time constraints and because there is not always data on what subtype the

disaster in question constitutes, but foremost to avoid situations with possibly arbitrary decisions

on my part based on my interpretation of the disaster description. This also increases the study’s replicability and is thus an advantage from a reliability perspective.

Finally, it is worth mentioning that with this approach, the inclusion of certain disasters might be questioned. For example, extreme temperatures arguably have a strong connection to climate change. However, it is possible that not all extreme temperature events, such as extreme winter conditions, are perceived by the public and officials to be connected to climate change. This is important because if a disaster is not perceived to be related to climate change, it is hard to argue that exposure to such disasters will induce climate change adaptation actions. These events have still been included in this thesis since it is not possible for me to determine how they are perceived, and by attempting to guess what perceptions officials may hold, and by excluding data based on my thoughts of others’ perceptions, I might introduce bias into the dataset. As I have already mentioned, I have decided to not delete subtypes of disaster but instead make a judgement on the disaster subgroup as a whole.

In sum, I have included all disaster subtypes of the subgroups climatological, hydrological, and meteorological disasters. The explanatory variables that this thesis will investigate are related to disaster frequency and to different kinds of impacts of disasters. The data for all of them is collected from CRED (see CRED/UCLouvain 2020).

3.4.1 Disaster frequency

In this thesis, I will use the same criteria for hazards to qualify as disasters as that used by the CRED, which is the definition used in the International Disaster Database (EM-DAT), i.e. the disaster data that constitutes part of the material for this study. This definition is thus useful in relation to the data that will be used in this study. For an event to be classified as a disaster by CRED and thus be collected in EM-DAT, at least one of the following criteria must be fulfilled:

(i) 10 or more fatalities reported, (ii) 100 or more people reported to be affected (iii) a state of

emergency has been declared (iv) there has been a call for international assistance (CRED

2020a). This variable consists of data from the EM-DAT database that register disasters impacts

on a national scale. The variable value is an aggregation of all the disasters experienced by a

country during the studied time period.

3.4.2 Disaster severity

In this thesis, I will study three variables measuring different kinds of disaster impacts. Disaster impact will thus be operationalised as (i) total fatalities, (ii) economic damages (iii) total affected.

The variable total fatalities consists of data on all those reported dead or missing after a disastrous natural hazard event (CRED 2020a). Total affected consists of all those homeless, injured and affected, i.e. the number of people in need of immediate assistance (CRED 2020a). The economic damage variable accounts for the estimated economic damage in current US$, recorded at the year of the disaster. It consists of estimated damage to property, livestock, and crops (CRED 2020a). All these variables consist of data in absolute numbers.

3.5 Control variables

Control variables are used for the purpose of isolating the relationships we are interested in. The control variables used in this thesis are GDP/Capita and population size. If these variables had not been included, the results would likely have been distorted.

3.5.1 GDP/Capita

It is widely acknowledged that there is a positive relationship between the level of development and the amount of resources a country devotes to safety, including proceedings aimed at lessening the negative effects of disasters (Toya & Skidmoore 2006, 20). According to Toya & Skidmoore, income is an important factor in two ways:

First, increases in income increases the private demand for safety. Higher income enables individuals (and by extension countries) to respond to the risks by employing additional costly precautionary measures. However, distinct from this private disaster–income–safety relationship is the existence of an underlying social/economic fabric that increases safety for all of society (2006, 21).

This suggests that a country’s economic status could be an important factor to take into account

when studying the relationship between disaster frequency and severity and adaptation actions

being taken. As mentioned in previous research, disaster losses are larger in developed countries

in absolute numbers, but constitute a greater proportion of GDP in developing countries. This

could be used as an argument to incorporate GDP into the economic damages variable. However,

I have decided to instead use it as a control variable, for reasons I will now discuss. There are

reasons to believe that a country’s economic development, which GDP or GDP/Capita can serve as a proxy for, could have an influence on other variables than the economic damages. It could likely influence the amounts of disasters experienced (not the amount of natural hazards that occur, but whether those hazards turn into disasters), how many are affected and the number of fatalities. In this thesis, I will use GDP/Capita rather than GDP, since GDP/Capita is related to living standards whereas GDP would say more about the total size of the whole economy. Since this variable is intended to work as a proxy for the level of development, I will be using GDP/

Capita. Moreover, GDP/Capita is likely also directly related to the dependent variable in a sense that higher income in itself likely leads to more adaptation actions due to the larger demand for safety and the greater capacity to respond to disasters. Furthermore, there are reasons to believe that disaster impact is less likely to induce risk-mitigating actions in countries with less capacity and thus induce fewer adaptation actions. Although imperfect, a country’s financial resources can work as a proxy for their capacity to act when faced with a disaster. Capacity, or lack thereof, is likely important for how a country responds to a disaster. It likely matters both for the immediate response and for how the country can prepare for future disasters. For example, imagine that a hurricane strikes a country with a relatively low level of economic development. The considerably smaller amount of resources available will then likely be devoted to immediate needs and not focused on the next disaster that might strike. In sum, we may suspect that the level of economic development in a country is relevant in the following ways: by influencing the values of the explanatory variables; by influencing whether the explanatory variables are likely to lead to more actions being taken; and by influencing the self-reporting to CDP of actions taken.

The level of economic development will be incorporated into the regression models in two different ways. In the first four regression tables where countries are separated into groups based on income, GDP/Capita will not be controlled for as a separate variable since it is already incorporated into the design; it is a country’s level of income that determines which income group it will be sorted into (see World Bank Group 2020c). Thus, controlling for GDP/Capita could be misleading. However, GDP/Capita will be included as a control variable in the fifth regression table which contains all countries.

The data on GDP/Capita is collected from the World Bank Group (World Bank Group 2020a). In

this study, the GDP/Capita used is that of 2018 because it ought to be most relevant for the study

since the dependent variable is measured in 2018 and 2019. Where data is missing for the year

2018, GDP/Capita data from the latest available year, from 2013 or later, is used. A threshold was chosen as to not include outdated data. 2013 was chosen since it is the beginning of the time span for studying the natural hazards. When there is no data from 2013 or later, the observation is treated as a missing value. GDP/Capita is in current US$.

3.5.2 Population size

I will also be controlling for population size. Population size is, albeit far from perfect, a proxy for country size. Larger countries likely experience more natural hazards, and the more populated the countries are, the larger the likelihood of those natural hazards affecting humans and thus becoming disasters. Therefore, population size likely affects the number of disasters experienced.

Additionally, population size likely affects the three severity measures investigated in this thesis as well: the larger and/or more populated a country, the more people are likely affected, killed and more property is likely damaged. Furthermore, countries with large populations likely have several large cities. Climate change is likely to have large impacts on urban areas (Georgeson et al. 2016, 584). Moreover, population size as an underlying variable likely interacts with the level of economic development. Georgeson et al. write:

Cities in developing countries are thought to be even more vulnerable to climate change owing to widespread poverty, lack of infrastructure, unplanned informal settlement and a lack of spending on adaptation (Georgeson et al. 2016, 584).

It is also highly likely that population size could have a direct impact on the dependent variable.

It appears likely that the larger the country, and the more administrative units or the more large cities, the more actions will likely be taken, simply because there are more actors and arenas where adaptation decisions can be made and implemented. An alternative to this approach would have been to somehow control for the number of administrative units. However, due to lack of available data, as far as I am aware, and given that population size likely is a decent, albeit imperfect, proxy for this I have not further investigated this possibility. Moreover, population size might be superior in the sense that simply counting administrative units would not take their size into consideration, a factor that likely is important.

Data for this variable is collected from the World Bank Group (World Bank Group 2020b). This

control will be used on both the regressions run on the separate income groups and on the global

data. Where population data was missing for the year 2018, it has been replaced with the most recent population data available (from the same data source) unless the latest available data is from before 2013, in which case it has been treated as a missing value so as not to include outdated data.

3.6 OLS regression

In this thesis, I intend to explore and test disaster frequency and severity in relation to the extent climate change adaptation actions are being taken using statistical methods. In this thesis, Ordinary Least Squares (OLS) regression analysis will be used to test the hypotheses formulated in the previous section. OLS regression has previously been used by other scholars who have studied the impact of natural hazards (see e.g. Zhou et al. 2014). OLS regression is useful for obtaining a measurement of the strength of the investigated relationship between two or more variables (Teorell

& Svensson 2007, 160). To be able to perform multivariate OLS regression analysis, variables are required to be either interval or ratio scale variables (Djurfeldt at al. 2018, 311), a condition that is met by all variables in this thesis. In this study, I will use a 95% significance level.

A possible downside of OLS regression is that extreme values tend to have a large effect on the

regression results since OLS attaches much weight to deviant observations (see Teorell & Svensson

2007, 167). Nonetheless, OLS regression is well suited to studies of exploratory nature, which this

one is. There are several focused case studies investigating the relationship of interest in this thesis,

but as far as I am aware no global overviews. OLS regression gives a good window into the

potential relationships. Moreover, this research method makes it possible to make full use of the

range of the data, since it is possible to carry out with the continuous dependent variable and does

not require condensing it to a binary yes/no - variable. Besides interpreting the correlation

coefficients, I have also included the adjusted R

values. R

is a measure of how much of the

variance that can be attributed to the explanatory variable(s) (Djurfeldt at al. 2018, 357). R

values

should mostly be interpreted as relative explanatory values which are useful when comparing

between different models (Teorell & Svensson 2007, 177). I will use adjusted R

with the purpose of

comparing across models and evaluating their relative fit.

3.7 Critical reflection

As previously mentioned, this study is limited to the time period 2013 - 2017 regarding disaster data and 2018 - 2019 for adaptation actions. The analysis includes cases where the dependent variable is present as well as cases where it is not in order to also include negative cases. Included in the final analysis are thus all countries that have experienced any natural hazards which qualify as disasters during the time period 2013 - 2017 regardless of whether they have reported any adaptation actions or not. The motivation behind including actions from both 2018 and 2019 is rooted in Birkland’s objection that the immediate period after a severe event might see fewer or no actions (see previous research). Exactly what constitutes the time immediately after a disaster is slightly unclear. However, by including actions taken in 2019 as well as 2018, at least over a year will have passed even after actions that have occurred in late 2017. There are of course limitations that stem from this design. For example, it is quite possible that certain actions that might have been influenced by disasters from early in the examined time period have been taken prior to 2018. These actions will be overlooked with this research design. However, the decision to design the study in this way is motivated by the notion that it would be unwise to include something that, in a regression, could be read as a result of something that actually occurred after the possible cause. In other words, this design is meant to ensure that cause precedes any possible effect. Although imperfect, it is arguably preferable to miss certain actions rather than to include them and be unable to determine the time order, which is necessary for any causal explanation.

Furthermore, it could be argued that in order to properly test the theory of cumulative learning, a

longer time period than five years ought to have been studied regarding the independent

variables. However, this would probably have required studying actions over a time period that

would overlap with the time period where disasters are studied, or else face the risk of missing

many actions taken. In a sense, there is a trade-off between studying many important actions and

being able to establish time order. In this study, being able to confirm that the independent

variables precede any actions taken has been prioritised. This design is intended to be able to give

rudimentary evidence regarding two theories and to give an indication of their importance. This

study can hopefully direct future research to study designs more specifically tailored to test a

certain theory.

A validity concern in this thesis is the suspected systemic bias in the data used for the dependent

variable, since it relies on self-reporting. What this bias means for the results has been discussed

at length above. By acknowledging that such a systematic bias likely exists in the data, it can be

taken into account when designing the study and when interpreting the regressions. As previously

mentioned, the suspected bias will be dealt with by separating all countries into four income

groups and running separate regressions, which also has the advantage of enabling an

investigation into whether different kinds of disaster impacts might be of greater importance in

countries with different levels of development, proxied by their income group. The global

regression results can thus be compared with the regression results for the separate income

groups. This design does not eliminate bias, but it makes it possible to work with the likely biased

data.

4. Results

In this section, I will present the regression results. Tables 1 to 4 depict regression results from the four separate income groups. Table 5 presents the results from the regression performed on all countries together.

Table 1. Low income countries

Signif. codes: '***' p value < 0.001, '**' p value < 0.01, * p value < 0.05 (Standard errors in parentheses)

None of the bivariate regressions for low income countries produce coefficients that are statistically significant. Neither are the coefficients in multivariate regressions in Model 5 or Model 6. The regressions give no evidence for any significant relationship between the variables.

However, lack of statistical significance need not mean that there are no relationships in reality.

In this table, it is worth noting that the number of observations is only 29. The relatively low number of observations is a possible reason for why no significant relationships were discovered, although it could also be that there are no relationships. It is also worth paying attention to the low, even negative, values for the adjusted R

. Although low values of adjusted R

are common in the social sciences, negative values do suggest that the models are a poor fit.

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Number of

disasters 0.117

(0.335) 1.809e-1

(5.364e-1) 1.267e-1

(5.346e-1) Economic

damages -1.254e-9

(3.761e-9) -2.901e-9

(4.848e-9) -3.817e-9

(4.878e-9)

Total affected 6.826e-8

(4.0045e-7) 1.609e-7

(4.684e-7) 8.014e-8

(4.703e-7)

Total fatalities 1.740e-3

(4.547e-3) 6.375e-4

(6.769e-3) 7.108e-4

(6.721e-3)

Population

size 9.540e-8

(8.224e-8)

Intercept 2.187

(2.742) 3.149

(1.838) 2.765

(1.976) 2.465

(2.105) 1.721

(3.115) 0.100

(3.394)

Adjusted R² -0.032 -0.033 -0.036 -0.031 -0.142 -0.126

N 29 29 29 29 29 29

Table 2. Lower middle income countries

Signif. codes: '***' p value < 0.001, '**' p value < 0.01, * p value < 0.05 (Standard errors in parentheses)

In the regression table for lower middle income countries, all independent variables appear to be significantly related to the dependent variable both in the bivariate regressions and when controlling for the other independent variables (Model 5) as well as when population is included as a control (Model 6). Note that the variables economic damages and total affected both appear to be positively related to adaptation actions in the bivariate regressions. But when holding the other independent variables constant (Model 5), as well as when also controlling for population (Model 6), total affected and economic damages instead both appear to be negatively associated with adaptation actions.

The adjusted R

values for Models 1 to 4 indicate that the models are at least a better fit for lower middle income countries than they were for low income countries. The relatively high adjusted R

values in Model 5 and 6 can to a large extent be attributed to the fact that more variables are included in the models.

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Number of

disasters 0.992***

(0.109) 7.316e-01***

(9.448e-02) 6.115e-01***

(1.165e-01) Economic

damages 1.151e-09**

(3.345e-10) -1.287e-09***

(1.613e-10) -1.450e-09***

(1.845e-10) Total

affected 1.550e-7**

(4.523e-8) -3.421e-07***

(3.190e-08) -4.625e-07***

(7.798e-08) Total

fatalities 5.607e-3***

(8.029e-4) 1.227e-02***

(1.039e-03) 1.357e-02***

(1.275e-03)

Population

size 3.224e-08

(1.915e-08) Intercept -2.4245

(2.1189) 4.882e+00

(2.894e+00) 5.975*

(2.815) 3.676

(2.182) -1.260e+00

(9.860e-01) -1.347e+00 (9.632e-01)

Adjusted R² 0.6698 0.2132 0.2117 0.5443 0.9385 0.9415

N 41 41 41 41 41 41

Table 3. Upper middle income countries

Signif. codes: '***' p value < 0.001, '**' p value < 0.01, * p value < 0.05 (Standard errors in parentheses)

For upper middle income countries, the bivariate regressions did not produce any significant results. However, when holding the other independent variable constant (Model 5), three significant relationships appear: number of disasters, economic damages, and total affected (although on different significance levels). This suggests that these variables are correlated. When controlling for population size, only number of disasters and economic damages remain significant. Note that economic damages in the multivariate models appears negatively correlated with the dependent variable. Again, the adjusted R

values are negative in several models, suggesting that the models fit poorly. The adjusted R

values increase but remain rather low even when more variables are included (Model 5 and 6).

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Number of

disasters 0.486

(0.554) 11.252***

(2.619) 9.718e+00**

(2.885e+00) Economic

damages 1.062e-10

(6.750e-10) -1.981e-8***

(4.51e-9) -2.135e-08***

(4.657e-09)

Total affected 2.322e-7

(4.740e-7) 7.198e-6*

(2.756e-6) 5.485e-06

(3.072e-06)

Total fatalities 8.151e-3

(0.022) -1.202e-1

(6.161e-2) -9.840e-02

(6.374e-02)

Population

size 4.654e-07

(3.773e-07) Intercept 23.895

(12.887) 2.822e+01*

(1.203e+01) 27.472*

(12.007) 27.121*

(12.424) -26.467

(15.921) -2.671e+01

(1.583e+01)

Adjusted R² -0.005 -0.020 -0.016 -0.018 0.325 0.333

N 50 50 50 50 50 50

Table 4. High income countries

Signif. codes: '***' p value < 0.001, '**' p value < 0.01, * p value < 0.05 (Standard errors in parentheses)

In the regression table for high income countries, three of the bivariate regressions produce significant results with total affected being the exception. When controlling for the other independent variables (Model 5), all independent variables are statistically significant. When controlling for population size (Model 6), all independent variables remain significant. Note that total affected and total fatalities are positive in the bivariate regressions but are negative in both multivariate models (Model 5 and 6), although total affected was not significant in the bivariate regression. Also worth noting in this table is that the correlation coefficient for economic damages remains positive in all models. The relationship is weaker in the multivariate regressions compared to the bivariate, but remains significant. The same is true for number of disasters. Adjusted R

values vary between the different models in this table: some are relatively high (Model 1 and 2) while others are lower (Model 3 and 4), but none are negative.

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Number of

disasters 9.781***

(0.555) 7.246e+00***

(1.195e+00) 6.880e+00***

(1.807e+00) Economic

damages 4.395e-09***

(2.232e-10) 1.862e-09***

(4.607e-10) 1.757e-09***

(4.767e-10)

Total affected 7.238e-5

(3.759e-5) -4.564e-05***

(1.298e-05) -4.812e-05***

(1.325e-05 )

Total fatalities 0.118*

(0.047) -3.854e-02*

(1.692e-02) -4.199e-02*

(1.741e-02)

Population

size 2.561e-07

(5.023e-07) Intercept -19.944

(10.048) 1.662e+01

(8.566e+00) 35.613

(23.897) 34.338

(23.111) -1.888e+00

(7.778e+00) -2.032e+00 (7.862e+00)

Adjusted R² 0.842 0.870 0.045 0.083 0.919 0.919

N 59 59 59 59 59 58

Table 5. All countries

Signif. codes: '***' p value < 0.001, '**' p value < 0.01, * p value < 0.05 (Standard errors in parentheses)

This fifth and final table contains the regressions that have been performed on all countries in the dataset. In the bivariate models, only number of disasters and economic damages produce significant results. They are both positive. When controlling for the other independent variables (Model 5), number of disasters is negative but no longer significant. All other independent variables are significant, although on different significance levels. Economic damages and total fatalities are positively related to adaptation actions while the correlation coefficient for total affected is negative. When controlling for GDP/Capita and population size (Model 6), total affected and total fatalities are no longer significant. Number of disasters is again significant and positively related to adaptation actions. Economic damages is also significant and positively correlated with adaptation actions. It is also worth noticing that population size is in itself significantly correlated with the number of adaptation actions (Model 6), which underlines that including it as a control was likely important. Interestingly though, the relationship appears to be negative in this model.

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Number of

disasters 3.536***

(0.426) -5.149e-01

(5.986e-01) 2.461e+00**

(9.347e-01) Economic

damages 3.696e-09***

(2.174e-10) 4.119e-09***

(3.502e-10) 3.693e-09***

(3.627e-10)

Total affected 1.151e-7

(2.610e-7) -1.154e-06***

(3.195e-07) 1.199e-06 (6.564e-07)

Total fatalities 8.275e-3

(6.270e-3) 1.800e-02*

(7.778e-03) -3.714e-03 (9.296e-03)

GDP/Capita 3.198e-04

(2.207e-04) Population

size -5.776e-07***

(1.424e-07) Intercept -3.254

(8.029) 1.168e+01*

(5.286e+00) 26.167**

(8.541) 24.133**

(8.639) 1.416e+01*

(6.024e+00) 6.537e+00 (7.204e+00)

Adjusted R² 0.276 0.618 -0.005 0.004 0.653 0.685

N 179 179 179 179 179 170

Again, adjusted R

values vary: it is low but not remarkably low for number of disasters,

relatively high for economic damages, very low for total fatalities and negative for total affected.