Climate variations, culture and economic behaviour of Chinese households

Societies adapt to climate variations and develop unique cultures that lead to distinctive economic behaviour across different regions. To estimate the climate-economic link and test the hypothetical role of culture, this paper uses a nationwide survey at the household level in China, together with historical temperature data at the prefectural city level for empirical analysis. The results show the significant role of local climate variations on consumption, savings and investment decisions by households. Harsh weather conditions are associated with lower consumption, lower income and higher savings. Such climate characteristics are also associated with a lower probability of purchasing risky financial assets. Using a sample of migrating families, we find strong evidence that culture is an important channel in the climate-economic relationship. Additional support for this view is found through the “catching up with the Joneses” effect documented in the economics literature. Overall, this research provides an alternative perspective for understanding the long-term behavioural impact of climate change.


Introduction
Climate change has had profound impacts on the global economy, and it has become one of the biggest threats to human sustainability. Increasing temperatures can raise sea levels (Domingues et al. 2008), cause more frequent and severe natural disasters (Aldunce et al. 2015), affect agricultural production (Falco et al. 2019) and restructure the industrial sectors in https://doi.org/10.1007/s10584-021-03145-6 an economy (Babiker 2005). Together, these effects will lead to substantial losses of household welfare (Skjeflo 2013;Ciscar et al. 2011). Policies and collective efforts to mitigate climate change are therefore critically important across the world (Edenhofer 2015).
For climate policies to be feasible and effective, it is important to explore broadly how climate factors affect human society, especially from a socioeconomic perspective. For example, Kriegler et al. (2012) elaborate the need to include this perspective in analysing the impacts of climate change and the importance of creating an integrated system that is incentive driven. Policymakers should realise that the effects of climate change not only apply to macroeconomics but also are fundamentally relevant to all people living on this planet. Although the impacts of climate change on aggregate socioeconomic outcomes have been extensively explored at the macro level, the climate-economic links at the micro-level remain underexamined.
In the literature, the relationship between climate and income has long been recognised since Huntington (1915). Evidence of the negative impact of global warming is abundant-for example, an increase in temperature of just 1°C is associated with a decline in per capita income of 8.5% (Dell et al. 2009). A substantial portion of the differences in income between countries can be explained by temperature alone. Comparing extremely hot or cold years, years in which the temperature is higher (i.e., by 1°C) have economic growth that is 1.1% lower (Dell et al. 2012). From a global perspective, high temperatures and extreme weather can slow the global economy by 0.25% (Carleton and Hsiang 2016), and further global warming is expected to lead to an additional 0.28% decrease. Katz and Brown (1992) suggest that climate variability is more important than average temperature. The economic impacts of climate variability are especially relevant to developing countries, which are less able to cope with environmental challenges (Jury 2002;Thurlow et al. 2012). Moreover, the agricultural sector, which is especially sensitive to changing weather conditions, is more important to these countries. China, as one of the fastest-growing developing economies, is in this category and thus worthy of investigation. Dell et al. (2009) suggest that an alternative method for understanding the time-series dimensions of climate variability is to examine the geographic differences in climate variability and the impacts on household economic outcomes. Several recent studies also focus on the effects of climate variability on households (Skoufias and Vinha 2013;Skjeflo 2013). Our study takes this approach in examining a nationwide household survey in China. Following the main idea in Dell et al. (2009), we examine the response to local climate variation among Chinese households across different regions. In addition, inspired by the idea that "climate creates cultures" (Van de Vliert 2007), we introduce a cultural element into the climateeconomic link. The logic behind this is that society develops its own unique culture as a way of adapting to climate variation. For example, cultural differences between Italians and the British might be due to environmental factors (Tavassoli 2009). These differences can be explained at least partially by contrasting attitudes toward pleasure, which differ between people in countries with a colder climate and those in warmer countries.
The study of culture and climate has been popular in interdisciplinary research for a long time (for a review, see Crate 2011). Human dimensions of climate change are explored in cultural theory (Pendergraft 1998), and cultural knowledge is useful for explaining people's heterogeneous responses to climate change (Nielsen and Reenberg 2010;Crona et al. 2013). Expanding on these concepts, Adger et al. (2013) demonstrate how culture can mediate the linkage between changes in society and climate. Climate change can affect material and lived aspects of a society, creating an important cultural dimension in climate-related perceptions.
Cultural adaptions to long-term climate change are shown to be important and valuable for understanding people's responses to future climate change (DeMenocal 2001;Leonard et al. 2013).
In China, climate variations range from the freezing Mongolian Plateau to the mild Sichuan Basin, and this great variation in local climates leads to a corresponding mix of regional cultures. For example, the Mongolians in the north have a substantially different culture to that of people in warmer southern regions. These cultural differences and their intrinsic links to climate variability translate into consumer behaviour through homeostatic processes (Parker and Tavassoli 2000).
To explore the climate-culture-economy link in China, we use a nationwide household survey, the China Household Finance Survey (CHFS). Combining micro-level survey information with local climate variations, we can effectively establish statistical linkage between climate change and household economic decisions. Climate variations across regions in China are captured by the long-term historical variations in temperatures at the prefectural city level, and then we combine these with data on individual households to determine whether and to what extent climate variability affects household economic outcomes.
Our empirical strategy also includes exploring the role of culture in the climate-economic linkage. Specifically, we use data on migrating families to test our hypothesis about cultural influences. We also extend the "catching up with the Joneses" hypothesis in the economics literature (Abel 1990) for further confirmation. Overall, this research provides an alternative perspective for understanding the long-run impact of climate change and contributes to the existing literature from an interdisciplinary perspective. Given that cultural influences tend to be long lasting; additional implications of climate change can be derived.
The rest of this paper is structured as follows. Section 2 gives a detailed description of the empirical data used in this research, together with the empirical strategies applied in the analysis. Section 3 reports the empirical results, together with a discussion. Section 4 concludes the paper.

Household survey
Household-level data are critical for our empirical analysis. They must have sufficient coverage and representativeness and, at the same time, contain relevant information to perform the empirical analysis. The CHFS survey, collected by the Survey and Research Centre for China Household Finance 1 , is a very good source of information and is well suited for this study. The initial round of the survey was conducted in 2011 and has been repeated every 2 years since then, providing information on consumption, savings, income and other important financial and demographic characteristics of Chinese households across the country. Its sample size has increased to cover more people and subjects, and it has become a very popular source for understanding household behaviour in China (Zhang 2016;Li et al. 2019;Wang et al. 2021).
The first round of the survey has only 8438 samples, so it is not regionally representative. Thus, we use the 2013 round for our research. In this round, the full-sample survey covers 29 provinces, 262 counties, 1048 communities and a total of 27,775 households across China 2 . Even within a province, substantial differences exist in terms of climate variations because of the large size of provinces in the country. Due to data availability and concerns that the regional level should not be too small to have a meaningful cultural difference, we decide to match households and climate information at the prefecture city level instead of counties. This approach gives us a total of 162 valid geographic regions.

Climate data
Given the fact that a long period of climate variability is needed to shape a particular culture, we also need historical data. In addition, human activity can cause climate change (Franzke 2014). It creates reverse causality that can affect statistical results. To avoid this issue, we use data from a reasonably long historical period for our assessment of regional climate variations.
Among all the climate factors, temperature is obviously a good measure among the various factors that determine climate change (Ji et al. 2014). We use data from the National Meteorological Information Center (NMIC) 3 from 1981 to 2010. The original source is taken from 2160 ground weather stations and then compiled to monthly frequency. We exclude temperature information from earlier years for two reasons other than availability. First, the Cultural Revolution (1966)(1967)(1968)(1969)(1970)(1971)(1972)(1973)(1974)(1975)(1976), a major structural change in China, ended a few years before our sample begins. It was a major disruption in almost all respects, especially culture. Therefore, anything related to this period needs to be excluded. Second, our analysis focuses on cross-sectional variability, and a coverage period of over 30 years provides a sufficient description of a particular region's climate.

Variable definitions and descriptive statistics
The key economic variables examined in this paper include disposable income, consumption, savings rate and investment decisions on risky assets 4 . Among these, impacts on consumption are of crucial interest as they relate directly to individual decision-making. When income is fixed, savings rates can be derived from the share of consumption. Notably, climate variations can also affect income at the same time; therefore, their impacts on savings rates might differ from consumption-side effects.
Moreover, consumption can be divided into necessary consumption (i.e. necessities) and discretionary consumption 5 (Halleröd et al. 1997), and each of these might respond to climate variations in different ways. According to the sample statistics, average annual household consumption is RMB 36,140 (about USD 5900 using the 2013 exchange rate). The highest value is as much as RMB 224,300, while the lowest is only RMB 1930, indicating that a large variation of living standards exist in China.
On average, necessary consumption accounts for around two-thirds of total consumption in China, and income from wages is the main source of revenue. Notably, the average ratio of consumption to income is around 257%, which is due to some extremely low values of reported disposable income. The savings rate is relatively low because of some samples have a negative savings rate. These are both distributional issues and do not affect the general conclusions. Moreover, these values are generally consistent with those in several recent studies (Gao and Liang 2019).
Many demographic factors can be extracted from the survey. As reported in Table 1, they include the household head's age, marital status, education level, health status, family size, employment status, political status, risk preference and residential status. These factors allow us to control for the large heterogeneity of household characteristics in the regressions and to make our key results comparable to those in existing studies.
Two variables are constructed from the NMIC data. The first is CV_std, the standard deviation of the monthly average temperature over 30 years (see Supplement Table 1 for the construction of this variable). A high value in this variable for a city indicates greater variation in that city. For example, the highest variation is 15.60, which corresponds to the city of Suihua, Heilongjiang Province. The lowest is 2.77, corresponding to Laibin, Guangxi Province. Figure 1 shows a map of the temperature variations in all 162 sample cities. Areas with Statistical values are calculated after winsoring at 1% and 99% levels (to remove outliers). Other incomes (Income_O) include business income, agricultural income and income from assets. Savings rate is calculated by (income−consumption)/income less climate variation, which are expected to be more comfortable to live in, are all located in southern or south-western China. Cities in the northeast and some northwestern areas show distinctly higher levels of climate variations. The second variable is CV_extreme, to measure the number of days with extreme weather conditions. Here, we define temperatures over 40°C or lower than − 30°as extreme weather conditions. The variable is a simple annual count of the number of days under these conditions; the annual counts are then averaged over 30 years. Given that CV_std and CV_extreme are highly correlated (correlation = 0.372), our analysis concentrates on the first variable, although we use the second variable for additional analysis.
Before discussing our methodology, we plot the link between climate variation and consumption in Figure 2, in which climate variations in the 162 prefectural cities are divided into five quantiles and then linked to the city-average consumption level in the boxplot. Q1 refers to cities with the lowest climate variation, whereas Q5 corresponds to those with the highest climate variation. A clear downward trend is indicated by whether the mean (red diamond) or the median value (black line in the box) is used in each quantile. In other words, cities with the mildest weather (Q1) have the highest average consumption level, whereas cities with the harshest weather (Q5) have the lowest average consumption level. Although the plot does not consider other factors, it still demonstrates the existence of climate-economic linkages in China.
It is also worth noting that the range of consumption in Q5, which covers households living in the harshest weather conditions, is significantly smaller than in others. The expectation of Figure 1 Map of climate variations across sample cities these households of higher variations in the future is stronger, and hence a majority of people in this quantile tend to make similar precautionary savings or behave in a more consistent way.

Research design
To ensure that the trend demonstrated in Figure 2 makes sense, several factors have to be controlled, so we need to use a formal econometric procedure. Here, the climate-economic links are examined by regressing economic factors at the household level on the proxy for local climate variation. We expected that people who live in areas with cold/harsh weather conditions have low sensitivity to pleasure from consumption (Tavassoli 2009), so they have a low propensity to consume. With all other factors being equal, households in areas with harsh weather conditions or regions with high temperature variations tend to consume less. Climate changes and harsh weather conditions can also cause changes in the perception of risk (Menapace et al. 2015). As a result, households develop distinctive risk preferences in the process of adapting to the local climate. Specifically, people living in areas with harsh weather Figure 2 Boxplot of climate variation against city-average consumption conditions must prepare for climate uncertainty, which means they increase their savings (Willows et al. 2003). In other words, we hypothesise that savings rates tend to be higher in areas with harsh weather conditions.
Although the existing literature suggests that countries in colder regions (Dell et al. 2009(Dell et al. , 2012 can have high growth or income, households within a country might have a very different experience. Extreme weather conditions can reduce agricultural output, damage production (Ciscar et al. 2011) and have negative impacts on household income or welfare (Skjeflo 2013;Skoufias and Vinha 2013). From this perspective, the following is a testable hypothesis: with all else being equal, harsh weather conditions are associated with lower income levels.
We test this hypothesis by applying the following linear regression model as a first step in examining the climate-economic relationship in households: where Y ij denotes the consumption/income/savings rates for household i in city j. CV j is the measurement of climate variation in city j, where X ij is a vector of demographic variables, and ε ij is the error term; α is the constant, Γ is a vector of coefficients and β is the key coefficient of interest. Empirically, we expect to see a negative value of β for consumption and income, but a positive value for savings. The second step is incorporating the cultural elements into our empirical model. Note that direct measurement of culture is difficult and often has multidimensional definitions (Beugelsdijk and Maseland 2010; Fellows and Liu 2013)-for example, Kroeber and Kluckhohn (1952) found 164 definitions. Nevertheless, a few clues from these definitions can help us to set up a mechanism for indirectly incorporating culture into climate-economic relationships. The first clue is that culture is formed on the basis of fundamental beliefs and assumptions (Schwartz and Bardi 2001), and thus it is stable over a long period. In other words, the shared values rooted in a culture are unlikely to change, even when people move around. Given this, we expect that migrating families tend to retain their original cultural traditions. If the cultural hypothesis were to be confirmed, then the economic behaviour of migrating families would be affected by climate variations in the region where they originally lived.
In our sample, approximately 8% of the households are migrating families, representing a total of 2113 sample households (for the distribution of this sample, see Figure 3). A small proportion of those samples had moved less than 1 year earlier (132) and between 1 and 3 years earlier (355), whereas a majority of the migrating families in our sample had moved between 3 and 10 years or more than 10 years earlier (663 and 473, respectively). An additional 490 migrating families did not report how long it has been since they had migrated. This distribution in the migrating samples allow us to examine the long-term effects of culture. Nonetheless, we omit the samples of households who had migrated only recently as a mean to check for robustness.
The second clue is that culture is formed collectively among human groups (Fellows and Liu 2013), often large groups and the consequential behaviour in one group distinguishes it from another (Hofstede 2001). Another related concept, which is often used in the asset pricing literature, is "catching up with the Joneses" (Abel 1990) or habit consumption (Campbell and Cochrane 1999). This means that an individual consumption decision can be affected by a habit level or the average level where they live. If the climate-economic link includes a cultural role, then the local average level of propensity to consume should play an important mediating role in individual household decisions. We captured this effect by introducing a city-level average share of consumption in total disposable income. This variable can be interpreted as the propensity to consume-an important decision for households. Our hypothesis is that climate variation can affect the individual propensity to consume and that this link is affected by the local average, which is a proxy for habits or culture.

Baseline regressions
In the climate-economic link, our main interest is in the impacts of climate variations on three main economic variables at the household level: consumption, income and savings rate. Combining city-level climate variations with the household survey data, we perform regressions based on Eq. (1), and the main empirical findings are reported in Table 2. The first three models (models (1)-(3)) study consumption-side impacts, whereas models (4)-(6) report results for income and its subcategories. The last two models are about savings.
To examine consumption, we distinguish necessary consumption from discretionary consumption. The results for models (1) and (2) are statistically significant at the 1% level, and the coefficients are all negative, which indicate that higher climate variation in a city of one standard deviation is associated with a reduction in average consumption of RMB 560. This effect arises mainly from necessary consumption. This finding is particularly worrying, as we believe that this type of consumption ensures basic household welfare.
The coefficients for the other control variables are generally consistent with our basic intuition. High-income families, larger families, married couples, families whose household head has a higher level of education, healthy families and families with party members tend to consume more. Consumption is negatively related to the age and the number of employed

137.320
Consumption consumption expenditure of the household, Consumption_N necessary consumption, Consumption_D discretionary consumption, Income disposable income, Income_W income from wage, Income_O income from other sources, saving savings rate, Risky risky financial assets to total financial assets ratio, CV_std climate variation in cities, size family size; GDP_city city-level per capital GDP in log term, structure city-level industrial structure (share of secondary industries). Robust standard errors are in brackets. Single, double and triple asterisks denote significance levels at 10%, 5% and 1%, respectively. Clustered standard errors are also checked and the results are consistent members in a household. Rural families have significantly lower levels of consumption. Citylevel control variables also have significant impacts, and they are also intuitively consistent with general beliefs. Households in more developed cities, as measured by the gross domestic product (GDP) per capita, consume more, but the industrial share has a negative impact on household consumption 6 . Not surprisingly, higher climate variation is associated with lower income. In particular, a temperature variation in a city that is one standard deviation higher can lead to an average reduction in disposable income of RMB 1200 for households in that city. The main effect comes from wage income, whereas income from other sources increases with climate variation. Presumably, harsh weather conditions reduce working hours and productivity. Other control variables also show a relationship with income. Families with members who are employed, married, better educated, healthy and party members all had a statistically significant and positive relationship with income. Some samples that are affected by these variables differed in wage income and other types of income. For example, party members have relatively higher wage incomes but no significant differences in terms of other types of income. Having a local hukou, or household registration of legal residence in a particular area, can increase wage income but is associated with a lower level of other types of income. Once again, city-level control variables are significant and intuitively consistent.
Reducing consumption naturally leads to the belief that savings rates will increase, but in this situation, the climate effect on income is also negative. Therefore, boosting the savings rate is not a simple consequence of decreasing consumption. We used both ordinary least squares (OLS) regression and the Tobit model (as the savings rate is truncated) to check for robustness, and we confirm a statistically significant relationship between climate variations and savings rates. The effect, however, is economically small. An increase in temperature variation of one standard deviation increased savings rates by only 0.553% (relative to the sample average of 11.51%). Nonetheless, the significant results support our initial hypothesis that people living in areas with harsh weather conditions have a greater need to save for the future because it is unpredictable. Over longer periods, this need will become ingrained as a habit. Of course, we need to interpret this result with caution, as the savings rate variable is calculated based on the difference between consumption and disposable income.
We also use CV_extreme as a key variable for measuring climate, and the results are generally consistent with the main findings above (see Table 3). A few differences, including a significant decrease in discretionary consumption and economically stronger impacts on savings rates, further support our main conclusions from Table 2. We also perform a set of additional analysis to control for other factors that were not covered in the baseline regressions-for example, including household compositions, additional city-level characteristics, quadratic forms of income and wealth etc. There are marginal numerical differences but have no significant impacts on our main conclusions, these results are reported in the Supplement. 6 City-specific control variables may be correlated and thus cause problems. We follow the comments of an anonymous reviewer to check correlations and the sensitivity of our results to these variables. It is found that the correlations are generally low. City level GDP is the most important variable but the main results stand irrespective to the choices (please refer to appendix 7 for more information).

Table 3
Using extreme days as a measure of climate Consumption consumption expenditure, Consumption_N necessary consumption, Consumption_O discretionary consumption, Income disposable income, Income_W income from wage, Income_O income from other sources, Saving savings rate, Risky risky financial assets to total financial assets ratio, CV_extreme extreme climate in cities, controls control variables used in the baseline regressions (Table 2). Robust standard errors are in brackets. Single, double and triple asterisks denote significance levels at 10%, 5% and 1%, respectively 3.2 2. Cultural effects: migrating families As mentioned above, we use two approaches to incorporate cultural elements into the climateeconomic links. The first is the use of migrating families. Cultural effects are persistent and do not change quickly (Lambert et al. 2020); in other words, migrating families tend to retain their original culture even after they move to another location. We explore this idea by looking at some possible ways in which this might occur. Given that consumption is the key decisionmaking variable, we concentrate on the impacts on consumption here. The results are reported in Table 4, which includes five models. The first four models use samples from migrating families, whereas model (5) is the sample without migrating families. The effects of recently migrated families are avoided by allowing models (2) and (4) to replicate models (1) and (3) by excluding families that had moved to a new city within the previous year. As the results show, the effects are generally consistent. First, the coefficients for the climate variation from the migrating families' city of origin (CV_Past) are significant and negative (−0.61/−0.60), whereas the coefficients for the climate variation from the city where they currently live (CV_Present) are not. Second, the effects from non-migrating families are statistically significant (−0.057). These results are generally consistent with our hypotheses and confirm the initial assumption that the climate-economic link includes a cultural role.
It is worth noting that migrating households may have different characteristics from native families (see Supplement 7 for detailed information), which raises a sample selection problem. To resolve this problem and possible biasedness, we adopt a simple propensity score matching (PSM) approach to redo Table 4 (see Supplement Table 7.4). Although the estimated coefficients changed slightly, they are generally consistent with the initial results and provide supporting evidence for the main story. Dependent variables are consumption expenditure (Consumption) in all models. CV_Past climate variation in the region before migrating, CV_Present climate variation in the current residential area, controls control variables used in the baseline regressions (Table 2). Models (2) and (4) are basically the same as models (1) and (3), but exclude migrating families moved within a 1-year period. Model (5) include all non-migrating families (natives). Robust standard errors are in brackets. Single, double and triple asterisks denote significance levels at 10%, 5% and 1%, respectively. Wald tests show no statistical differences between the coefficients estimated for CV_Past in models (1)/(2) and (5)

Cultural effects: catching up with the Joneses
The second approach for modelling cultural effects is to use the local average consumption share to represent local culture. This strategy is also in line with literature on habit formation or "catching up with the Joneses." In other words, individual consumption decisions are often affected by the choices of neighbours. Culture also refers to the behaviour of a reasonably large population group (Fellows and Liu 2013); therefore, using the city-level average to represent local consumption culture is an option. If the climate-economic link has a cultural element, then the share of city-level average consumption should be a significant mediating variable. Table 5 shows the empirical results for the consumption share, as well as the share for necessary consumption, of disposable income. The results in Table 5 provide strong support for our hypothesis. First, a one standard deviation in city climate variation can reduce the average consumption share in a city by 3.64%. Second, a statistically significant mediating role is found in models (1) to (3) and in models (4) to (6). Regardless of the total consumption shares or the share of necessary consumption, we can comfortably confirm a statistically significant role of the city-level average. In other words, a "catching up with the Joneses" effect exists in China, which further confirms the impact of cultural elements.

Conclusions
This paper explores the link between climate variation and household economic decisions using a sample of over 20,000 households in China, together with long-term data on temperature variation at the level of 162 prefectural cities. Our findings contribute to recent debates on the socioeconomic consequences of climate change using detailed micro-level data. In particular, we incorporate cultural elements into this climate-economic link by examining Dependent variables are shares of consumption over disposable income (CS), and necessary consumption share over disposable income (NCS). They are at household level (hh) or city level (city). Controls control variables used in the baseline regressions (Table 2). Robust standard errors are in brackets. Single, double and triple asterisks denote significance levels at 10%, 5% and 1%, respectively migrating families and the "catching up with the Joneses" hypothesis. By revealing the cultural effects, our research offers new insights into the mechanisms through which climate change could bring about long-term behavioural change and, in turn, have significant economic impacts. Our empirical analyses support the hypothesis that a region's harsh climate can lower household income and necessary consumption while raising the household savings rate. An increase of one standard deviation in climate variation can reduce income by RMB 1200, lower consumption by RMB 560 and increase savings rates. These results are robust to an alternative climate measurement.
Using migrating families and a city's average propensity to consume indirectly confirms the role of culture in the climate-economic linkage. Specifically, migrating families, even long after they have settled in their new city, retain their sensitivity to climate variations in their original location. Moreover, the cultural effects are stronger for harsh weather conditions than for milder ones. The average propensity to consume at the city level plays a significant mediating role in the climate-economic link for individual households. We offer further elaboration of the concept of "catching up with the Joneses" by providing additional support for cultural elements.
This research has policy implications from a bottom-up perspective. First, we present similar arguments to support actively addressing climate change at a global scale, conditioned on the fact that global warming leads to greater climate variation and more extreme weather conditions. Second, the results show that households suffer significant welfare losses due to climate variation. These variations can therefore cause further income inequality and, more importantly, a reluctance to consume, which in turn can have a negative economic impact on the macro-economy. Policy supports are needed for regions with harsh weather conditions. Lastly, the long-term effects of cultural elements indicate that the behavioural consequences of climate variation can have long-lasting effects, and thus they are difficult to change. Efforts to cope with and adapt to climate change therefore have both long-lasting and persistent effects.
A few limitations of the current study are worth clarification. First, the sampling issues should be taken with caution. Although we have tried our best to make sure the econometric analyses are robust, the research is constrained by data availability. For example, we need to know more about migrating families to avoid the impacts of non-observable characteristics. Second, the survey is not designed for a study such as this, which thus affect our ability to make any precise analysis. In the future, when long-term data are available, technically more advanced methods can be applied to the same set of questions. Third, further analyses can be done in the future using the rich information of this micro-survey to provide more explanations to some of the results here; for example, a more accurate matching between household and climate conditions. This, of course, is subject to data availability.