1. Introduction
In a world facing climate change, improving understanding of how climate affects well-being is important if policy makers are to develop appropriate adaptation and resilience responses. This paper explores the influence of climate factors on the individual well-being of people residing within Australia while controlling for the impact of demographic and socioeconomic variables.
The study of human well-being, happiness, and individual life satisfaction1 has become an important field of research in economics (Ferrer-i-Carbonell 2013). The dominant orthodoxy that utility could not be measured has been challenged by many economists who consider that self-reported subjective well-being may be a valid proxy for experienced utility (Kristoffersen 2010; Welsch and Ferreira 2014). Although many conventional economists still cast doubt over the validity and predictive power of experienced utility (Kristoffersen 2017), new evidence based on longitudinal studies shows an association between measures of human feelings and decision utility across different domains and different countries (Kaiser and Oswald 2022). The new interest in measuring subjective well-being originates from the growing realization that relying solely on income measures such as gross domestic product (GDP) has limitations (Daly and Cobb 1989; Stiglitz et al. 2009). The argument that human well-being cannot simply be narrowed to economic welfare has gained traction (Costanza et al. 2009; Easterlin 1974; Nordhaus and Tobin 1973), and empirical evidence appear to suggest that while GDP per capita has increased dramatically since the 1970s, other measures of well-being have failed to follow the same trajectory, particularly in developed economies (Easterlin 1995; Easterlin et al. 2010; Posner and Costanza 2011) although these findings have been challenged by some (Stevenson and Wolfers 2008).
The independent variables in the model are factors expected to contribute either positively or negatively to individual LS. These can be categorized into broad groups: genetic, demographic, and contextual factors (Diener and Lucas 1999; Emmons and Diener 1986; Larsen and Ketelaar 1991). Existing research indicates that personality traits contribute between 55% and 80% of the variations in LS (Lykken and Tellegen 1996; Røysamb et al. 2018; Tellegen et al. 1988), while demographic variables (age, gender, and marital status) account for 20% and contextual variables including income less than 15% (Diener and Lucas 1999). The positive or negative influence of natural environment variables on human well-being has been relatively well investigated; the literature that has examined the contribution of weather and climate to LS is in comparison more limited.3
This paper proposes alternative LS regression models incorporating climate variables. We then conduct a spatial analysis using geographically weighted regression to examine the variations in the model(s) across different regions of Australia. The outcomes from this research can inform predictions about how the anticipated changes in climate are likely to affect well-being, and so assist in designing policies to build an appropriate adaptive capacity. An understanding of how contextual climate variables interact with other factors to influence well-being will also help in explaining why people move between regions and between countries.
The remainder of the paper is structured as follows: Section 2 reviews the existing literature on climate and LS; section 3 presents the method adopted for this study. In section 4, we describe our data sources and explain how the data was selected and prepared for the purpose of this research. Section 5 lays out the specification of the regression models, while results are presented in section 6. A discussion of our findings follows in section 7; in section 8, we conclude and suggest an agenda for further research.
2. Climate, weather, and LS
Apart from research on climate and LS in the field of economics, a significant body of the broader literature has discussed the effects of weather and climate4 on humans: physiological effects (e.g., heat related diseases) (Vargas and Magana 2020); psychological effects (e.g., mood disorders) (Yang et al. 2021; Zander et al. 2019), society effects (culture, migrations) (Parker 1995). Economic impacts on activities such as agriculture, tourism, labor productivity, and natural disaster management and recovery have also been examined (Chaiechi 2020; Parsons 1993).
Investigations of how weather conditions at the time of survey impact on LS rating by respondents have led to inconclusive results. Higher precipitation and higher temperature may have a negative effect on reported LS scores (Connolly 2013), as may recent cloud cover (Barrington-Leigh 2008); but Levinson (2009) found that precipitation and temperature had a positive effect while Lucas and Lawless (2013) reported little evidence of an association between weather variables and LS judgments. In a longitudinal study conducted over 18 months undertaken in Osaka, Japan, temperature was found to weakly affect happiness while other meteorological variables had no effect (Tsutsui 2013). In their Australian study, Feddersen et al. (2016) show a positive and significant influence of solar exposure but negative effects of barometric pressure and wind speed on LS ratings. They also note that weather bias is more prominent for cognitively complex questions (such as evaluation of overall LS) and that bias declines with panel experience.
Studies that focused on the influence of climate (as opposed to weather) on LS have often used country-level data. Temperature and precipitation variables explain some of the variations in well-being between countries; however, the effects are not uniform across seasons or climate types. It seems that a rise in temperature during colder months increases LS, while a similar rise during hotter months has a negative effect (Frijters and Van Praag 1998; Rehdanz and Maddison 2005). Likewise, warmer temperatures improve LS for people living in cool climates but lower LS in hot climates (Maddison and Rehdanz 2011). In a single-country study conducted in Ireland, increases in winter minimum and summer maximum temperatures are both associated with higher levels of LS (Brereton et al. 2008). On the other hand, a recent Australian study found that heat related stress had no significant impact on either overall LS or momentary happiness (Zander et al. 2019). Precipitation was often determined to be nonsignificant; however, more months with very little rainfall appear to reduce LS (Rehdanz and Maddison 2005). Drought and risk of drought are also shown to have a negative effect on LS, especially among the poorer section of the population (Berlemann and Eurich 2021a). In Ireland, increased mean rainfall has a slightly positive effect on LS. According to the authors, this result may be driven by a positive correlation between high rainfall and scenic beauty (Brereton et al. 2008).5 Also from the Ireland study, wind speed emerges as a significant but negative contributor to LS, and so does surprisingly sunshine duration. But as sunshine duration is negatively associated with rainfall, it is possible that some hidden characteristics of rainfall (frequency, intensity) may bias the results (Brereton et al. 2008).
Many discrete climate variables will have a greater influence on LS when combined, as their individual impacts may be compounded. A study undertaken in Russia shows that strong winds have a significant negative effect on well-being in January when associated with cold temperatures (wind-chill factor); likewise, “stickiness” (interaction between high temperature and high humidity) is found to reduce well-being (Frijters and Van Praag 1998). Other evidence suggests that climate variables may also combine with other factors that amplify or moderate their effect on well-being. For instance, national wealth qualifies the relationship between temperature and happiness: people living in less temperate countries tend to be happier if they are richer and less happy if they are poorer (Van de Vliert et al. 2004).
The impact of single climate events at specific locations has also been researched: drought (Berlemann and Eurich 2021a; Carroll et al. 2009), floods (Fernandez et al. 2019; Luechinger and Raschky 2009; Sekulova and van den Bergh 2016), extreme cold weather (Kraehnert and Fluhrer 2021), extreme heat (Nitschke et al. 2011), and hurricanes (Calvo et al. 2015). The evidence suggests that beyond the obvious immediate effect on welfare, these extreme weather events were detrimental to the LS of people in the affected regions, often for a long period of time. It also shows that the mere risk of natural hazards such as hurricane or tornado decreases LS (Berlemann and Eurich 2021b).
Country-level studies allow the comparison of climate factors in different climatic conditions, but they overlook that the climate is likely to be perceived at a local rather than a national level (Brereton et al. 2008). While the Ireland study addresses this issue, it was conducted in a small country with a relatively homogeneous climate that precludes the analysis of climatic variations between regions. The present study examines the impact of climate variables on LS in Australia, a continent-size country with distinctly different climate zones. While the incidence of specific climate factors has been investigated in the United States (Berlemann and Eurich 2021a,b), to our knowledge, this is the first study that analyses the incidence of a wide range of climate factors in a large country since the Frijters and Van Praag (1998) Russia study.6
3. Method
a. OLS empirical model
b. Choice of spatial scale
As the primary focus of this study is to investigate spatial variations, individual data are aggregated by spatial unit. The spatial scale retained for this study is the Statistical Area Level 2 (SA2) based on the hierarchical scales developed by the Australian Statistical Geography Standard (ASGS). SA2 is defined as an economic and social community of around 10 000 people (Australian Bureau of Statistics 2016). This choice is justified based on the total number of geographical units in Australia and the average number of respondents within each unit: 2019 SA2 areas of a total of 2303 are represented in this study, each containing an average of 86 individual responses. Where the variable is continuous, the mean is calculated for each SA2. For ordinal and categorical variables (e.g., gender), a proportion is calculated for a particular value (e.g., male).
c. Geographic information systems and geographically weighted regression
The use of a geographical information system (GIS) as a tool of investigation seemed appropriate for this study. Within the LS economics literature, examples of studies where a GIS was used include Brereton et al. (2008), Jarvis et al. (2017), and Kubiszewski et al. (2019a). The software used for this study is ArcGIS-ArcMap, version 10.7.1.
While climate is assumed to be fixed over time, it typically varies across regions; therefore, it is highly likely that specific climate factors will impact LS differently at different geographic locations. In statistical terms, this means that the coefficients of climate variables in LS models are expected to vary depending on geographic location, a phenomenon known as spatial nonstationarity. A technique widely used to investigate how spatial nonstationarity affects the relationship between variables across different locations is geographically weighted regression (GWR; Fotheringham et al. 2002). Within the framework of a global model, GWR allows the estimation of the local relationship between contributing variables and the dependent variable (LS). For this purpose, it estimates the value of regression coefficients at each location using a matrix of variable values in the vicinity of the core location (Wheeler and Tiefelsdorf 2005). While GWR is a very useful tool, the presence of local multicollinearity between variables can impair the model’s explanatory power and coefficient estimation (Fotheringham and Oshan 2016; Wheeler and Tiefelsdorf 2005). Prior to undertaking GWR, it is essential to ensure that the overall level of spatial nonstationarity in the OLS model warrants the exploration of spatial heterogeneity in the relationship between variables and to check for possible spatial autocorrelation. Spatial nonstationarity is measured by the Koenker Studentised Breusch Pagan (BP) statistic, and spatial autocorrelation is measured by the global Moran’s index.
4. Data
a. Data sources
The source for LS and personal demographic data used in this study is the Household Income and Labor Dynamics in Australia (HILDA) survey (Melbourne Institute 2022). The data was collected through the 10 latest waves (waves 11–20) of the HILDA survey spanning the period 2011–20. Climate data including rainfall, temperature, solar exposure, humidity, and wind were obtained from The Australian Bureau of Meteorology (BoM) (Bureau of Meteorology 2013). Weather observations cover the period 2000–20 for temperature and rainfall, and 2000–17 for solar exposure, humidity, and wind. By averaging weather observations over a relatively long period of time, we assume that the resulting data represents a climate trend that is fixed over time (NOAA 2021). Contextual socioeconomic indicators were sourced from the Australian Bureau of Statistics (ABS), in particular, the ABS socio-economic index for Australia (SEIFA) from the 2016 census (Australian Bureau of Statistics 2018). Last, the normalized difference vegetation index (NDVI) available from BoM was used as a proxy for natural capital (Bureau of Meteorology 2021). More details about data sources are available in appendix A.
An assessment of the sample representativeness of sociodemographic variables in relation to the 2016 census data was conducted. The results show that overall, the sample was representative of the sociodemographic characteristics of the general Australian population. However, respondents in the sample had, on average, a higher income and were slightly older. The percentage of respondents who own their house (outright or with a mortgage) was also markedly lower than the national average. See appendix B for detailed information about sample representativeness.
b. Data preparation
1) Selection of sociodemographic and contextual variables
The selection of personal demographic and economic variables was in line with the factors determined to be significant in previous LS studies (Ambrey and Fleming 2014; Kubiszewski et al. 2019a). The standard deviation of LS in each area (LSSD) was included as a key explanatory variable pursuing outcomes from previous LS studies that used spatial units, that heterogeneity in levels of happiness within a region itself impacts people’s happiness (Kubiszewski et al. 2019b).
The socioeconomic environment of each SA2 can be described by the decile rankings for the four SEIFA indexes computed by the ABS: relative social disadvantage (IRSD), relative social advantage and disadvantage (IRSAD), economic resources (IER), and education and occupation (IEO) (Australian Bureau of Statistics 2018). The SEIFA deciles were not directly incorporated in the model because of their high level of multicollinearity as the indices are based on similar underlying variables. Instead, we used variables representing the interaction between personal variables and the socioeconomic context: average household income × SEIFA decile for IER (relative income) and proportion with higher education × SEIFA decile for IEO (relative education status). This is justified on the empirical ground that relative income (and to some extent relative education status) matters more to individual happiness than absolute levels of income (Blanchflower and Oswald 2004; Frey and Stutzer 2000; Shields et al. 2009). The list of sociodemographic variables preselected for this study is shown in Table A1 of appendix A.
2) Selections of climate variables
According to biometeorologists, six categories of variables provide a complete climate/weather picture: temperature, precipitation, cloud cover/sunshine, humidity, wind, and atmospheric pressure (San-Gil et al. 1991). The first five categories are represented in this study. Atmospheric pressure may be of interest as an instant weather variable (see Feddersen et al. 2016); however, considering its high variability, it is not believed to be relevant as a climate variable (Keller et al. 2005). Variables describing the climate can be either average measures or extreme values, for example, the maximum temperature in summer. Another form of measurement is based on counts, for example, the number of days with a temperature under 5°C.
Previous research provided guidance for selecting appropriate climate variables. The model should include variables representing each climate characteristic: rainfall, temperature, sunshine, humidity, and wind (Brereton et al. 2008). Multicollinearity should be minimized by avoiding the simultaneous inclusion of variables with a high degree of correlation. Extreme values are often more relevant than averages as they are more easily perceived (Cushing 1987; Van de Vliert et al. 2004). Crossover variables that reflect the recurrence of specific weather conditions should be introduced: for example, wind-chill factor or stickiness (Frijters and Van Praag 1998). The model specification should have an overall consistency; that is, include extreme cold and extremely hot, rather than extreme cold and average maximum. Where appropriate, the likelihood of nonlinear relationship should be addressed by including variables either in squared (Maddison and Rehdanz 2011; Van de Vliert et al. 2004) or log form (Frijters and Van Praag 1998). The list of climate variables preselected for the purpose of this study is shown in Table A2 of appendix A.
3) Dimension reduction
The selection of relevant climate variables is a delicate process, that can lead to some important variables being omitted. Rather than trying to select variables based on heuristics, an alternative approach is to adopt a dimension reduction technique allowing the computation of summary variables based on statistical criteria (Frijters and Van Praag 1998). We used principal component analysis (PCA) available in SPSS, version 26, which allowed the identification of two main composite variables: climate factor 1 representing hot, dry and sunny climate features; climate factor 2 representing both wet and warm climate features. The process and outcome of the PCA are described in more detail in appendix C.
5. Specification of regression models
A dataset containing 2006 valid observations is analyzed using five separate OLS regression models (hereinafter referred to as models 1–5). Each model examines how climate factors in different forms contribute to LS, while sociodemographic variables remain the same. Model specifications are described in Table 1.
Explanatory variables used in the specifications for models 1–5; here, ln indicates the natural logarithm.
Model 1 is based on the premise that extreme climate features influence LS more than average measures and partly replicates the first model specification in Rehdanz and Maddison (2005). Model 2 relies on the concept of heating degrees and cooling degrees as a measure of the discomfort created by extreme temperatures. Following the method used by Maddison and Rehdanz (2011), the model includes heating degree months (HDM) and cooling degree months (CDM). Instead of the North American baseline of 65°F (18.3°C), this study adopts a baseline of 20°C that better reflects the higher average temperatures in Australia. Model 3 uses counts rather than the actual climate measures. The rationale for this approach is that people experience the weather in terms of its duration as well as its intensity (Rehdanz and Maddison 2005).
The use of descriptive climate variables in their primary form implies the assumption of a linear interaction between the climate factor and LS. We can relax this assumption and allow for the possibility of nonlinear relationships by transforming the explanatory variables (Frijters and Van Praag 1998; Maddison and Bigano 2003). Different transformations were considered: square value; primary value + square value; natural log value; a combination of the above. After several iterations, a combination of primary values, square values and log values of various climate factors were included in model 4. Model 5 includes the two climate factors determined through the PCA process described in section 4b(3) as well as two crossover factors.
Prior to estimating OLS regressions, initial checks were conducted for collinearity and behavior of residuals. Scatterplots did not show convincing patterns of collinearity between dependent variables (DV) and independent variables (IV). Correlations between the DV and individual IVs were usually very low (R < 0.1). There was no excessive collinearity between IVs. Residuals plots indicate that the assumption of homoscedasticity holds. Histograms and normal probability plots indicate that the assumption of normality in the distribution of residuals holds. In short, the Gauss–Markov conditions necessary for properly interpreting the series of OLS held.
6. Results
a. OLS regression results
All five OLS models were estimated in SPSS. For each model, two versions were assessed: one version used the specifications described in Table 1, and the second version also included dummy variables representing each of the eight Australian states and territories. A summary of the regression results for the second version is presented in Table 2.
OLS Regression coefficients for models 1–5 (with state dummies). Coefficient significance levels: one, two, and three asterisks indicate p < 0.1, p < 0.05, and p < 0.01, respectively. ACT is the Australian Capital Territory and the other abbreviations are found in the text, n = 2006 for each model, and the model fit adjusted R2 values are 0.227, 0.227, 0.227, 0.229, and 0.219 for models 1–5, respectively.
A preliminary model estimation without climate variables indicates that sociodemographic factors explain 21% of the variance in LS. The inclusion of climate variables increases the explanatory power of the model (measured by adjusted R2) by a value ranging from 0.010 to 0.012; that is, climate variables contribute 1%–1.2% to the total variance in LS.
When state/territory dummy variables are incorporated, adjusted R2 rises by a further 0.08 to 0.13. Models 1–3 have an adjusted R2 = 0.227, model 4 is slightly higher at 0.229, and model 5 is slightly lower at 0.219.
Demographic and socioeconomic variables behave consistently across the models, with similar coefficients and levels of significance. This suggests structural stability and robustness in the relationship between these variables and LS. All variables contribute positively to LS except “being male” and “having a university degree.” The LSSD has a significantly negative coefficient. Relative education status is a significantly positive socioeconomic contributor to LS, relative income is nonsignificant when dummy location variables are included; the natural capital proxy variable (NDVI) is not significant except in model 5.
Variables representing extreme rainfall values are significant (models 1 and 2). Count variables for both dry and wet months are significant (model 3). All factors have positive coefficients except for rainfall in the driest month. Average rainfall in quadratic form [x + (1/1000)x2] is significant (model 4) with a negative coefficient. The results for temperature variables are inconclusive: minimum in the coldest month is significantly negative (model 1), but both HDM and CDM are significantly positive (model 2), while “number of days over 30°C” is significantly positive (model 3). The natural log of “average mean temperature,” and “difference in temperature between coldest and hottest month” are both significant and negative (model 4). The natural log of sunshine hours is significantly positive. Stickiness (positive) and wind chill (negative) are only significant when calculated from log values (model 4). Climate factor 1 (hot and dry) is found to be significant and positive, but climate factor 2 (wet and warm) is not (model 5).
Regression coefficients for the state and territory dummy variables are generally non-significant except for Western Australia (WA), that was consistently significant across all five models. This would suggest a different pattern of relationship between IVs and DV in that state.
b. Geographic investigation results
1) General evaluation
The estimation of the five global OLS models in ArcGIS shows similar results to the SPSS regressions.7 For all models, the Koenker BP statistic is highly significant, indicating the likely presence of nonstationarity in the relationship between variables, thus warranting a GWR investigation. The global Moran’s index is found to be significant in three out of five models implying that residuals may not be randomly spatially distributed (spatial autocorrelation). Clustering of residuals tends to occur in remote statistical areas where the number of respondents is relatively small; hence, averages in those areas are more sensitive to extreme values.
Besides autocorrelation, multicollinearity between factors is another problem that can plague GWR, in particular local multicollinearity that is difficult to detect (Wheeler and Tiefelsdorf 2005). In ArcGIS, multicollinearity (global or local) causes the model to fail. Bearing in mind those constraints, successive trial-and-error estimations led to the identification of a GWR model that is a simplified version of model 5 with 10 explanatory variables consisting of 8 demographic variables and 2 composite climate factors. A comparison between the OLS and GWR models is shown in Table 3. The overall model fit for the GWR model (adjusted R2 = 0.254; the mean value of local R2 = 0.235) is markedly improved from the OLS model with the same specification (R2 = 0.192).8 The mean value of local coefficients for the explanatory variables were close to the OLS coefficients in the original model 5. Only education status was markedly different presumably because relative education status was dropped in the GWR model.
Comparison of regression results between OLS and GWR models. An asterisk indicates that the OLS coeficient is significant at the 90% level.
2) Life satisfaction and sociodemographic variables
Mean LS scores by Statistical Area Level 3 (SA3)9 average 7.9, ranging from a minimum of 6.2 to a maximum of 8.6 (Fig. 1). The relationship between the mean LS score and the LSSD within the area is always negative, with the coefficient ranging from −0.393 to −0.053, and strong negative values in southeast regions and the southern part of WA.
Mean LS scores aggregated by SA3, where LS score of 10 represents the highest possible level of satisfaction with life overall, and 0 represents the least.
Citation: Weather, Climate, and Society 15, 1; 10.1175/WCAS-D-22-0063.1
“Having no long-term health condition” is always significantly and positively related to LS and almost always the largest contributor. The factor is particularly strong in northern Australia and in outback areas. Age (squared) is also a significant positive factor but weaker than health. The strongest impact of age on LS is in western and central Australia. “Being male” is overwhelmingly a negative contributor to LS, particularly so in WA and South Australia (SA). “Being in a relationship” is a significantly positive contributor in a broad arc spanning from SA to central Queensland (QLD), an area containing about 85% of the Australian population.
Household income is a small but significant contributor to LS in most of New South Wales (NSW), QLD, and Tasmania (TAS), in the areas to the east of the city of Melbourne as well as the far northern and far southern WA. House ownership is a significantly positive contributor in NSW, Victoria (VIC), QLD, and most of WA. “Having a university degree” is a negative contributor to LS everywhere except for most of WA where it is not significant.
3) Climate variables
Climate factor 1 is a composite factor that scores positively in regions with a hot, dry and sunny climate and negatively in cool and cloudy areas (Fig. 2). Climate factor 1 is a significantly negative contributor to LS in the hot regions of central Australia and northern WA where the scores are highly positive. Conversely, it is a significantly positive contributor to LS in southeast regions where the factor is neutral (low positive values) reflecting more temperate conditions (Figs. 3 and 4).
Scores for climate factor 1 by SA3; higher scores indicate hotter, drier, and sunnier regions, and lower scores indicate cooler and cloudier regions.
Citation: Weather, Climate, and Society 15, 1; 10.1175/WCAS-D-22-0063.1
Significance of local coefficient for climate factor 1 aggregated by SA3.
Citation: Weather, Climate, and Society 15, 1; 10.1175/WCAS-D-22-0063.1
Local regression coefficient values for climate factor 1 aggregated by SA3.
Citation: Weather, Climate, and Society 15, 1; 10.1175/WCAS-D-22-0063.1
Rainfall during wet months is the main constituent variable in climate factor 2; however, HDM (a proxy for cold temperatures) is also a significant negative constituent. So, regions with a pronounced wet season and warm winters will score positively, while regions with a low rainfall peak and cool to cold winters have negative scores (Fig. 5). Climate factor 2 was a significantly positive contributor to LS in tropical north QLD with warm winters and a pronounced wet season in summer. Conversely, it was a significantly negative contributor to LS in southern regions (SA, western NSW, VIC, and TAS) that experience low or moderate rainfall in summer and cool to cold winters (Figs. 6 and 7).
Scores for climate factor 2 by SA3, with higher scores indicating regions with wetter and warmer climates and lower scores indicating drier and cooler climates.
Citation: Weather, Climate, and Society 15, 1; 10.1175/WCAS-D-22-0063.1
Significance of local coefficient for climate factor 2 aggregated by SA3.
Citation: Weather, Climate, and Society 15, 1; 10.1175/WCAS-D-22-0063.1
Local regression coefficient values for climate factor 2 aggregated by SA3.
Citation: Weather, Climate, and Society 15, 1; 10.1175/WCAS-D-22-0063.1
7. Discussion
There are significant spatial variations in mean LS scores across Australia, supporting findings from previous research (Kubiszewski et al. 2019a). The distribution of mean LS scores reveals a general pattern where LS is higher in regional areas than in metropolitan areas, a fact also noted by Ambrey and Fleming (2014). Some caution needs to be exercised in interpreting this result because of the low number of observations in some remote areas. The coefficient for the LSSD variable is significantly negative across all areas, implying that mean LS was lower in areas with greater heterogeneity in LS scores; this is in line with the conclusions in Kubiszewski et al. (2019a).
The strength of the relationship (measured by local R2) between contributing factors and LS varies significantly between regions. While R2 values range between 0.20 and 0.25 in most areas of eastern and southern Australia, they rise to around 0.30 in the top end of the Northern Territory (NT) and the Kimberley region (northern WA) and to well above 0.40 in southwest WA. Kubiszewski et al. (2019a) had similar results with an even wider range of variation (from 0.12 to 0.78). Overall, LS explanatory models for WA tend to deviate from the general pattern in terms of explanatory power and regression coefficients.10 This invites the following questions:
-
Are there confounding variables that impact on the way sociodemographic and climate variables interact while contributing to LS?
-
Is there a possibility of reverse causality where, for instance, LS has an impact on the way people perceive the climate?
A WA-specific study may shed some light on why factors in this region behave differently than in other regions of Australia. The remainder of this discussion is focused on climate factors.
The global OLS regression models show a small but significant overall contribution of climate variables to LS: low rainfall and high temperatures are both negative contributors, in line with previous findings (Brereton et al. 2008; Rehdanz and Maddison 2005). There is also some indication from the results in model 4 that the relationship between climate variables and LS may not be linear, a conclusion also reached by Frijters and Van Praag (1998) and Van de Vliert et al. (2004).
Model 4 showed a very promising configuration between climate variables and LS; regrettably, local multicollinearity issues prevented further GWR analysis for that model. Another possible limitation is that data from 10 consecutive HILDA surveys were pooled together and that responses collected at different points in time were treated as contemporaneous. It also means that the data include responses from individuals who have participated in multiple surveys. This could bias results in rural areas where the total number of responses is relatively low.
The GWR analysis confirms that the impact of climate factors on LS is generally small and patchy across Australia. The spatial distribution of coefficients for climate factor 1 suggests that where the climate is hot, “hotter” impacts negatively on LS, however, a rise in temperature in milder climates has a positive effect corroborating previous results from Rehdanz and Maddison (2005). Excessive heat in tropical cities can be exacerbated by the heat island phenomenon that may bring daytime temperatures to over 45°C (Chaiechi and Tavares 2019). This is particularly concerning given that the negative effects of extreme heat on mortality and morbidity have been abundantly documented (Hondula and Barnett 2014; Kovats and Hajat 2008; Nitschke et al. 2011). Excessive heat has also been shown to be the third most important motivation behind the intention to leave a tropical city in Australia (Zander and Garnett 2020). The spatial analysis for climate factor 2 indicates that in north QLD the wet and warm climate has a positive influence on LS. It is worth noting that in the Kimberley region of WA and the top end of the NT, regions with a similar tropical climate, the coefficient for climate factor 2 is not significant. This could be linked to the fact that whereas rainfall in tropical north QLD is reliable year-round, it is far more erratic (and sometimes catastrophic) in other northern regions. Conversely, climate factor 2 is a negative contributor to LS in regions where it has negative values (low peak rainfall, cold temperatures): it is not clear whether the cold temperatures, the lack of rain or the combination of both have the most negative effect on LS. Interestingly, the coefficient for factor 2 has significant but weak negative values in regions where temperatures are generally cool, and rainfall is moderate (TAS and VIC) but strong negative values in areas with long dry periods and experiencing cold temperatures (SA and southwestern NSW). These results appear to support earlier conclusions that drought and risk of drought decrease happiness (Berlemann and Eurich 2021a), particularly in farming areas (Carroll et al. 2009), whereas higher and reliable rainfall has a positive impact on LS (Brereton et al. 2008).
A possible limitation to the GWR estimation conducted in this study is that the use of climate factors determined through the PCA process may have led to some important climate variables being omitted. A total of five variables representing hot and cold temperatures, rainfall, and sunshine contribute to the composite factors used in the GWR regression; however, other variables such as windiness and humidity are not included.
8. Conclusions
It transpires from this study that temperature, precipitation, and sunshine likely influence individual LS to some degree, while the effect of other climate variables such as wind speed and humidity is limited. The spatial analysis reveals that for people who live in a hot and dry area, more heat and drought is associated with lower individual LS. This finding does not bode well for the future if the climate does become hotter and droughts more frequent (Lawrence et al. 2022). A possible challenge arising from these results is to build the resilience and adaptive capacity of people in affected communities so that climate change has a less adverse impact on their LS. Current climate change strategies at either federal (Australian Government 2021) or state (Queensland Government 2017) level make a number of strategic recommendations about how to adapt to the effects of climate change. However, while economic costs are abundantly discussed, there are few references to the consequences of these changes on individual well-being. We hope that research on this issue such as this study will help inform future policy initiatives.
The outcomes from this study form the basis of an agenda for future research about the contribution of climate factors to LS. First, we need to further explore the importance of contextual climate in shaping the relationship between specific climate variables and LS. This could be achieved by comparing relationship patterns within contrasting climates (e.g., temperate and tropical climates). Second, spatial differences identified in this study suggest that people may perceive climate in different ways depending on their environment and their circumstances. This prompts the following questions:
-
How do people perceive the climate in their immediate area and how does this perception relate to both actual climate conditions and LS?
-
Is individual perception of climate conditions influenced by specific factors such as occupation, outdoor activity, and level of engagement with nature?
Additional data from rural and remote areas with diverse climates would help our understanding of the spatial variations in LS explanatory models.
Although they are conceptually distinct, the terms “life satisfaction,” “subjective well-being,” and “happiness” are often used interchangeably in the literature (Easterlin 2003). For further discussion of these concepts, see Organisation for Economic Co-operation and Development (2013).
The validity and comparability of subjective well-being measurements have been abundantly discussed in the literature. See, for instance, Kahneman and Krueger (2006), Veenhoven (2008), and Kristoffersen (2010).
For a critical review of the literature on the influence of natural capital on well-being, see Jarvis et al. (2023).
“Climate is what you expect, weather is what you get.”—Mark Twain. Weather refers to short-term changes in the atmosphere; climate depicts what the weather is like in a specific area over a long period of time (National Center for Environmental Information 2018).
It should not be concluded from this correlation that high rainfall is necessarily associated with scenic beauty, although lush green vegetation is generally associated with increased rainfall. In Ireland, regions in the west of the country with high mean rainfall also happen to have spectacular landscapes.
For a deeper review of the relevant literature, see Lignier et al. (2022).
When repeating the OLS models within ArcGIS the state/territory dummy variables were excluded, because, as a result of the spatial basis of the software, the inclusion of categorical variables representing locations is not recommended. ArcGIS also calculates the corrected Akaike information criterion (AIC), a related measure that also evaluates model fitness. More information on the AIC can be found online (https://www.sciencedirect.com/topics/social-sciences/akaike-information-criterion).
AIC also improved from 2670 (OLS model) to 2538 for the GWR model.
To make interpretation easier, the results were aggregated at SA3 levels. Note that this is an aggregation for presentation purposes only; this is not the same as conducting a regression analysis at SA3 level. Of a total 340 spatial SA3 statistical areas, 331 were represented in the sample. SA3s often closely align with large urban local government areas. In the country they represent areas that have distinct identity and socioeconomic characteristics (Australian Bureau of Statistics 2016).
An OLS model with the same specifications as the GWR model was estimated on WA data alone. The variance inflation factors were all < 2, ruling out multicollinearity as the source of endogeneity at least at the global level. Thus, the most likely cause of endogeneity appeared to be either a missing variable or reverse causality. Reverse causality in LS models is discussed in Frijters and Beatton (2012).
Acknowledgments.
We thank James Cook University—in particular, the College of Business, Law and Governance—for supporting our research.
Data availability statement.
Because of the proprietary nature of the Household Income and Labour Dynamics in Australia (HILDA) data, supporting data cannot be made openly available. Further information about the data and conditions for access are available at the Melbourne Institute (https://melbourneinstitute.unimelb.edu.au/hilda). Climate datasets used from this study are available from the Australian Bureau of Meteorology (http://www.bom.gov.au/climate/data-services/) and can be purchased at a moderate cost.
APPENDIX A
Data Sources
a. HILDA survey data
Sociodemographic data are mostly derived from individual respondents’ responses except for household income and housing status compiled by households. The data for this study include responses from 174 857 individual respondents linked to 102 224 households. The data from 10 consecutive years (2011–20) were pooled as if they had been collected through a single survey.
Self-reported overall LS scores are collected through individual questionnaires from answers to the question: “All things considered, how satisfied are you with your life?” Respondents are prompted to choose a score between 0 (totally dissatisfied) and 10 (totally satisfied). There has been an ongoing debate in the happiness economics literature about whether this score should be treated as an ordinal measure or a cardinal measure (Kristoffersen 2017). Cardinality (the distance between individual scores is presumed to be constant) has important consequences in terms of statistical analysis. Despite the valid arguments presented by proponents of the ordinality position (Katzner 1998; Veenhoven 1984), most researchers in the field argue that the theoretical and empirical basis for assuming cardinality is strong (Kristoffersen 2017; Ng 1997). See Table A1 for a list of sociodemographic variables considered for this study.
Preselected demographic, economic, and environment variables. ATSI indicates Aboriginal and Torres Strait Islander. The asterisk indicates a zero value where the number of respondents in SA2 was 1.
b. Natural environment variable
The NDVI is a proxy for natural capital and was used for the same purpose by Kubiszewski et al. (2019a,b). NDVI measures the vegetation density and greenness based on the absorption of visible light and the reflection of infrared radiation (http://www.bom.gov.au/climate/austmaps/about-ndvi-maps.shtml). NDVI scores were collated as the measure of the index on 31 January of each year and then averaged over the period from 2000 to 2019.
c. Climate data
Climate data were compiled from daily observations by weather station for each variable. The compilation of descriptive statistics for the whole period was assisted by the availability of a specialized software tool: ClimateQuery (https://github.com/adamrehn/ClimateQuery). Climate data are in point format (each station being represented by specific coordinates). To calculate average climate values for each SA2, each station’s data needed to be converted into polygon datatype by using the Thiessen Polygon mapping tool available from ArcGIS. Each polygon was then intersected with the geographic SA2s. Descriptive climate variables preselected for this study are listed in Table A2.
Descriptive climate variables preselected for the purpose of this study.
APPENDIX B
Sample Representativeness
The average mean values of key demographic and economic variables used for this study were compared with national averages from the 2016 census (midpoint of the period considered for this study) (Table B1).
Sample representativeness [source: Australian Bureau of Statistics, 2016 Census QuickStat (https://quickstats.censusdata.abs.gov.au/census_services/getproduct/census/2016/quickstat/318?opendocument)].
APPENDIX C
Dimension Reduction
PCA available in SPSS was used as a method of variable reduction. Variables that can be potentially summarized by principal factors must have a degree of correlation between them; however, the degree of correlation must not exceed R = 0.8 because this would create multicollinearity issues. Adequacy of sampling is measured by the Kaiser–Meyer Olkin (KMO) index, which must be > 0.6, as well as the Bartlett’s test of sphericity, which must be highly significant (p < 0.01) (https://statistics.laerd.com/spss-tutorials/principal-components-analysis-pca-using-spss-statistics.php).
To minimize the multicollinearity problems, climate variables were categorized into three separate perception constructs: temperature, rainfall, and sunshine. Stickiness and wind chill have been previously defined as cross factors between two specific variables and therefore were excluded from the pool of candidate variables for factor reduction. A trial-and-error process was initiated to try to determine the pool of variables that would lead to meaningful principal components with the best fit.
The optimal result identified a set of five variables categorized into two principal components. The corresponding rotated component matrix showing the principal constituents and respective contributing variables is shown in Table C1. KMO for this model is 0.703, and the Bartlett test of sphericity is highly significant (99%). The two components explain 82.27% of the total variance of the constituent variables. Variables related to “hot temperature,” prolonged dry weather, and sunshine hours load strongly and positively into component 1, thereafter labeled “climate factor 1” representing hot, dry, and sunny climate features. The main variable represented in component 2 is “rainfall in wettest month,” which loads strongly and positively; however, HDM (a proxy for “extreme” cold temperatures) also loads strongly but negatively. This composite variable, thereafter labeled “climate factor 2,” represents both wet and warm climate features. Standardized scores for each component (shown at the bottom of Table C1) were generated by SPSS using the “regression method.”
Principal component analysis: structure of climate factor components, for the rotated component matrix. The rotation method is Varimax with Kaiser normalization. The rotation converged with 3 iterations.
REFERENCES
Ambrey, C. L., and C. M. Fleming, 2014: Life satisfaction in Australia: Evidence from ten years of the HILDA survey. Soc. Indic. Res., 115, 691–714, https://doi.org/10.1007/s11205-012-0228-0.
Australian Bureau of Statistics, 2016: Australian statistical geography standard: Volume 1-Main structure and greater capital city statistical areas, July 2016. Australian Bureau of Statistics, accessed 2 September 2021, https://www.abs.gov.au/AUSSTATS/abs@.nsf/DetailsPage/1270.0.55.001July%202016?OpenDocument.
Australian Bureau of Statistics, 2018: Socio-Economic Indexes for Areas (SEIFA)—2016. Australian Bureau of Statistics, accessed 20 September 2021, https://www.abs.gov.au/ausstats/abs@.nsf/mf/2033.0.55.001.
Australian Government, 2021: National climate resilience and adaptation strategy: 2021–2025. Australian Bureau of Statistics Doc., 47 pp., https://www.dcceew.gov.au/sites/default/files/documents/national-climate-resilience-and-adaptation-strategy.pdf.
Barrington-Leigh, C. P., 2008: Weather as a transient influence on survey-reported satisfaction with life. University of British Columbia MPRA Paper 25736, 38 pp., https://core.ac.uk/download/pdf/213921453.pdf.
Berlemann, M., and M. Eurich, 2021a: Does drought risk depress expected well-being? Appl. Econ. Lett., 29, 1229–1233, https://doi.org/10.1080/13504851.2021.1922582.
Berlemann, M., and M. Eurich, 2021b: Natural hazard risk and life satisfaction—Empirical evidence for hurricanes. Ecol. Econ., 190, 107194, https://doi.org/10.1016/j.ecolecon.2021.107194.
Blanchflower, D. G., and A. J. Oswald, 2004: Well-being over time in Britain and the USA. J. Public Econ., 88, 1359–1386, https://doi.org/10.1016/S0047-2727(02)00168-8.
Brereton, F., J. P. Clinch, and S. Ferreira, 2008: Happiness, geography and the environment. Ecol. Econ., 65, 386–396, https://doi.org/10.1016/j.ecolecon.2007.07.008.
Bureau of Meteorology, 2013: Climate data online. BoM, accessed 24 June 2021, http://www.bom.gov.au/climate/data/.
Bureau of Meteorology, 2021: NDVI average for Australia. BoM, XX XXX XXXX, http://www.bom.gov.au/jsp/awap/ndvi/index.jsp.
Calvo, R., M. Arcaya, C. F. Baum, S. R. Lowe, and M. C. Waters, 2015: Happily ever after? Pre-and-post disaster determinants of happiness among survivors of Hurricane Katrina. J. Happiness Stud., 16, 427–442, https://doi.org/10.1007/s10902-014-9516-5.
Carroll, N., P. Frijters, and M. A. Shields, 2009: Quantifying the costs of drought: New evidence from the life satisfaction data. J. Popul. Econ., 22, 445–461, https://doi.org/10.1007/s00148-007-0174-3.
Chaiechi, T., Ed., 2020: Economic Effects of Natural Disasters: Theoretical Foundations, Methods and Tools. Academic Press, 664 pp.
Chaiechi, T., and S. Tavares, 2019: Urban growth, heat islands, humidity, climate change: The costs multiply in tropical cities. The Conversation, https://theconversation.com/urban-growth-heat-islands-humidity-climate-change-the-costs-multiply-in-tropical-cities-120825.
Connolly, M., 2013: Some like it hot and not too wet: The influence weather on subjective well-being. J. Happiness Stud., 14, 457–473, https://doi.org/10.1007/s10902-012-9338-2.
Costanza, R., M. Hart, S. Posner, and J. Talberth, 2009: Beyond GDP: The need for new measures of progress. Boston University Pardee Papers No. 4,, 46 pp., https://www.bu.edu/pardee/files/documents/PP-004-GDP.pdf?PDF=pardee-paper-004-beyond-gdp.
Cushing, B. J., 1987: A note on specifications of climate variables in models of population migration. J. Reg. Sci., 27, 641–649, https://doi.org/10.1111/j.1467-9787.1987.tb01186.x.
Daly, H. E., and J. B. Cobb Jr., 1989: For the Common Good. Beacon Press, 482 pp.
Diener, E., and R. E. Lucas, 1999: Personality and subjective well-being. Well-Being: The Foundation of Hedonic Psychology, D. Kahneman, E. Diener, and N. Schwarz, Eds., Russell Sage Foundation, 213–229.
Easterlin, R. A., 1974: Does economic growth improve the human lot? Nations and Households in Economic Growth: Essays in Honour of Moses Abramovitz, P. David and M. Reder, Eds., Academic Press, 89–125.
Easterlin, R. A., 1995: Will raising the incomes of all increase the happiness of all? J. Econ. Behav. Organ., 27, 35–47, https://doi.org/10.1016/0167-2681(95)00003-B.
Easterlin, R. A., 2003: Explaining happiness. Proc. Natl. Acad. Sci. USA, 100, 11 176–11 183, https://doi.org/10.1073/pnas.1633144100.
Easterlin, R. A., L. A. McVey, M. Switek, O. Sawangfa, and J. S. Zweig, 2010: The happiness-income paradox revisited. Proc. Natl. Acad. Sci. USA, 107, 22 463–22 468, https://doi.org/10.1073/pnas.1015962107.
Emmons, R. A., and E. Diener, 1986: An interactional approach to the study of personality and emotion. J. Pers., 54, 371–384, https://doi.org/10.1111/j.1467-6494.1986.tb00400.x.
Feddersen, J., R. Metcalfe, and M. Wooden, 2016: Subjective wellbeing: Why weather matters. J. Roy. Stat. Soc., 179A, 203–228, https://doi.org/10.1111/rssa.12118.
Fernandez, C. J., N. Stoeckl, and R. Welters, 2019: The cost of doing nothing in the face of climate change: A case study, using the life satisfaction approach to value the tangible and intangible costs of flooding in the Philippines. Climate Dev., 11, 825–838, https://doi.org/10.1080/17565529.2019.1579697.
Ferrer-i-Carbonell, A., 2013: Happiness economics. SERIEs, 4, 35–60, https://doi.org/10.1007/s13209-012-0086-7.
Fotheringham, A. S., and T. M. Oshan, 2016: Geographically weighted regression and multicollinearity: Dispelling the myth. J. Geogr. Syst., 18, 303–329, https://doi.org/10.1007/s10109-016-0239-5.
Fotheringham, A. S., C. Brunsdon, and M. Charlton, 2002: Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. John Wiley and Sons, 269 pp.
Frey, B. S., and A. Stutzer, 2000: Happiness, economy and institutions. Econ. J., 110, 918–938, https://doi.org/10.1111/1468-0297.00570.
Frey, B. S., and A. Stutzer, 2012: The use of happiness research for public policy. Soc. Choice Welfare, 38, 659–674, https://doi.org/10.1007/s00355-011-0629-z.
Frijters, P., and B. M. S. Van Praag, 1998: The effects of climate on welfare and well-being in Russia. Climatic Change, 39, 61–81, https://doi.org/10.1023/A:1005347721963.
Frijters, P., and T. Beatton, 2012: The mystery of the U-shaped relationship between happiness and age. J. Econ. Behav. Organ., 82, 525–542, https://doi.org/10.1016/j.jebo.2012.03.008.
Hondula, D. M., and A. G. Barnett, 2014: Heat-related morbidity in Brisbane, Australia: Spatial variation and area-level predictors. Environ. Health Perspect., 122, 831–836, https://doi.org/10.1289/ehp.1307496.
Jarvis, D., N. Stoeckl, and H.-B. Liu, 2017: New methods for valuing, and for identifying spatial variations, in cultural services: A case study of the Great Barrier Reef. Ecosyst. Serv., 24, 58–67, https://doi.org/10.1016/j.ecoser.2017.02.012.
Jarvis, D., P. Lignier, I. Kubiszewski, and R. Costanza, 2023: Natural capital, ecosystem services and subjective wellbeing: A systematic review. Toward an Integrated Science of Wellbeing, E. Rieger et al., Eds., Oxford University Press, in press.
Kahneman, D., and A. B. Krueger, 2006: Developments in the measurement of subjective well-being. J. Econ. Perspect., 20, 3–24, https://doi.org/10.1257/089533006776526030.
Kaiser, C., and A. J. Oswald, 2022: The scientific value of numerical measures of human feelings. Proc. Natl. Acad. Sci. USA, 119, e2210412119, https://doi.org/10.1073/pnas.2210412119.
Katzner, D. W., 1998: The misuse of measurement in economics. Metroeconomica, 49 (1), 1–22, https://doi.org/10.1111/1467-999X.00038.
Keller, M. C., B. L. Fredrickson, O. Ybarra, S. Côté, K. Johnson, J. Mikels, A. Conway, and T. Wager, 2005: A warm heart and a clear head: The contingent effects of weather on mood and cognition. Psychol. Sci., 16, 724–731, https://doi.org/10.1111/j.1467-9280.2005.01602.x.
Kovats, R. S., and S. Hajat, 2008: Heat stress and public health: A critical review. Annu. Rev. Public Health, 29, 41–55, https://doi.org/10.1146/annurev.publhealth.29.020907.090843.
Kraehnert, K., and S. Fluhrer, 2021: Sitting in the same boat: Subjective well-being and social comparison after an extreme weather event. Beiträge zur Jahrestagung des Vereins für Socialpolitik 2021: Climate Economics, ZBW—Leibniz Information Centre for Economics, https://www.econstor.eu/bitstream/10419/242379/1/vfs-2021-pid-49601.pdf.
Kristoffersen, I., 2010: The metrics of subjective wellbeing: Cardinality, neutrality and additivity. Econ. Rec., 86, 98–123, https://doi.org/10.1111/j.1475-4932.2009.00598.x.
Kristoffersen, I., 2017: The metrics of subjective wellbeing data: An empirical evaluation of the ordinal and cardinal comparability of life satisfaction scores. Soc. Indic. Res., 130, 845–865, https://doi.org/10.1007/s11205-015-1200-6.
Kubiszewski, I., D. Jarvis, and N. Zakariyya, 2019a: Spatial variations in contributors to life satisfaction: An Australian case study. Ecol. Econ., 164, 106345, https://doi.org/10.1016/j.ecolecon.2019.05.025.
Kubiszewski, I., N. Zakariyya, and D. Jarvis, 2019b: Subjective wellbeing at different spatial scales for individual satisfied and dissatisfied with life. Peer J., 7, e6502, https://doi.org/10.7717/peerj.6502.
Larsen, R. J., and T. Ketelaar, 1991: Personality and susceptibility to positive and negative emotional states. J. Pers. Soc. Psychol., 61, 132–140, https://doi.org/10.1037/0022-3514.61.1.132.
Lawrence, J., and Coauthors, 2022: Australasia. Climate Change 2022: Impacts, Adaptation, and Vulnerability, H.-O. Pörtner et al., Eds., Cambridge University Press, 1581–1688, https://www.ipcc.ch/report/ar6/wg2/downloads/report/IPCC_AR6_WGII_Chapter11.pdf.
Levinson, A., 2009: Valuing public goods using happiness data: The case of air quality. NBER Working Paper 15156, 34 pp., http://www.nber.org/papers/w15156.
Lignier, P., D. Jarvis, D. Grainger, and T. Chaiechi, 2022: Geography, climate and life satisfaction. Community Empowerment, Sustainable Cities and Transformative Economies, T. Chaiechi and J. Wood, Eds., Springer, 451–473.
Lucas, R. E., and N. M. Lawless, 2013: Does life seem better on a sunny day: Examining the association between daily weather conditions and life satisfaction judgements. J. Pers. Soc. Psychol., 104, 872–884, https://doi.org/10.1037/a0032124.
Luechinger, S., and P. A. Raschky, 2009: Valuing flood disasters using the life satisfaction approach. J. Public Econ., 93, 620–633, https://doi.org/10.1016/j.jpubeco.2008.10.003.
Lykken, D., and A. Tellegen, 1996: Happiness is a stochastic phenomenon. Psychol. Sci., 7, 186–189, https://doi.org/10.1111/j.1467-9280.1996.tb00355.x.
Maddison, D., and A. Bigano, 2003: The amenity value of the Italian climate. J. Environ. Econ. Manage., 45, 319–332, https://doi.org/10.1016/S0095-0696(02)00052-9.
Maddison, D., and K. Rehdanz, 2011: The impact of climate on life satisfaction. Ecol. Econ., 70, 2437–2445, https://doi.org/10.1016/j.ecolecon.2011.07.027.
Melbourne Institute, 2022: HILDA Survey. Melbourne Institute, accessed 20 January 2022, https://melbourneinstitute.unimelb.edu.au/hilda.
National Center for Environmental Information, 2018: What’s the difference between weather and climate? NOAA, accessed 7 August 2020, https://www.ncei.noaa.gov/news/weather-vs-climate.
Ng, Y.-K., 1997: A case for happiness, cardinalism and interpersonal comparability. Econ. J., 107, 1848–1858, https://doi.org/10.1111/j.1468-0297.1997.tb00087.x.
Nitschke, M., G. R. Tucker, A. L. Hansen, S. Williams, Y. Zhang, and P. Bi, 2011: Impact of two recent extreme heat episodes on morbidity and mortality in Adelaide, South Australia: A case-series analysis. Environ. Health, 10, 42, https://doi.org/10.1186/1476-069X-10-42.
NOAA, 2021: What is the difference between weather and climate? National Ocean Service, https://oceanservice.noaa.gov/facts/weather_climate.html.
Nordhaus, W. D., and J. Tobin, 1973: Is growth obsolete? The Measurement of Economic and Social Performance, M. Moss, Ed., National Bureau of Economic Research, 509–564.
Organisation for Economic Co-operation and Development, 2013: OECD Guidelines on Measuring Subjective Wellbeing. OECD Publishing, 270 pp., https://doi.org/10.1787/9789264191655-en.
Parker, P. M., 1995: Climate Effects on Individual, Social and Economic Behavior: A Physioeconomic Review of Research Across Disciplines. Greenwood Press, 304 pp.
Parsons, K., 1993: Human Thermodynamic Environments: The Effects of Hot, Moderate and Cold Environments on Human Health, Comfort and Performance. CRC Press, 359 pp.
Posner, S. M., and R. Costanza, 2011: A summary of ISEW and GPI studies at multiple scales and new estimates for Baltimore City, Baltimore County and the State of Maryland. Ecol. Econ., 70, 1972–1980, https://doi.org/10.1016/j.ecolecon.2011.05.004.
Queensland Government, 2017: Pathways to a climate resilient Queensland: Queensland climate adaptation strategy 2017–2030. Queensland Department of Environment and Heritage Protection Rep., 35 pp., https://www.qld.gov.au/__data/assets/pdf_file/0017/67301/qld-climate-adaptation-strategy.pdf.
Rehdanz, K., and D. Maddison, 2005: Climate and happiness. Ecol. Econ., 52, 111–125, https://doi.org/10.1016/j.ecolecon.2004.06.015.
Røysamb, E., R. B. Nes, N. O. Czajkowski, and O. Vassend, 2018: Genetics, personality and wellbeing: A twin study of traits, facets and life satisfaction. Sci. Rep., 8, 12298, https://doi.org/10.1038/s41598-018-29881-x.
San-Gil, J., J. L. Gonzalez de Rivera, and J. Gonzalez, 1991: Biometeorology of psychiatric disorders. The European Handbook of Psychiatry and Mental Health, A. Seva Diaz, Ed., Anthropos, 40–47.
Sekulova, F., and J. C. J. M. van den Bergh, 2016: Floods and happiness: Empirical evidence from Bulgaria. Ecol. Econ., 126, 51–57, https://doi.org/10.1016/j.ecolecon.2016.02.014.
Shields, M. A., S. W. Price, and M. Wooden, 2009: Life satisfaction and the economic and social characteristics of neighbourhoods. J. Popul. Econ., 22, 421–443, https://doi.org/10.1007/s00148-007-0146-7.
Stevenson, B., and J. Wolfers, 2008: Economic growth and subjective well-being: Reassessing the Easterlin paradox. NBER Working Paper 14282, 102 pp., http://www.nber.org/papers/w14282.
Stiglitz, J. E., A. Sen, and J.-P. Fitoussi, 2009: Report of the Commission on the Measurement of Economic Performance and Social Progress. Commission on the Measurement of Economic Performance and Social Progress Rep., 292 pp., https://ec.europa.eu/eurostat/documents/8131721/8131772/Stiglitz-Sen-Fitoussi-Commission-report.pdf.
Tellegen, A., D. T. Lykken, T. J. Bouchard, K. J. Wilcox, N. L. Segal, and S. Rich, 1988: Personality similarity in twins reared apart and together. J. Pers. Soc. Psychol., 54, 1031–1039, https://doi.org/10.1037/0022-3514.54.6.1031.
Tsutsui, Y., 2013: Weather and individual happiness. Wea. Climate Soc., 5, 70–82, https://doi.org/10.1175/WCAS-D-11-00052.1.
Van de Vliert, E., X. Huang, and P. M. Parker, 2004: Do colder and hotter climate make richer societies more, but poorer societies less happy and altruistic? J. Environ. Psychol., 24, 17–30, https://doi.org/10.1016/S0272-4944(03)00021-5.
van Praag, B. M. S., P. Frijters, and A. Ferrer-i-Carbonell, 2003: The anatomy of subjective wellbeing. J. Econ. Behav. Organ., 51, 29–49, https://doi.org/10.1016/S0167-2681(02)00140-3.
Vargas, N., and V. Magana, 2020: Warm spells and climate risk to human health in the Mexico City metropolitan area. Wea. Climate Soc., 12, 351–365, https://doi.org/10.1175/WCAS-D-19-0096.1.
Veenhoven, R., 1984: Conditions of Happiness. Springer, 462 pp.
Veenhoven, R., 2008: Comparability of happiness across nations. Int. Meeting of the Psychometric Society, Tokyo, Japan, Kwansei Gakuin University, 211–234, https://core.ac.uk/download/pdf/143632798.pdf.
Welsch, H., and S. Ferreira, 2014: Environment, well-being and experienced preference. University of Oldenburg Department of Economics Oldenburg Discussion Papers in Economics V347-14, 44 pp., https://www.econstor.eu/bitstream/10419/105055/1/V-367-14.pdf.
Wheeler, D., and M. Tiefelsdorf, 2005: Multicollinearity and correlation among local regression coefficients in geographically weighted regressions. J. Geogr. Syst., 7, 161–187, https://doi.org/10.1007/s10109-005-0155-6.
Yang, Z., B. Yang, P. Liu, Y. Zhang, and X.-C. Yuan, 2021: Impact of temperature on physical and mental health: Evidence from China. Wea. Climate Soc., 13, 709–727, https://doi.org/10.1175/WCAS-D-21-0038.1.
Zander, K. K., and S. T. Garnett, 2020: The importance of climate to emigration intentions from a tropical city in Australia. Sustainable Cities Soc., 63, 102465, https://doi.org/10.1016/j.scs.2020.102465.
Zander, K. K., S. Moss, and S. T. Garnett, 2019: Climate change–related heat stress and subjective well-being in Australia. Wea. Climate Soc., 11, 505–520, https://doi.org/10.1175/WCAS-D-18-0074.1.
