Extreme weather events and pro-environmental behavior: evidence from a climate change vulnerable country

ABSTRACT

Experiencing an extreme weather event and its consequences might make the risks associated with climate change more tangible, easier to evaluate, and more salient. Consequently, those experiences might translate into the adoption of pro-environmental behaviours. Understanding this relationship is fundamental for the successful design of policies aimed towards promoting the adoption of climate change adaptation and mitigation measures. This work contributes to the literature by showing that there is in fact a positive relation between experiencing an extreme weather event and willingness to take pro-environmental actions. The prevailing available evidence is for developed countries. Our empirical analysis is based on a nationally representative sample of households from Mexico, a developing country that is highly vulnerable to the effects of extreme weather events.

KEYWORDS:

• Extreme weather event
• pro-environmental behaviour
• willingness to adopt
• Mexico
• climate change

JEL CODES:

• Q2
• Q5
• Q54
• D80

I. Introduction

The frequency and intensity of extreme weather events (EWE), such as droughts, heatwaves, and hurricanes, are expected to increase because of climate change (Seneviratne, Nicholls, and Easterling et al. 2012). These events can have severe economic (Botzen, Deschenes, and Sanders 2019) and ecological consequences (Nelson et al. 2013). The risks and impacts of these events are unevenly distributed across the planet; local contexts determine who is affected, how, and for how long (Olsson et al. 2014). Individuals in developing countries, especially the poor and disenfranchised, face the highest risk of being negatively affected (Olsson et al. 2014).

If they are to reduce the threats of climate change, households and individuals will need to engage in adaptive actions and support and implement mitigation strategies (Van Valkengoed and Steg 2019). One of the multiple factors that explain individuals’ willingness to adopt these behaviours is their perception about climate change and the risks it entails (Grothmann and Patt 2005). Individuals’ exposure to EWE can be particularly important in terms of explaining perception of climate change (Demski et al. 2017; Ogunbode, Doran, and Böhm 2020). Individuals can use a simple observable outcome, the EWE, as a substitute for a broader understanding of climate change (Gärtner and Schoen 2021). In this way, personal experience of an EWE can act as an attribute substitute for climate change, making the risks associated with it more tangible, easier to evaluate and more salient (Broomell, Budescu, and Por 2015; Gärtner and Schoen 2021).

Although the link between EWE experiences and climate change beliefs has been thoroughly studied, there is a need for more research in developing countries, especially those that are highly vulnerable to climate change; most of the available analyses refer to developed countries and there is almost no research for either Africa or Latin America (Broomell, Budescu, and Por 2015; Van Valkengoed and Steg 2019). We contribute to this literature by analysing the case of Mexico, a country that due to its location, geographic characteristics, low levels of adaptive capacity, and high levels of poverty is highly vulnerable to the negative effects of EWE (Arceo-Gómez, Hernández-Cortés, and López-Feldman 2020). We look at the relationship between direct personal experience with an EWE and willingness to adopt a series of pro-environmental actions related to water, energy and waste management.

II. Materials and methods

We use data from the Household and Environmental Module (MOHOMA) of Mexico’s National Income and Expenditure Survey (ENH). The MOHOMA was designed and implemented by Mexico’s National Institute of Statistics, Geography and Informatics to be representative of the whole country. The questionnaire was applied during 2017 and captures information for 2016 (INEGI 2017). In addition to basic individual and household characteristics, the questionnaire captures information related to households’ experience with EWE during 2016; this is our variable of interest. Thirteen percentage of the households in our sample responded that they had experienced an EWE during the previous year; of those, 95% said that they had been directly affected in a negative way (e.g. dwelling damaged, loss of job, health afflictions) by that event. Additionally, questionnaire respondents were asked about their willingness to adopt, in an unspecified future date, different pro-environmental actions related to water, energy and waste. With this information, the count and binary variables that act as dependent variables in the empirical analysis were created (see Table 1).

Table 1. Descriptive statistics of the variables included in the statistical analyses.

Variable	Description	Mean	Standard deviation
Dependent variables
Water-related actions	Number of water-related actions (pay a higher price for water, instal water-saving technology, fix water leaks, and check water bills for consumption levels) that respondents reported they were willing to take. Maximum is 4 actions.	2.95	0.64
Energy-related actions	Number of energy-related actions (e.g. pay a higher price for energy, buy energy-efficient equipment, etc.) that respondents reported they were willing to take. Maximum is 5 actions.	3.89	0.62
Waste-related actions	Number of waste-related actions (e.g. pay for waste disposal based on the amount generated, waste separation, etc.) that respondents reported they were willing to take. Maximum is 5 actions.	4.06	1.07
Willingness to pay more for water, energy, or waste	= 1 if the respondent said she would pay a higher price for water, energy, or waste disposal	0.50	0.50
Variable of interest and covariates
Experience	= 1 if during last year the household was affected by an extreme weather event (e.g. drought, hurricane)	0.13	0.34
Individual and household level covariates
Age	Age in years	43.01	16.87
Gender	= 1 if female	0.54	0.50
Elementary	= 1 if individual has only elementary school education	0.26	0.44
Middle school	= 1 if individual has only middle school education	0.27	0.45
High school	= 1 if individual has only high school education	0.19	0.39
University	= 1 if individual has a university degree	0.18	0.39
Wealth	Wealth index constructed using household characteristics and Principal Component	5.49	2.54
Household size	Number of household members	3.64	1.83
Locality and municipality level covariates
Rural	= 1 if the locality where the household is located has less than 2,500 inhabitants	0.24	0.43
Poverty	Percentage of population below the poverty line at the municipal level	42.34	19.01
Density	Population density at the municipal level (person per square kilometre)	1069.5	2620.6
Regional level covariates
Hot	= 1 locality is in a hot region	0.28	0.45
Template	= 1 if locality is in a template region	0.48	0.50
Tropical	= 1 if locality is in a tropical region	0.25	0.43

Authors’ estimations using data for 2017 from the Household and Environmental Module (MOHOMA) of Mexico’s National Income and Expenditure Survey (ENH), and from Mexico’s National Council for the Evaluation of Social Development Policy.

The association between experiencing an EWE and willingness to take pro-environmental actions is analysed by estimating a series of Poisson and probit models. For each one of the three variables that measure the number of actions that an individual is willing to take, we estimate four different specifications of the Poisson model.¹ In the first specification, the only independent variable is the binary variable for experiencing an EWE. For the second specification, we add the following individual and household level variables: age, gender, elementary school, middle school, high school, university, wealth and household size (see Table 1 for a description of each one of the covariates). For the third specification, the following locality and municipal level variables are added: rural, poverty and density. Finally, for the fourth specification, we add region indicators; this allows for the possibility that our results could be affected by unobserved geographic regional factors. In this way, we try to isolate the effect of experiencing an EWE from other confounding factors. The full, and preferred, model is the following:

$E\left[y{}_i^j\right]=exp(\beta {}_0^j+\beta {}_1^{j}experienc{e}_i+X_i\delta {}^j+regio{n}_i)$

where $y{}_i^j$ is the number of actions that individual $i$ is willing to take in one of the $j$ categories (water, energy or waste management), $experience$ is equal to one if the household in which individual $i$ lives was affected by an EWE, $X_i$ is a vector that includes all the individual, household, locality and municipal controls previously mentioned, and $regio{n}_i$ are regional indicator variables. To complement this analysis, we estimate the association between experiencing an EWE and the probability that an individual will say that she is willing to pay more for water, energy or waste disposal. To do this, we estimate probit models for the same four specifications explained above.

III. Results

Table 2 shows the results of estimating Poisson regressions to look at the association between experiencing an EWE and the number of pro-environmental actions that an individual is willing to take. Panel A shows results for pro-environmental actions related to water. In the specification without any covariates (column 1), we find that experiencing an EWE increases the number of water-related pro-environmental actions that an individual is willing to take. The statistical significance of this effect remains very high when individual and household covariates (column 2), municipal and locality covariates (column 3), and region indicators (column 4) are included. Results for actions related to energy are a lot weaker (Panel B). The estimated coefficient is only weakly significant for the specifications that include municipal and locality covariates (column 3), and regional indicators (column 4). Results for waste-related actions (Panel C) are highly statistically significant across all specifications. Although there are no studies directly comparable to ours, the results presented here are in line with those by Demski et al. (2017), who show that experiencing an EWE can have effects on behavioural intentions not directly related to the event.

Table 2. Estimations of the impacts of experiencing an extreme weather event (EWE).

	(1)	(2)	(3)	(4)
Panel A: Water-related actions
EWE experience	0.019***	0.025***	0.028***	0.028***
	[0.005,0.033]	[0.011,0.039]	[0.014,0.043]	[0.013,0.042]
Panel B: Energy-related actions
EWE experience	0.001	0.004	0.008*	0.009*
	[−0.008, 0.010]	[−0.005, 0.013]	[−0.001, 0.018]	[−0.001, 0.018]
Panel C: Waste-related actions
EWE experience	0.049***	0.047***	0.045***	0.045***
	[0.034, 0.063]	[0.033, 0.062]	[0.030, 0.060]	[0.030, 0.060]
Panel D: Willingness to pay more for water, energy, or waste
EWE experience	0.200***	0.191***	0.133***	0.129***
	[0.118, 0.283]	[0.107, 0.274]	[0.048, 0.219]	[0.044, 0.215]
Individual/household level covariates	No	Yes	Yes	Yes
Locality/municipal level covariates	No	No	Yes	Yes
Regional indicators	No	No	No	Yes
Number of households	14502	14502	14502	14502

Authors’ estimations using quasi-maximum likelihood Poisson estimations (panels A to C) and probit estimations (Panel D). Data for 2017 from the Household and Environmental Module (MOHOMA) of Mexico’s National Income and Expenditure Survey (ENH). Each panel represents a different dependent variable. For panels A to C, the dependent variable is the number of actions that the individual said she was willing to take. For Panel D, the dependent variable is equal to one if the individual said she was willing to pay a higher price for water, energy, or waste disposal, and zero otherwise. Each coefficient represents an estimate from a separate regression. Individual and household level covariates are: age, gender, education indicators, wealth and household size. Locality and municipal level covariates are: rural, poverty and density. Robust 95% confidence intervals are shown in brackets. * p < 0.10, ** p < 0.05, *** p < 0.01.

Figure 1 shows the average marginal effect that experiencing an EWE has on the expected value of the different pro-environmental groups of actions; for all the estimations we use the coefficients from the full specification. An individual affected by an EWE during the previous year is willing to take, on average, almost 0.1 more pro-environmental water-related actions than an individual from a household not affected by an EWE. The point estimates for the average marginal effects on waste actions are almost twice as big. Meanwhile, the marginal effects on energy are much smaller in magnitude and only statistically significant at a 10% level.

Figure 1. The effects of experiencing an extreme weather event on the number of pro-environmental measures that an individual is willing to adopt across different categories. Estimates of marginal effects shown with 95% robust confidence intervals. The Poisson models behind each estimation are those described in Panels A to C in Table 2.

The results for the relationship between experiencing an EWE and willingness to pay more for water, energy or waste disposal are shown in Panel D of Table 2. The estimations are very robust across the different specifications and show a positive relationship. Average marginal effects, using the full specification, show that individuals affected by an EWE are five percentage points (95% CI: 1.7, 8.3) more likely to say that they are willing to pay more for water, energy or waste disposal, than those not affected by an EWE.

IV. Conclusions

This work contributes to the literature by showing that experiencing an EWE can increase individuals’ willingness to take pro-environmental actions that are not directly related to the nature of the event experienced. Nevertheless, an important limitation of our work is that there is nothing to say about the effect that experiencing an EWE has on actual behaviour. Another limitation is that, with the data available, we cannot discriminate between different hypotheses that can explain the underlying mechanisms behind our results. Finally, identifying causal effects is restricted by the fact that we do not have longitudinal data or data resulting from an experimental setting.

Our analysis is based on a nationally representative sample of households from a developing country that is highly vulnerable to the effects of EWEs. Improving our understanding of the factors that explain pro-environmental decisions in settings like this, can be very relevant for the design of policies aimed towards promoting the adoption of climate change adaptation and mitigation measures. In this sense, the results presented here suggest that targeting some mitigation and adaptation policies on households recently affected by an EWE is something that might be worth exploring. For example, one policy intervention could be to provide affected households with information that links climate change with the occurrence of EWEs similar to the one that they have experienced. This could be complemented by an intervention that provides information about climate change adaptation and locally available mitigation options. Arguably, these interventions could reinforce households’ perception and awareness of climate change, while also modifying their behaviour in terms of adoption of mitigation and adaptation actions. In any case, more research and additional data is needed to better understand the causal relationship between experiencing EWEs and the adoption of pro-environmental actions.

Acknowledgments

We thank Danae Hernández-Cortés, Isael Fierros, participants at the 16^th annual meeting of the Environment for Development Initiative, and an anonymous reviewer, for their valuable comments and suggestions.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.

Notes

¹ We use a quasi-maximum likelihood Poisson estimator, which has the advantage of being consistent regardless of the form of the variance (Wooldridge 2010). As a robustness check we estimated binomial models to consider the fact that our count data is conditional on an upper bound. The results of that model are qualitatively the same as those presented here; the marginal effects are almost indistinguishable across the different estimations. Finally, since our data does not show overdispersion, estimating negative binomial models leads to the same results as estimating Poisson models.