Impacts of Climate Variation on Rural Populations: Evidence from Vietnam

ABSTRACT Climate change and increasing climatic variability have exacerbated the climate-related effects on various aspects of human socioeconomic activity. The effects of droughts on crop production, household expenditure, rural poverty and internal migration in Vietnam were investigated using a unique panel dataset while controlling for potential issues that could lead to biased estimations. We find comprehensive evidence that droughts not only caused a significant decline in rice productivity and rural household consumption, but also pushed them into poverty. Furthermore, droughts are strongly linked to rural people's decisions to migrate to other areas, pushing migration decisions through the productivity, consumption and poverty channels. Understanding the complex relationship between climate change, particularly drought, and its consequences is critical for developing policy interventions to help rural populations cope with climate change risks and uncertainties. These findings emphasize the importance of increasing rural populations’ resilience to climate-related extreme events to mitigate their interconnected effects.


Introduction
Climate change has been a global threat for the past several decades, resulting in an increase in the frequency and intensity of extreme weather events such as droughts (Seneviratne et al. 2012;Nguyen, Ancev, and Randall 2019;IPCC 2012).Undoubtedly, the impact of climate change, in general, and droughts in particular, is devastating.Hsiang (2010) estimated the loss caused by climate change and suggested that 'a temporary 1°C increase in surface temperature is associated with contemporaneous 2.5% reductions in economic output' (15367).Natural disasters slow the growth and development of disaster-prone nations by one percentage point compared to their counterparts (Cantelmo, Melina, and Papageorgiou 2023).Developed countries experience greater absolute economic losses as a result of natural disasters and climate change than low-income countries, but the latter's relative impact is often striking because of its higher share of GDP losses due to natural disasters (Arouri, Nguyen, and Adel 2015).Furthermore, the lack of safety nets and risk-sharing mechanisms makes lowincome economies more vulnerable to extreme climatic events, particularly droughts and floods (Chaijaroen 2019).Climate change is likely to threaten agricultural production in developing countries, including Vietnam, owing to the direct exposure to climatic variables (Grörer and Zyleberberg 2016).
Vietnam is considered one of the countries most prone to natural disasters, ranking 13th in the Climate Change Vulnerability Index (Lohmann and Lechtenfeld 2015).There were long-term trends in climatic variables, indicating the effects of climate change throughout Vietnam.Nguyen, Ancev, and Randall (2019) confirmed spatio-temporal climate variability and changes in the country and suggested that existing long-term changes in temperature and precipitation patterns have detrimental effects on agricultural production and farming households' welfare across regions.Climate variability increases both the frequency and intensity of natural disasters in Vietnam, especially the most frequent climatic events such as droughts, floods and typhoons.Floods, typhoons and droughts were the most common natural disasters in Vietnam between 1900 and 2016 (see Nguyen et al. 2020a and Appendix 1).It is estimated that environmental shocks may affect 10-20% of Vietnam's population and lead to an approximately 10% loss of GDP (Nguyen, Ancev, and Randall 2019).According to Lohmann and Lechtenfeld (2015), an economic loss caused by a single drought in 2005 was estimated approximately 0.2% of Vietnam's GDP (UNISDR 2011).Using a nationally representative household survey called VARHS in 2014 and 2016, Huynh, Saito, and Nguyen (2022) indicated that drought is the most frequent natural disaster, accounting for approximately 30% of the number of shocks affecting surveyed households.Thus, as droughts are emerging threats to rural households, there is increasing research interest in the impact of climate-related events on human life.
Increasing climatic variability and changes have intensified other climate-induced consequences on various aspects of human socio-economic activities such as decreased crop productivity, diminished quality of life and forced migration.The impacts of climate variability and extreme events, particularly droughts, have been thoroughly examined in the literature.Numerous empirical studies have investigated the impact of climate change on crop productivity (Korres et al. 2017;Zhao et al. 2017;He et al. 2020;Nguyen et al. 2022), household welfare (Bui, Nguyen, and Pham 2014;Boansi et al. 2021;Ngoma et al. 2021), poverty (Bangalore, Smith, and Veldkamp 2019;Skoufias et al. 2020), human development (Rodriguez-Oreggia et al. 2013) and so forth.However, few studies have simultaneously considered multiple climate-induced consequences associated with a single extreme weather event such as drought, in Vietnam and globally.This leaves a gap in the literature and this study sets apart previous studies by exploring the comprehensive impacts of droughts on various aspects of rural households.Specifically, we attempted to address the following three main research questions: Does drought negatively affect household crop production and consumption?Is drought linked to household poverty?Does exposure to drought cause household members to migrate?.
This study sought to provide solid empirical evidence of drought-induced effects on rice productivity, household consumption, poverty and migration, using a unique Vietnamese household panel dataset.We employed instrumental variables with a fixed-effects estimation to explore the connections between droughts and the outcome variables of interest at the household level.Previous studies have identified potential endogeneity issues when measuring household exposure to drought using self-reported data, owing to the fact that data reported by respondents could suffer from reporting biases.As argued, measurement errors and reporting biases can occur when respondents are asked to recall the natural disasters they have experienced over a certain period (Lohmann and Lechtenfeld 2015;Berlemann and Steinhardt 2017).Thus, identifying individuals' exposure to natural disasters based on selfreported data presents a significant endogeneity issue, which may lead to biased estimates.This study is the first to address the endogeneity of household exposure to drought in Vietnam using temperature and precipitation anomalies as instrumental variables, while controlling for commune-and year-fixed effects.We complemented the surveyed household data with weather data for climatic variability and anomalies, which sets it apart from previous studies of the effects of natural disasters on crop production, household welfare and internal migration.To the best of our knowledge, this is one of the few empirical studies that has considered multiple climate-induced consequences on rural populations while controlling for potential issues that could lead to biased estimates.
The remainder of this paper is organized as follows.A relevant literature review is presented in Section 2. Section 3 presents the conceptual framework, data and empirical model.Section 4 presents our estimation results.Finally, Section 5 provides the concluding remarks.

Literature review
The impact of natural disasters has been examined in a large body of socioeconomic literature (Cantelmo, Melina, and Papageorgiou 2023).Natural shocks such as floods and droughts have detrimental impacts on household income and consumption and are strongly correlated with poverty.According to Adego, Simane, and Woldie (2019), climate variability results in substantial welfare loss for smallholder farmers.Bui, Nguyen, and Pham (2014) investigated the impact of natural disasters on household income, expenditure, poverty and inequality.To address the endogeneity of dummy variables for natural disasters, this study employed fixed effects and Instrumental Variable (IV) estimation.Arouri, Nguyen, and Adel (2015) emphasized that the level of resilience of households and communities influences the nexus between natural disaster impacts and household welfare, while Lohmann and Lechtenfeld (2015) focused on the impact of drought on health outcomes and expenditure.This empirical evidence shows that extreme climate events significantly reduce household welfare proxied by expenditure and income, deteriorate health, and exacerbate poverty and inequality levels.Akaakohol and Aye (2014) shed light on the welfare effect of diversification on farming households.Most farmers are smallholders who lack government support in response to natural disasters such as drought, floods and global warming.Therefore, some farm households diversify their income sources and participate in nonfarm activities.Based on Ordinary Least Square (OLS) estimation, the study provided robust evidence that income diversification is associated with household welfare, as indicated by food and nonfood consumption patterns.
Several recent studies have examined the economic impact of climate change.Boansi et al. (2021) investigated the effects of rainfall shocks on the welfare and crop productivity of farmers in Ghana.The estimation results show that the expected total household income decreased by 7.3% and 42.5%, respectively, under the lack-of-rainfall scenario.Manuel et al. (2021) 2021) provided a different finding regarding the impact of rainfall shocks on various crop productivities, indicating that excess rainfall increased the yield of maize, millet, rice and groundnut crops.He et al. (2020) also provided empirical evidence that climate change is associated with the growth of rice yields in China in the short-run.
Overall, there is consensus on the causal relationship between climatic variability and international migration flows.Most previous studies have confirmed the impact of climatic factors on international migration to OECD countries (Coniglio and Pesce 2015;Cattaneo and Peri 2019).For instance, Coniglio and Pesce (2015) confirmed that past climatic shocks had direct and indirect push effects, increasing out-migration flows from poor to wealthy nations, particularly from countries of origin with large agricultural sectors.According to Beine and Parsons (2015), long-term climatic conditions do not directly influence international migration flows, but rather have indirect impacts through wage disparities.The agricultural sector and wage differences play significant roles in the climate-international migration nexus.As for internal migration at the micro-level, some robust evidence confirms the push effect of climatic factors on internal migration in developing climatic-prone countries such as Sub-Saharan Africa (Barrios, Bertinelli, and Strobl 2006;Marchiori, Maystadt, and Schumacher 2011;Henderson, Storeygard, and Deichmann 2017;Mueller et al. 2020).Additionally, Gray and Mueller (2012b) investigated the effect of drought on population mobility in the rural Ethiopian highlands.Empirical research indicates opposite findings for men and women, in which drought significantly induces long-distance male migration and labor motives, but extreme weather events reduce female migration for those with marriage motives.Sedova and Kalkuhl (2020) used weather data to construct longterm weather anomalies for the environment-induced migration in India.The study is the first to reveal how migrant characteristics affect the climate-migration nexus.The estimation results show that climateinduced migrants tend to have lower literacy or skills and come from agriculture-reliant households, whereas more educated people are likely to stay.Using logistic regression estimates of rural male out-migration, Debnath and Nayak (2022) confirmed that droughtinduced temporary migration is an adaptation of small landholders and poor households in India.
Despite a growing body of literature on the influence of climate change on migration, there is still a lack of consistent evidence on whether climate change significantly induces migration, especially at the household level.Cattaneo and Peri (2019) confirmed the significant gaps in our understanding of the complex relationship between climate change and migration.It is still unclear how climatic factors induce migration, particularly in rural-urban migration.Beine and Parsons (2015) found little or no direct effect of climate change on migration.However, Gray and Mueller (2012a) and Koubi et al. (2016) provided contradictory findings.Koubi et al. (2016) analyzed the separate impacts of slow-onset and sudden-onset types of natural disasters and people's perception of risk leading to their migration decisions.It was concluded that slow-onset events (droughts) reduce individual movements because people learn to adapt to these shocks over time.The complexity of the climate-migration nexus suggests the mediating role of possible channels.Falco, Donzelli, and Olper (2018) indicated that climatic factors indirectly drive migration decisions through their effects on agricultural productivity, rural livelihoods and food (in-) security.Several empirical studies stressed the importance of agricultural channels.Using a household dataset in Nepal, Arslan et al. (2021) highlighted the push impact of weather shocks on migration through weather-induced direct effects on farming income.Worsening climatic conditions with lower rainfall result in lower agricultural productivity, which, in turn, drives the migration of rural households.
The empirical evidence on climate-induced migration in Vietnam is inconclusive.Few studies have examined the direct effects of natural disasters on internal migration (Grörer and Zyleberberg 2016;Berlemann and Tran 2020).Grörer and Zyleberberg (2016) interpreted out-migration due to a major typhoon in 2016 -the Ketsana Typhoon-for income generation purposes from remittances received from migrant labor networks.Financial mechanisms can smooth consumption and income losses from ex-post typhoons.By contrast, other studies find indirect impacts in which natural disasters can induce displacement through possible channels such as income reduction (Nguyen 2020), livelihood stress (Dun 2011) or agriculture production (Luong et al. 2023).Koubi et al. (2016) reported mixed results for different types of natural hazards.Individual exposure to sudden-onset and short-term environmental events such as floods significantly pushed decision to migrate while long-term and slow-onset natural events such as droughts reduced rural-urban migration.Luong et al. (2023) emphasized the importance of agricultural production in interpreting mixed effects in the climate-change-migration nexus.Rainfall shocks caused people to stay because of high agricultural productivity in years with excess rainfall; in contrast, temperature shocks caused people to leave their rural towns because of low crop yields in dry years.

Conceptual framework
There are strong linkages between agricultural households, climate change, agriculture and poverty (Hertel and Rosch 2010;Falco, Donzelli, and Olper 2018).According to Morton (2007), Black et al. (2011) and Warren, Diaz, and Hurlbert (2016), there are a variety of research trends in the literature investigating the impacts of climate change on agriculture such as the performance of key smallholder crops, ecosystems used by smallholder farmers and socio-economic vulnerability.The framework proposed by Morton (2007) focuses on understanding the impact of climate change on smallholders and subsistence agriculture by considering the impacts of climate change in a comprehensive setting that covers the vulnerability of smallholder households to both climatic shocks and nonclimate-related stresses.This framework has been frequently used in studies on the impact of climaterelated changes (Black et al. 2011;Warren, Diaz, and Hurlbert 2016).In this study, we extend this conceptual framework to investigate the effects of climate variation on rural households that rely mainly on farming production, focusing on their vulnerability to drought captured by climate anomalies on rice crop productivity, expenditure, poverty and migration (Figure 1).

Data
This study used a dataset of 2,200 households from the Thailand Vietnam Social Economic Panel (TVSEP), which was funded by the German Research Foundation (DFG).The household survey was conducted as part of the Poverty Dynamics and Sustainable Development (2015-2024) research project (Nguyen et al. 2020b).Household demographics, geography, shocks, risks, expectations, subjective assessments and behavioral traits were all included in the TVSEP dataset.Vietnamese datasets were gathered by surveying rural households in three central coastal provinces: Ha Tinh, Thua Thien Hue and Dak Lak (Figure 2).More specifically, the study sites were Ha Tinh and Thua Thien Hue provinces, where typhoons and storms are common, and Dak Lak, located in the central highlands, where droughts are common (Chaudry and Ruysschaert 2007).A panel dataset was constructed for three years (i.e. 2010, 2013 and 2016).This study also used weather data to calculate the rainfall, temperature anomalies and other droughtrelated variables.The two most important climatic variables, rainfall and temperature, were obtained from nine land-based weather stations from Vietnam's National Center for Hydrometeorological Forecasting located in the three surveyed provinces.
Tables 1 and 2 briefly describe the dependent and independent variables used in the study.There are four dependent variables: rice productivity, consumption per capita, poverty and migration.To identify 'poverty', this study employs the official national poverty line established by the GSO for rural areas.Households are considered poverty-poor if their percapita consumption falls below the poverty line.From 2005 to 2010, the GSO's poverty lines for rural areas were set at 200,000 VND, from 2011 to 2015 at 400,000 VND, and from 2016 to 2020 at 700,000 VND per person per month.The migration variable was calculated as the ratio of the number of migrants who reported leaving the household and relocating to a different province divided by the nuclear size of the household.
This study used a dummy variable for household perceptions of drought.The dummy variable 'drought' indicates whether a household perceived their experiences of drought events in the year prior to the survey.Table 1 presents the descriptive statistics of self-reported claims for experiencing drought during the surveyed years.This study uses the TVSEP household questionnaire for the types of shocks or events experienced by the households surveyed over a three-year period.Specifically, the specific question asked the respondent in the questionnaire is 'When considering the time period between 5/2013 and 4/2016, has there been any event (drought, flood, typhoon …) causing a big problem (shock) affecting the household?'.The duration of the drought episodes varied among the surveyed years.For instance, 60% of the drought events in 2013 occurred in the current year and 22.27% of the drought incidents occurred in the previous year.Regarding the drought events in 2016, the one-year duration accounted for nearly 49%, and the remaining 50% of the drought outbreaks were short-term events with a duration of less than one year.Simultaneously, drought was captured using instrument variables that measured the temperature and precipitation anomalies in local districts.
Table 2 presents the explanatory and control variables at the household and district level.Several household characteristics, including household size, dependence ratio and location, were controlled.Regarding the characteristics of household heads, we used education level, gender, ethnicity and age.District-level factors indicate the infrastructure and geographical characteristics of the municipalities.Specifically, commune infrastructure consists of the distance to town and the distance to the provincial capital from the surveyed districts in kilometers, as well as the types of accessibility (i.e.paved roads with one or two lanes).Geographical characteristics were used as dummy variables to indicate whether the commune landscape consisted of coastlines, mountains, or plains.
Table 2 provides the summary statistics for the outcome variables (household rice productivity, per capita consumption, poverty status and migration), explanatory variables (drought) and control variables.As shown in Table 2, the average rice productivity was approximately 377.6 kg/1000 m 2 .Regarding household welfare measures, annual per capita consumption is relatively low, averaging approximately 40 million VND (equivalent to $1,700 USD in annual living expenses as of the 2021 average exchange rate), and 60% of the surveyed households live in poverty.The analysis also revealed that this group of households was susceptible to drought, as approximately 20 percent had been affected in the previous years.In addition, one of every five household members lived away from their families, indicating a significant out-migration rate.The majority of the study sample consisted of middle-aged male Kinh household leaders.They average seven years of schooling, which is lower than the Vietnamese average of 8.2 years (Chaudry and Ruysschaert 2007).The dependency ratio was fairly high at roughly 1.5 or 2.0.Regarding district-level control variables, the majority of districts are situated in close proximity to a town, but are distant from the provincial capital.Seventy percent of the districts have paved roads with one or two lanes, suggesting that the infrastructure in most of the study region is solid.In terms of geographical features, more than 50 percent of the examined districts' landscapes were comprised of mountains and plains.The descriptive statistics of the IVs are presented in Table 2, and scatter graphs of the weather data for climatic variability and anomalies are provided in Appendix 4.   (2) .Drought-poverty nexus + g 1 edu ijt + g 2 ethnic ijt + g 3 sex ijt + g 4 age ijt + g 5 social tie ijt + g 6 dependent member ijt + g 7 household size ijt + w 1 distance town jt + w 2 distance centre jt + w 3 road jt + w 4 mountain jt + w 6 plain jt + w 8 coast jt + 1 ijt ) (3) where: G is a function that takes values strictly between zero and one: 0 , G(z) , 1, for all real numbers of zvalues.For the probit model, G is the standard normal cumulative standard normal distribution function (Wooldridge 2018).
. Drought-migration nexus There are other outcome variables: poverty ijt presents household i's status of being 'poor', and the ratio of number of migrants devided to household size is denoted by migration ijt .The empirical models also include several district-level control variables such as distance town jt and distance centre jt , which represent the distance from surveyed district j to town and that to provincial capital (i.e. the central area); road jtthe types of roads; and mountain jt , plain jt and coast jt ., which denote landscape features such as mountains, plains, and coastal lines, respectively.The error term is 1 ijt , and lastly, d t and d j are included in the fixedeffects estimation models, denoting time and communes fixed effects, respectively.

Instrumental variables
Following Sedova and Kalkuhl (2020), we constructed temperature and rainfall anomalies, which were subsequently used to address the endogeneity problem associated with self-reported drought information.We tested the validity of the selected instrumental variables, and the results are presented in Appendix 3.This studyollowed the Sedova and Kalkuhl (2020) formulas to measure weather anomalies based on Equations ( 5) and ( 6) andhen used them as instrumental variables for 'drought'.
Climate negative anomalies j, t−3 = Climate positive anomalies j, t−3 where: t is the surveyed year (i.e.t = 2010, 2013, or 2016).x i represents the number of months with climate-positive anomalies, whereas y i denotes the number of months with climate-negative anomalies compared to the long-run climate level.Climate positive anomalies j, t−3 , represent the negative anomalies of temperature or rainfall in district j for each lagged year (t-3).Climate level j, t−3 is the level of temperature or rainfall in district j in a given month of the lagged year (t-3) .m LR (Climate level j ) represents the long-run average temperature or rainfall of district j in a given month from 1974 to the lagged year (t-3) .s LR (Climate level j ) is the long-run standard deviation of the average temperature or rainfall of district j in a given month from 1974 to the lagged year (t-3).
Equations ( 5) and ( 6) were used to construct climatepositive anomalies in district j.It is constructed by the sum of the district's positive weather deviation (i.e.rainfall or temperature) in a given month of the lagged year (t-3) from its long-run mean m LR , divided by the district's month-specific standard deviation s LR .The climate negative anomalies in district j are constructed by the sum of the absolute value of the district's negative weather deviation (i.e.rainfall or temperature) in a given month of the lagged year (t-3) from its long-run mean m LR , divided by the district's month-specific standard deviation s LR .

Drought-induced impacts on rice productivity
The estimated results of the impact of drought on rice productivity are presented in Table 3.The coefficients for all specifications are significantly negative, suggesting that drought exposure reduced rice productivity.According to the fixed-effects estimate, if a household has experienced drought incidence in previous years, rice productivity will be reduced by approximately 8.6%.At the 1% significance level, the fixedand random-effects with IVs estimations offer evidence of the detrimental effects of droughts on productivity.These findings support the large body of research showing that agricultural production is inherently vulnerable to climate-related events such as drought.
Regarding the control variables, being in 'Kinh' ethnic groupthe dominant ethnic group in Vietnamis associated with better cropping, as measured by a 13.6% increase in rice productivity in the random-effects IVs model.Additionally, the geographical distance from the town is statistically significant in all specifications, indicating that rural villages located further from towns or urban areas have a higher productivity for rice crops.In contrast, all three specifications show negative and significant coefficients for 'road-type' variable.This result implies that households in communes with paved or easily Table 3. Estimation results of drought-induced impact on rice productivity. (1) (  accessible roads are likely to produce less rice than those in communes with dirt roads.

Drought-induced impacts on expenditure
Table 4 shows the estimated effects of drought on expenditures.The fixedand random-effects IVs estimations show a highly significant and negative effect of drought on household welfare as measured by consumption per capita.Household exposure to drought reduced consumption per capita by approximately 12.4% in the fixed-effects model.This finding is consistent with those of previous studies (e.g.Bui, Nguyen, and Pham 2014;Arouri, Nguyen, and Adel 2015) which found that extreme events such as drought caused a significant decrease in household welfare, whether measured in terms of income or consumption.
In addition, the higher education level of the household head significantly increases expenditure per capita.All estimations also show that male and aged household heads consume more than their female and younger counterparts.By contrast, the dependent ratio and household size significantly reduced per capita consumption.For example, research estimations show that if a household had one or more economically dependent members, per capita consumption would have decreased significantly, by approximately 13% to 15%.In addition, well-being is lower in local communities that are more rural and farther away from cities or towns.As shown in the fixed-effects model, distance from urban areas (i.e. the nearest town) had a negative robust effect on household consumption, which was significant at the 1% level.

Drought-induced impacts on poverty
The subsequent analysis investigates whether the negative impact on consumption pushes the affected  households into poverty.Table 5 shows the marginal effects of regressing the estimation results of Equation (3).The number of households exposed to drought in previous years increased the likelihood of poverty by 18,9% at the significance level of 1%.This finding is consistent with that of Arouri, Nguyen, and Adel (2015), who discovered that living in flood-prone areas in Vietnam increases the chance of falling into poverty.However, the magnitude of the impact of the weather events was much smaller, 1,8%.This conclusion is in line with empirical evidence from other developing countries (Cater et al. 2007;Rodriguez-Oreggia et al. 2013).
Regarding the control variables, the level of education proved to eradicate poverty as it significantly reduced the likelihood of falling into poverty by 3.0%.Furthermore, the 'Kinh' ethnic group was less likely to be 'poor' than other ethnic minority groups in Vietnam.Households with more dependent members are likely to be poorer than those with fewer members.Larger household size is also harmful as it considerably increases the likelihood of poverty by 9%.The effects of commune characteristics on household poverty are comparable to those of the previous welfare measures of per capita consumption.Remote rural households are more likely to be in poor condition.The estimation results from the probit model also demonstrate that variables such as distance to town, distance to center and road infrastructure in the neighborhood significantly increase a household's likelihood of falling into poverty.Regarding communes' landscape features, households residing along coastlines had a lower chance of being 'poor' than those living in mountainous villages or flat land areas.

Drought-induced impacts on internal migration
Lastly, Table 6 depicts the causal relationship between drought and rural-urban migration.The IVs and random effects with IVs models clearly show that household exposure to drought in particular causes more household members to migrate.Specifically, the ratio of migrants to household size increases by approximately 52.8% and 80.9%, respectively, at the 1% significance level.This evidence suggests that drought acts as a push factor that increases internal migration.It is consistent with the findings of Gray and Mueller (2012a) and Koubi et al. (2016).The positive-signed nexus between drought and internal migration implies that Vietnamese rural households opt for an ultimate risk-taking strategy, allowing their members to migrate for economic purposes to cope with the consequences of natural shocks in welfare and productivity.
Furthermore, the estimation results in Table 6 show that several household head characteristics significantly affect internal migration.Education level and 'Kinh' ethnic had positive effects on household migration in all specifications at the 1% level of significance.A male-headed household was less likely to have members who live away in urban areas than its female counterparts, which was proven significant (at the 5% significance level) in specifications ( 2) and (3).Furthermore, distance to town and distance to provincial capital significantly increased internal migration and the effects were highly significant in all specifications.Surprisingly, those whose hometowns are farther from a town or provincial capital are more likely to migrate.Thus, this suggests that out-migration is becoming more common in rural villages located farther away from urban areas (i.e.towns or capital).

Concluding remarks
This study aims to shed light on the relationship between extreme climatic events such as droughts and rural household welfare in Vietnam's central and highland provinces.Rural households are facing an increasing number of unforeseeable events that jeopardize their well-being and livelihoods.We attempted to simultaneously consider multiple climate-induced consequences in rural populations using a rich panel dataset, while controlling for potential issues that could lead to biased estimates.Specifically, this empirical study investigated the effects of drought on crop production, household expenditure, rural poverty and internal migration in the study area.
Overall, we find robust evidence that droughts reduce rice productivity in farming households.Moreover, the estimated results show that droughts not only caused a significant decline in the consumption of rural households, but also pushed them into poverty.The findings of this study confirmed a significant welfare loss if households experienced climatic shocks such as droughts.Furthermore, these findings confirm the relationship between climatic factors and migration.Droughts are significantly associated with the decision of rural people to migrate to urban areas, pushing their migration decisions through connections between productivity, consumption and poverty channels.As observed in the poor group, this empirical evidence suggests that migration is a key adaptation strategy to climatic shocks in Vietnam.Understanding the multifaceted relationship between climate change, particularly drought, and its consequences is critical for the development of policy interventions to help rural populations coping with risks and uncertainties.These findings highlight the importance of enhancing rural populations' resilience to extreme climate-related events in order mitigate their negative effects.However, due to data constraints (a relatively short-duration dataset with few study sites), we were unable to uncover the long-term interactions between rural household welfare and external factors such as droughts.Future studies could aid in understanding the dynamic nature of this relationship as more data becomes available.

Graph 3 :
Precipitation-positive anomalies Graph 4: Precipitation-negative anomalies (Source: Author's own calculation from weather data collected from hydrometeorological stations located in Ha Tinh, Thua Thien Hue, and Dak Lak provinces, in three surveyed years: 2010, 2013, and 2016).

Table 1 .
Number of shocks occurred at the surveyed districts.There is no information on the type of shocks in 2010 at the district level because of a lack of data availability in the village dataset for the 2010 wave.Drought-crop productivity nexusrice productivity ijt = a + b 1 drought ij(t−n) + g 1 edu ijt+ g 2 ethnic ijt + g 3 sex ijt + g 4 age ijt + g 5 social tie ijt with IVs and random-effects with IVs.This study also controlled for several covariates for household characteristics (e.g.household head education, age, ethnicity and household size), as well as commune-level factors (e.g.distance to the nearest town, infrastructure proxied by type of roads and geographical features: mountain, .+g 6 dependent member ijt + g 7 household size ijt + w 1 distance town jt + w 2 distance centre jt + w 3 road jt + w 4 mountain jt + w 6 plain jt + w 8 coast jt + d t + d j + 1 ijt(1)

Table 2 .
Descriptive statistics of the variables used in the study.Drought-expenditure nexusexpenditure ijt = a + b 1 drought ij(t−n) + g 1 edu ijt+ g 2 ethnic ijt + g 3 sex ijt + g 4 age ijt + g 5 social tie ijt + g 6 dependent member ijt + g 7 household size ijt + w 1 distance town jt + w 2 distance centre jt + w 3 road jt + w 4 mountain jt + w 6 plain jt + w 8 coast jt + d t + d j + 1 ijt (Source: Author's estimation from TVSEP panel data in the 2010, 2013, and 2016 waves).DEVELOPMENT STUDIES RESEARCH.
g 1 edu ijt + g 2 ethnic ijt + g 3 sex ijt + g 4 age ijt + g 5 social tie ijt + g 6 dependent member ijt + g 7 household size ijt + w 1 distance town jt + w 2 distance centre jt + w 3 road jt + w 4 mountain jt + w 6 plain jt + w 8 coast jt+ d t + d j + 1 ijt (4)where: rice productivity ijt denotes the logarithmic form of household i's land productivity for rice crops in district j at time t.As for the explanatory variable, drought ij(t−n) denotes households' exposure to drought in previous years (t-n) .Household-level control variables include household head education, edu ijt ; household head's ethnicity, ethnic ijt ; gender of household head, gender ijt ; age: age ijt members of social or political groups, social tie ijt ; dependent member ijt and household size ijt .

Table 4 .
Estimation results of the drought-induced impact on expenditure.

Table 5 .
Estimation results of the drought-induced impact on poverty.

Table 6 .
Estimation results of the drought-induced impact on internal migration.