the varying income effects of weather...
TRANSCRIPT
Policy Research Working Paper 7764
The Varying Income Effects of Weather Variation
Initial Insights from Rural Vietnam
Ulf Narloch
Development EconomicsEnvironment and Natural Resources Global Practice Group &Climate Change Cross-Cutting Solutions Area July 2016
Climate Change and Poverty in Vietnam
Background Paper
WPS7764P
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
ed
Produced by the Research Support Team
Abstract
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Policy Research Working Paper 7764
This paper is a product of the World Bank Environment and Natural Resources Global Practice Group and the Climate Change Cross-Cutting Solutions Area and is a background paper for the World Bank work on “Climate Change and Poverty in Vietnam.” It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at [email protected].
To estimate the impact of weather on rural income changes over time, this study combines data from the panel sub-sample of the latest Vietnam Household Living Standard Surveys 2010, 2012, and 2014 and gridded weather data from the Climate Research Unit. The analyses show: (i) crop cultivation, livestock management, forestry and fishing activities, and agricultural wages remain important income sources in rural Vietnam—especially for poorer households; (ii) rural communes are exposed to substantial inter- and intra-annual weather variation, as measured by annual, seasonal, abnormal, and extreme weather conditions and weather events; and (iii) these types of weather variation are indeed related to income variation. In particular, warmer temperatures and heat extremes can have negative income effects in all climate contexts and for all socioeconomic
groups and most income activities. Only staple crops, forestry, and fishing seem to be less sensitive to hotter con-ditions. The effects of rainfall are more difficult to generalize. Some findings indicate that more rainfall is beneficial in drier places but harmful in wetter places. Interestingly, the incomes of poorer households seem to be negatively affected by wetter conditions, while those of wealthier households are more impacted by drier conditions. An increase in rainfall levels and flood conditions between 2012 and 2014, which were relatively wet years, is related to reduced income growth between these two years. Altogether these findings suggests that greater attention has to be paid to making rural livelihoods more resilient to weather variation which, is very likely to increase because of climate change.
TheVaryingIncomeEffectsofWeatherVariation:InitialInsightsfromRuralVietnam*
Ulf Narloch
Sustainable Development Practice Group, World Bank, Washington, DC, USA
Keywords: Climate Change, Consumption, Households, Incomes, Livelihoods, Poverty, Shocks,
Vulnerability
JEL: I30, I32, O10, Q10, Q54, R20
* Acknowledgements: This work is part of the programmatic AAA on Vietnam Climate Resilience and Green Growth(P148188) and was developed under the oversight of Christophe Crepin. It contributed to the global program onClimate Change and Poverty (P149919) under the oversight of Stephane Hallegatte. I am very thankful to the WorldBank Vietnam team for providing the Vietnam Household Living Standards Survey (VHLSS) data and to Linh HoangVu and Ha Thi Ngoc Tran for helping with data questions. Mook Bangalore prepared the weather and geo‐spatialdata. Maros Ivanic and Anne Zimmer contributed to earlier data preparations. Very helpful suggestions andcomments on earlier versions of this work were received from Tuan Anh Le, Diji Chandrasekharan Behr, GabrielDemombynes, Linh Hoang Vu, Christopher Jackson, Frederik Noack, Pamela McElwee, and Maurice Rawlins.
1.Introduction
Especially for poor people with climate‐sensitive livelihoods, weather variation can be linked to living
standards. Despite significant progress in poverty reduction, around 20 percent of Vietnam’s population
still lived in extreme poverty in 2010 – about 27 percent in rural areas (World Bank, 2012). At the same
time, Vietnam is also particularly sensitive to increasing climate hazards, including short‐lived natural
disasters and inter‐ and intra‐annual variation in weather conditions, both of which are likely to be
exacerbated by climate change.1 Such climate hazards can affect poor and other vulnerable people
through various channels, such as agriculture and ecosystems, natural disasters, and health (Hallegatte et
al., 2016).
Many of the poverty impacts will unfold through changes in household incomes, which are hardly
quantified. Existing studies for Vietnam investigate the effects of global warming on agricultural
production using crop and hydrological models (Gebretsadik et al., 2012; Van Hoang et al., 2014; Yu et al.,
2010). Other work estimates the macroeconomic cost of climate change in Vietnam through sectoral
impacts (Arndt et al., 2012; World Bank, 2010). Global‐level work, including Vietnam‐specific results,
estimates poverty impacts of generalized climate shocks using micro‐simulation techniques and
Computable General Equilibrium (CGE) modeling (Ahmed et al., 2009; Hertel et al., 2010; Rozenberg and
Hallegatte, 2015). Recent studies have shown the current income and welfare impacts of short‐lived
natural disasters or extreme events on Vietnamese households (Arouri et al., 2015; Bui et al., 2014;
Thomas et al., 2010). More work, however, is needed to understand household‐level income effects of
more subtle variation in rainfall and temperature conditions and gradual changes to this variation.
A growing number of studies in a variety of contexts investigates how observed weather conditions affect
economic outcomes, mainly in an attempt to quantify the potential economic impacts of future climate
change (Auffhammer et al., 2013; Dell et al., 2014).2 Yet a quantification of climate change impacts
remains challenging even when the current income effects of weather variation are understood. Firstly,
how weather conditions will alter due to climate change is highly uncertain and even existing Global
Circulation Models (GCMs) bring diverging forecasts for some regions and countries. Moreover, income
1 In line with existing literature, this paper defines weather variation to describe shorter‐run temporal variation (variation between and within years), while climate is used for longer‐term variability of conditions and changes decades (Dell et al., 2014). 2 Many of these studies are based on cross‐country analyses comparing country income and production and weather conditions at various points in time (e.g. Burke et al., 2015, 2011; Dell et al., 2012; Hsiang, 2010; Schlenker and Lobell, 2010). Another strand of the literature focuses on the impacts on agricultural profits mostly using US data at the county level (Deschênes and Greenstone, 2012, 2007; Fisher et al., 2012; Mendelsohn et al., 1994; Schlenker et al., 2006; Schlenker and Roberts, 2009). Other studies use weather variables to explain household income in order to test income effects on other variables (Feng et al., 2010; Hidalgo et al., 2010; Yang and Choi, 2007). Interesting new studies estimate current weather impacts on incomes and living conditions (Baez et al., 2015; Noack et al., 2015; Park et al., 2015). Yet results from these micro‐level studies have not yet been used to simulate income effects under future climate change.
3
changes in the future are subject to adaptive responses and socioeconomic changes, such as the shift to
less climate‐sensitive activities.
While it is hard to predict what happens in the future, it is even difficult to establish the links between
weather and incomes in the present. While some activities are negatively affected by wetter or warmer
rainfall conditions, others may benefits from it. And Income changes do not only depend on direct weather
impacts on output, but also on indirect impacts such as price or production adjustments. Decreases in
crop income through yield declines could be reduced or even offset through resulting increases in prices
and wages (Hertel et al., 2010; Jacoby et al., 2014). Different income activities also play various functions.
Some may actually serve as a coping strategy to compensate income shortfalls from other activities so
that incomes from these activities increase in times of adverse weather conditions (Noack et al., 2015). In
addition, weather effects depend on socioeconomic factors that determine the extent to which
households choose risker (but more profitable) activities and can manage negative impacts through
production adjustments (e.g. use more irrigation).
Another difficulty results from identifying the types of weather variation that are relevant for income
fluctuations. Many studies use temperature and rainfall levels defined as annual means (Burke et al.,
2011; Dell et al., 2009; Feng et al., 2010; Schlenker and Lobell, 2010). Yet weather impacts depend on
their timing and the seasonality of income activities, so that some studies measure seasonal weather
conditions (Hsiang, 2010; Mendelsohn et al., 1994; Welch et al., 2010; Yang and Choi, 2007). Moreover,
excess rain and heat waves can be as harmful as rainfall scarcity and cold spells. To capture this non‐
linearity, some studies break rainfall and temperature levels into different intervals or only measure them
above or below a certain threshold (Deryugina and Hsiang, 2014; Deschênes and Greenstone, 2011;
Schlenker and Lobell, 2010).3 To control for abnormal values, some studies measure the deviation from
the long‐term mean normalized for location‐specific variability (Baez et al., 2015; Hidalgo et al., 2010;
Noack et al., 2015). Weather extremes measured by minimum and maximum temperature and rainfall
levels matter too (Welch et al., 2010). Very importantly, all of these forms of weather variation can have
different impacts in different places depending on location‐specific climate conditions (Park et al., 2015).
In order to define recommendations for poverty eradication in face of increasing climate risks in Vietnam,
this paper attempts to disentangle the current impacts of weather variation on income changes. To better
understand the income effects of different types of weather variation, the analyses differentiate between
annual, seasonal, abnormal and extreme weather conditions and weather events related to rainfall and
temperature. They also investigate how weather impacts vary by socioeconomic group, climate zones and
income activities. By covering this breadth of impacts, this study adds to the existing literature showing
that income effects depend on the type of weather variation and different contexts.
These analyses can build on two novel methodological aspects. First, this study combines data from the
latest Vietnam Household Living Standard Surveys (VHLSS) 2010, 2012 and 2014 and gridded weather data
3 Related to this approach is the concept of degree days, which assumes a piecewise‐linear function in temperatures defines as the sum of degrees above a lower threshold and below an upper threshold intervals (Schlenker and Lobell, 2010; Schlenker and Roberts, 2009) . .
4
from the Climate Research Unit (CRU). Second, this work takes advantage of the panel structure of this
data set, which includes about half of the households in at least two of the three survey rounds.
Regression techniques based on the panel data set can estimate the weather impact on income changes
over time while reducing omitted variable bias. Despite the strength of these data and methods, a
quantification of future impacts subject to uncertain climatic, environmental and socioeconomic changes
is beyond reach.
The remainder of this paper is structured as follows: Section 2 explains the data and methods applied for
the analyses. Section 3 shows that rural households remain highly reliant on agriculture and other
ecosystem‐based activities. Section 4 demonstrates the extent of weather inter‐ and intra‐annual
variation communes are exposed to. Section 5 presents the estimated income effects of weather variation
from the various regression analyses. Section 6 concludes that in the face of climate change greater
attention needs to be paid to make rural livelihoods more resilient to weather variation.
2.Dataandmethods
Based on data from the Vietnam Household Living Standard Survey (VHLSS) collected in 2010, 2012, and
2014 with gridded weather data from the Climate Research Unit (CRU), this study fits a number of
regression models to explain income differences.
2.1Householdandcommunedata
Information on incomes and socioeconomic conditions is derived from the household and commune data
from the VHLSS 2010, 2012, and 2014. These surveys are conducted by the General Statistics Office (GSO)
with technical support from the World Bank in Vietnam. They are nationally representative and contain
detailed information on individuals, households and communes. In total ca 9,400 households nationwide
are included in each round with about half of the households in each round also being surveyed in the
previous round so that the data set includes a short‐term panel.
These analyses focus on rural households and communes leaving a data set of about 20,000 household
observations from ca. 2,250 communes. Each survey round covers about 6,600 and 6,700 rural
households. About 1,400 households were interviewed in all three rounds, 1,600 in 2010 and 2012, and
1,400 in 2012 and 2014.
The household surveys include a wide array of socioeconomic data. At the individual level these data cover
demographics, education, employment, health, and migration. At the household level the data comprise
information on income and expenditures, employment and self‐production, durables, assets, and
participation in government programs. Consumption estimates are based on per‐capita expenditure as
calculated by the World Bank and GOS to determine the national poverty line. Incomes are calculated
based on the raw data in line with classifications from the GSO (Section 3). All consumption and income
values are expressed in 2010 prices using data on the Consumer Price Index from the World Development
5
Indicators. Other variables controlling for household demographics and assets are constructed from the
data sets.
The commune surveys collect information about the occurrence of emergencies in the last 3 years listed
by type and month in addition to information on population, economic conditions, agriculture and land.
These data allow identifying communes exposed to weather events, such as storms, floods and droughts.
Although the data allow estimating nationally‐representative consumption and income estimates and
their changes over time, several shortcomings for the purpose of this study are to be noted. First, the
study relies on observations from only three years within a five year time period and only a subsample of
the included households is observed in several years. This short‐term panel does not allow to understand
time‐variant household factors or structural changes over longer time horizons. Second, little information
is available about labor allocation, production inputs and returns, which are important to explain income
differences over time. The lack of such information can limit the explanatory power of the regression
models. Yet many of these variables could be determined by weather conditions and thus create potential
endogeneity biases.
2.2Weatherdata
In addition to these self‐reported weather events in the commune surveys, this study uses weather data
from the CRU of the University of East Anglia to control for rainfall and temperature conditions. From the
global CRU TS3.21 data set, monthly time series of rainfall, minimum, mean and maximum temperature
from 1961 to 2014 is available at 0.5 x 0.5‐degree grid.4 This balanced panel of weather data was produced
using statistical interpolation based data from 4,000 individual weather stations (Harris et al. 2014).
These data are merged at the commune level to construct current and long‐term weather variables for
each month. For each commune, the household interview dates are specified and the current weather
variables are defined for the 12 months before that interview date (e.g. if the interview date was in May
2012, the current weather variables cover June 2011‐ May 2012). The long‐term means and standard
deviation of monthly rainfall, minimum, mean and maximum temperature are calculated for 30 years
before these 12 months (e.g. in the above example June 1981– May 2011). Based on these data several
variables measuring current weather conditions are constructed as explained in section 4.
Although gridded data set, such as CRU are commonly used in economic studies as they provide a
balanced panel that adjusts for missing data and spatial factors (e.g. elevation), they suffer from some
limitations for assessing weather variation at subnational level. Optimally such data would be measured
based on ground station data, which, however, is not readily available for all of Vietnam. The precision of
the CRU data at the subnational level depends on the interpolation method and the availability of station
data for the areas of interest (Dell et al., 2014). Although this is an important concern that deserves
further investigation, this is beyond the scope of this paper. Moreover, only monthly average rainfall and
temperature data are available from CRU. Given high intra‐annual variations, the distribution of
4 http://www.cru.uea.ac.uk/cru/data/hrg/
6
temperature and precipitation within each day could include important information to identify extreme
or unusual weather conditions (Deschênes and Greenstone, 2007; Fisher et al., 2012; Schlenker et al.,
2006; Schlenker and Lobell, 2010; Schlenker and Roberts, 2009).
2.3Regressionanalyses
In the literature there are two econometric approaches used to estimate the income effects of weather
variation (Dell et al., 2012; Deschênes and Greenstone, 2012, 2007; Fisher et al., 2012; Mendelsohn et al.,
1994): Differences between households and locations are estimated from a cross‐section of data observed
at one point in time following a Ricardian approach. Alternatively, changes over time are estimated from
panel data observing the same households or locations at various points in time.
The data set used for this study allows estimating the following regression model:
where Y denotes per‐capita income observed for individual i in commune j in year t (i.e. 2010, 2012, 2014).
W measures weather conditions n commune j in year t using five sets of weather variables as described
in section 4. β is the parameter of interest that indicates the income effects of weather variation. X is a
set of household‐specific controls that vary over time, such as education, labor and land endowments.5 T
measures time‐fixed effects to neutralize common trends over time. Z includes commune‐specific effects
that do not change over time. U is a random, idiosyncratic error term.
A particularity of this data set is that for some households information is available for two or three survey
years. These households form a Panel data set. Other households were only observed in one of the three
years. Using this cross‐section data as well and treating all observation as independent observations
provides a Pooled data set.
A main concern when fitting models to estimate weather impacts on economic outcomes is endogeneity
bias (dell et al., 2014). Reverse causation is unlikely to be a problem as weather conditions are
exogenously determined. Yet the model is likely to suffer from omitted variable bias caused by the
potential correlation of weather variables with other commune characteristics (e.g. long‐term climate,
geo‐graphical, agro‐ecological conditions) that determine living standards. To the extent that such
variables cannot be measured, the estimates of β will be biased.
To minimize this omitted variable bias, a fixed‐effects (FE) linear model is fitted by using a within‐
regression estimator based on the Panel data set (Panel FE).6 These analyses allow investigating the
determinants of income changes between 2010 and 2012 and 2014 due to differences in weather
5 The data sets allow to control for land and labor inputs of some income activities. These are, however, not included in the regression analyses due to potential endogeneity biases. 6 A within‐regression estimator is always consistent, but a General Least Squares estimator can be more efficient. The GLS –estimator based on a random‐effects model was also tested, but a Hausman test rejects that the GLS estimator is efficient.
7
conditions over time or other time‐variant household factors. The model regresses the household specific
difference of the income in one specific year from the mean in years with observations using the
difference of weather conditions in one year from their means in all observed years.7 By taking differences,
all time‐invariant commune and household‐fixed effects are taken out eliminating any omitted variable
bias caused by time‐invariant factors.
To take advantage of the full data set and test the robustness of the results, additional models are fitted
based on the Pooled data set, which estimate the determinants of income differences between
households. First an Ordinary Least Squares model (OLS) is fitted that controls for observable commune
characteristics in Z (Pooled OLS). These commune controls include long‐term rain and temperature levels
and variability calculated from the CRU data and other location‐specific factors (e.g. altitude, slope, soil
conditions, tree cover, distance from city and roads) calculated from other geo‐spatial data sets.8 Whereas
the set of commune controls reduces omitted variable bias caused by time‐invariant commune
characteristics, it is unlikely to eliminate it. To further reduce this bias, a fixed‐effects linear model is fitted
by including a dummy for each commune in Z (Pooled FE). This specification can control for commune‐
fixed effects, but not for unobservable household‐fixed effects.9 Where unobserved household‐factors
are correlated with weather deviation (e.g. cultural preferences or risk aversion), they can also lead to
omitted variable bias, which can only be reduced through the Panel FE model.
These regression models are estimated for various sub‐samples to disentangle differences between
socioeconomic groups and climate zones. To identify weather impacts on different income activities, the
regression models are also applied to the household sub‐sample participating in the respective income
activity.10 Each model is calculated separately for each of the five sets of weather variables specified in
section 4. Robust standard errors are estimated by clustering at the commune‐level in order to account
for spatial correlation. The natural logarithm of the outcome variables is used in the regression to
7 For households with observations for only two years, this approach corresponds to regressing the difference in incomes between the two years on the difference in weather conditions. 8 The advantage of using observed time‐invariant commune factors is that they allow estimating their impact on income differences, which is not possible in a fixed effects model using the Pooled Cross‐section or Panel data set. 9 The effects of other observable, time‐invariant commune factors cannot be estimated. 10 In the Panel FE model, only those households are included that have reported the respective income activity in both years. Although it would also be interesting to analyze how weather variability affects the choice to take‐up a new activity or to drop an existing activity, the limited number of households with such changes does not allow for a meaningful analysis. For the Pooled data set selection models were tested, as the factors that drive participation in an activity can be very different from those that determine income levels. Such selection biases can be corrected by a Heckman selection model using limited information maximum likelihood (LIML) or full information maximum likelihood (FIML) estimators (Heckman, 1979). Yet such estimators are not robust when the model is not correctly specified or subject to collinearity (Puhani, 2000). The levels of collinearity found in this data when estimating selection models for the different incomes exceed critical levels as defined in other work (Leung and Yu, 1996). This collinearity is possibly caused be the lack of sufficient exclusion restrictions, i.e. variables that determine income levels but not participation in the income activity. In the presence of such collinearity problems OLS for the sub‐samples without correction of election biases provide more robust estimates (Puhani, 2000).
8
normalize the skewed distribution of outcomes (i.e. many observations of low income levels and a few
observations of very high income levels).11
3.Ruralincomesandchanginglivingconditions
This section shows the composition of rural income portfolios across different socioeconomic groups and
regions, as well as changes in living conditions and the impediments to increasing prosperity – as
perceived by households.
3.1Incomeportfolios
As calculated based on the VHLSS data, the following income categories are of importance in rural
Vietnam:
1) crop income: the output value minus production costs including rice, other staple crops, industrial
crops12, fruit trees and crop by‐products;13
2) livestock income: the value of animals and animal products produced (for selling and self‐
consumption) minus production costs;
3) forestry income: the value of harvested trees and other forest products (such as firewood), hunted
animals, as well as incomes from , forest protection and management minus the production costs;
4) fishing income: the value of production and catch of fish and shrimp minus production costs;
5) wage income: any cash and in‐kind wages and salaries including from agricultural wage
employment and non‐agricultural wage‐employment in unskilled and skilled occupations;
6) business income: revenues from businesses outside agriculture, forestry, aquaculture minus cost;
7) transfer income include remittances, emergency assistance, insurance, donations, social benefits,
support in health care and education;
8) other income: including returns from investments, and earnings from weddings and funerals.
The data demonstrates that agriculture and other ecosystem‐based incomes remain an important income
source for rural households. Crop, livestock, forestry and fishing incomes and agricultural wages make ca
45 percent of incomes of all rural households (Figure 1). Whereas a larger share of rural households is
engaged in income activities related to crop cultivation, livestock, forestry or fishing, the incomes earned
from wage‐employment in skilled occupations and from businesses are much higher (Table A.1).
Interestingly, more than 90 percent of households across all groups receive some transfer income. Overall,
there is little difference in the income composition for different socioeconomic groups between 2010,
2012 and 2014.
11 Taking the natural logarithm also improves the explanatory power of the various models and brings ease in interpreting the coefficients as percentage change of the outcome variable. 12 In order of importance, these include peanut, coconut, coffee, soya beans, tea, sugarcane, cashew, pepper. 13 While revenues are indicated for each crop, costs are only given for all crop production. To calculate incomes for different crops, it is assumed that each crop’s cost share is equivalent to it share in total crop revenues.
9
For poorer households, the role of crop cultivation, livestock, forestry, fishing and agricultural wages is
much more important than for wealthier households. Altogether they make more than 60 percent of the
income of the poorest quintile and even more than 70 percent for ethnic minorities compared to less than
30 percent for the wealthiest households (Figure 1). Remarkably, about 60 percent of the poorest quintile
and 80 percent of ethnic minorities are engaged in forestry activities earning about 10 percent of their
income from these activities (Table A.1). For other groups these incomes are negligible. While agricultural
wages and non‐agricultural wages in unskilled occupations are more important for poorer quintiles,
wealthier quintiles receive a higher income share from skilled wage employment or business activities.
Figure 1: Incomesharesbysocioeconomicgroup,2010,2012and2014
Notes: Weighted average value of household income share by group: Q1‐Q5= Consumption quintiles based on weighted per
capita expenditure; All=All rural households; Min=Ethnic minorities; Fem=Female headed households
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014.
There is also considerable variation in the importance of different incomes across regions. For example,
incomes from crop cultivation, livestock, forestry and fishing amount to 60 percent in the North West, but
are below 30 percent in the South East (Figure 2). Generally, income portfolios in the North East and North
West are much more diverse with high a large percentage of households engaged in rice cultivation, other
crops, livestock and forestry (Table A.2). Maize, the second most important crop in Vietnam in terms of
cultivation area is planted by more than 50 percent of households in these regions. In the Red River Delta
and the South Central Coast, which have the highest income levels, non‐agricultural wage employment in
skilled activities is the most important income source (Table A.2). In the Central Highlands income from
industrialized crops – mostly coffee – makes about 30 percent of incomes. And in the Mekong River Delta
fishing incomes play an important role with more than 40 percent of households engaged in some
aquaculture activity (Table A.2).
10
Figure 2: Incomesharesbyregion,2010,2012and2014
Notes: Unweighted average of household income shares per region. Source: Author’s calculation based on VHLSS 2010, 2012 & 2014.
3.2Changesinlivingconditions
Although the average income portfolios have not changed much between 2010 and 2014, at the
household‐level changes in living conditions are observed. Using the Panel sub‐sample and comparing the
consumption quintiles for these households observed in at least two years reveals great mobility between
quintiles. In both time periods more than 50 percent of households have changed their consumption
quintile; 28 percent have moved up one or more consumption quintiles; and 26 percent have moved down
at least one quintile. These data show relative changes in living conditions compared to other households.
Figure 3. MobilitybetweenconsumptionquintilesforPanelhouseholds,2010‐12and2012‐14
Notes: N indicates the number of Panel households in the 2010 (2012) consumption quintiles. The percentage values indicate the share of these households in the 2012 (2014) consumption quintiles. Green color coding = moving‐up consumption quintiles. Red color coding = moving down consumption quintiles. Grey color coding = same consumption quintile. Q1‐Q5= Consumption quintiles based on weighted per capita expenditure. Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 data.
n n
639 62.4% 24.4% 9.7% 1.7% 1.4% 592 63.7% 22.8% 9.0% 3.5% 1.0%
593 21.8% 34.7% 27.5% 11.3% 4.7% 542 24.0% 33.9% 21.0% 15.9% 5.2%
590 8.3% 22.4% 33.2% 25.4% 10.3% 588 9.0% 25.0% 34.2% 19.6% 12.2%
636 3.0% 12.4% 22.5% 36.8% 24.4% 563 3.0% 12.1% 20.8% 37.3% 26.8%
625 0.5% 2.6% 10.9% 21.9% 62.7% 541 1.3% 4.8% 9.2% 22.4% 62.3%
2012
Q1
Q2
Q3
Q4
Q5
2014
Q1 Q2 Q3 Q4 Q5
2012
Q4Q1 Q2 Q3 Q5
Q2
Q3
Q4
Q5
2010
Q1
11
In the VHLSS survey households are asked whether their living conditions have changed in absolute terms
compared to five years ago. The majority of households indicate improved living conditions, while only
very few have experienced worse living conditions (Figure A.1). The share of households with worse or
the same living conditions is highest in the lowest quintiles. These households rank the reasons for not‐
improved living conditions indicating three reasons in order of importance.
Among the most important reasons, natural and income related factors rank high especially for poorer
households. On average natural events (including droughts, floods, pests and harvest failures affecting
production) and other natural factors, such a livestock epidemics and changes in land and water
conditions play a limited role, which is declining over time. Yet as shown in Figure 4, they are relatively
more important in the lowest quintile and for ethnic minorities. Low incomes and other income‐related
reasons, such as production costs and selling prices, rank among the most important reasons too.
Interestingly, other factors, such as consumption prices and illness rank highest among wealthier
households.
Figure 4. Mainreasonsfornotimprovedlivingconditionsreportedbyhouseholds,2010,2012and2014
Notes: Unweighted average value by household group: Q1‐Q5= Consumption quintiles based on weighted per capita expenditure;
All=All rural households; Min=Ethnic minorities; Fem=Female headed households
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 data.
While such data are highly subjective, they offer some insights into the perceived impediments to greater
prosperity. Exposure to natural factors and income related reasons seem to prevent households from
improving living conditions. As all these factors can be conditioned by weather conditions, the extent to
which the households are exposed to weather variation is evaluated next.
4.Weathervariationinruralcommunes
The communes within the different regions of Vietnam represent different climate zones with varying
rainfall and temperature conditions. Communes are categorized as dry versus wet and cold versus hot
12
zones based on the long‐term (30 years) annual mean of monthly rainfall (157mm) and mean temperature
(25.4°C) across the communes in the data set (Figure A2) in order to define Dry‐cold, Dry‐hot, Wet‐cold
and Wet‐hot climate zones. Long‐term rainfall and temperature conditions as well as their inter‐annual
variation greatly varies by zone (Figure A.3). Different climate zones mostly coincide with the different
regions (Table A.3). In what follows current weather variation is described by five sets of variables
measuring annual, seasonal, abnormal and extreme weather conditions and weather events.
4.1Annualweatherconditions
Variables measuring annual rainfall and temperature levels control for the effect of average annual
conditions and variation between years. The use of annual values is common in the literature exploring
linkages between weather and economic outcomes (Burke et al., 2011; Dell et al., 2009; Feng et al., 2010;
Schlenker and Lobell, 2010).
Based on the CRU data, annual values are calculated as the mean of the monthly rainfall and mean
temperature values in the 12 months prior to the survey. Accordingly, 2010 values are not limited to the
2010 calendar year, but measure conditions in the 2009/10 season.
Figure 5 Annualweatherconditionsinruralcommunesbyclimatezonein2010,2012and2014
a. Annual rain b. Annual temperature
Notes: Box plot shows the distribution of commune observations for each climate zone. The boxes illustrate the 25 to 75 percentile with the median value represented by the line in the box. The whiskers indicate the lowest and highest adjacent value with the points outside below or above that identifying outlier observations. Values are measured as the mean of monthly rainfall levels and mean temperature in the last 12 months. Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
Annual weather conditions vary between the zones and years. 2010 was the driest and warmest year with
considerable variation between and within zones (Figure 5). Although on average 2012 is the wettest year
in all zones, 2014 has a greater number of communes in the Wet‐cold and Wet‐hot zones which have
rainfall levels far above the zone’s average. This skewed distribution reflects the heavy rainfalls that
occurred in the Central Provinces at the end of 2013. These rainfalls resulted in severe flooding and
livelihood damages (UN, 2013).
010
020
03
004
00
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
mo
nth
ly r
ain
fall
in m
m
1520
25
30
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
mon
thly
tem
pera
ture
in C
13
4.2Seasonalweatherconditions
Variables measuring seasonal rainfall and temperature levels can measure the impact of intra‐annual
variation and their changes between years in order to test how income effects depend on the timing of
weather shocks. Many studies measure temperature and rainfall levels at various points in the production
year to reflect the seasonality of weather conditions and many income activities (Hsiang, 2010;
Mendelsohn et al., 1994; Welch et al., 2010; Yang and Choi, 2007).
Based on the long‐term intra‐annual weather patterns (Figure A.3), this study differentiates between
three seasons: January – April, which are drier and colder (S1), May – August with wet months and the
highest temperatures (S2) and September – December with wet months and lower temperatures (S3).
This division is also broadly in line with the growing cycle of some crops. For each of these periods in the
last 12 months prior to the survey date, the mean of monthly rainfall and mean temperature level is
calculated.
Seasonal conditions do not only vary between years but also zones. All survey years are affected by large
intra‐annual variation in rainfall and temperature conditions (Figure A.4). In all three years the seasonal
variation between zones and within zones is highest for temperature S1 and for rainfall in S3 (Figure 6). In
2010 dry season temperatures in S1 are higher than in the other years, while in 2012 and 2014 rainfall
levels in S3 are larger than 2010. The rainfall outliers in S3 in the Wet‐cold and Wet‐hot zones capture the
extensive rainfalls in November 2013 in the Central provinces. 14 Within 3 days some provinces
experienced up to 400‐973 mm of rain (UN, 2013).
14 The weather variables are constructed so as to reflect the weather conditions in the 12 months prior to the survey. For example, for a household interviewed in March 2012, the first month indicates January 2012, while the 12th months indicate December 2011. Accordingly, the December values in the 2014 graph show the rainfall in December 2013.
14
Figure 6 Seasonalweatherconditionsinruralcommunesbyclimatezonein2010,2012and2014
a. Seasonal rain S1 (Jan‐Apr) b. Seasonal temperature S1 (Jan‐Apr)
c. Seasonal rain S1 (May‐Aug) d. Seasonal temperature S2 (May‐Aug)
e. Seasonal rain S3 (Sep‐Dec) f. Seasonal temperature S3 (Sep‐Dec)
Notes: Box plot shows the distribution of commune observations for each climate zone. The boxes illustrate the 25 to 75
percentile with the median value represented by the line in the box. The whiskers indicate the lowest and highest adjacent value
with the points outside below or above that identifying outlier observations. Values are measured as the mean of rainfall levels
and mean temperature in the respective months.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
050
01,
000
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
mon
thly
rai
nfal
l in
mm
1525
35
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
mon
thly
tem
pera
ture
in C
050
01
,000
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
mon
thly
rai
nfa
ll in
mm
1525
35
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
mon
thly
tem
pera
ture
in C
05
001,
000
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
mon
thly
rai
nfal
l in
mm
1525
35
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
mon
thly
tem
pera
ture
in C
15
4.3Abnormalweatherconditions
Not only absolute rainfall and temperature levels matter, but also the extent to which these levels differ
from long‐term normal climate conditions. Unusually wet, dry, hot and cold conditions can all have
detrimental impacts, which depend on their timing and location. For example, more rain may be beneficial
in a dry month and dry locations, but harmful in wet months and wet locations. Existing studies define
unusual weather conditions as the deviation from the long‐term mean (Hidalgo et al., 2010; Baez et al.,
2015; Noack et al., 2015). These studies put the calculated deviation into the context of the location‐
specific variability by normalizing by the location’s long‐term standard deviation (Lobell et al., 2011).
Figure 7Abnormalweatherconditionsinruralcommunesbyclimatezonein2010,2012and2014
a. Wet months b. Hot months
c. Dry months d. Cold months
Notes: Histograms show the distribution of commune observations for each climate zone. Abnormal months are measured by the deviation of the current month’s value from the 30 years mean being greater than 1.5 standard deviations. Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
To measure the number of wet, dry, hot and cold months, this study calculates the current deviation from
the long‐term mean for each month (Figure A.5). Wet conditions are defined as positive deviation from
0.2
.4.6
.81
0.2
.4.6
.81
0.2
.4.6
.81
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
2010, dry_cold 2010, dry_hot 2010, wet_cold 2010, wet_hot
2012, dry_cold 2012, dry_hot 2012, wet_cold 2012, wet_hot
2014, dry_cold 2014, dry_hot 2014, wet_cold 2014, wet_hot
Den
sity
number of months
0.2
.4.6
.81
0.2
.4.6
.81
0.2
.4.6
.81
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
2010, dry_cold 2010, dry_hot 2010, wet_cold 2010, wet_hot
2012, dry_cold 2012, dry_hot 2012, wet_cold 2012, wet_hot
2014, dry_cold 2014, dry_hot 2014, wet_cold 2014, wet_hot
De
nsity
number of months
0.2
.4.6
.81
0.2
.4.6
.81
0.2
.4.6
.81
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
2010, dry_cold 2010, dry_hot 2010, wet_cold 2010, wet_hot
2012, dry_cold 2012, dry_hot 2012, wet_cold 2012, wet_hot
2014, dry_cold 2014, dry_hot 2014, wet_cold 2014, wet_hot
Den
sity
number of months
0.2
.4.6
.81
0.2
.4.6
.81
0.2
.4.6
.81
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
2010, dry_cold 2010, dry_hot 2010, wet_cold 2010, wet_hot
2012, dry_cold 2012, dry_hot 2012, wet_cold 2012, wet_hot
2014, dry_cold 2014, dry_hot 2014, wet_cold 2014, wet_hot
Den
sity
number of months
16
the long‐term rainfall level and dry conditions as negative deviation. Hot conditions are measured by the
month’s positive deviation of current maximum temperature from the long‐term mean of maximum
temperature and cold conditions by the month’s negative deviation of current minimum temperature
from the long‐term mean of minimum temperature. For all measures each month’s deviation is then
divided by the month’s long‐term standard deviation. Abnormal months are defined as months with a
deviation exceeding 1.5 standard deviations.
Abnormal weather conditions vary mostly by year (Figure 7). The limited variation between zones results
from accounting for some of the regional variation through the normalization by the commune’s standard
deviation. In all zones number of wet months is highest in 2014 and the number of dry months in 2010.
Hot months prevail in 2010, while the number of cold months is largest in 2014.
4.4Extremeweatherconditions
To control for extreme weather conditions maximum and minimum values in rainfall and temperature
levels can be measured. A limited period without any rain or with intensive rain can be as harmful to
livelihoods as a limited number of days with heat extremes or with frost conditions. For example, a recent
study for Vietnam defines excessive rainfall if it exceeds 300, 450 or 600mm within a 5 day period (Thomas
et al., 2010). Another study from Asia differentiates between daily minimum and maximum temperatures
to control for temperature variation (Welch et al., 2010).
Based on the monthly rainfall and temperature values (Figure A.4), extreme weather conditions are
measured as follows: the maximum of rain as the precipitation level in the wettest month, the minimum
of rain as measured by the precipitation level in the driest month, the maximum of temperature as the
maximum temperature in the hottest month, and the minimum of temperature as the minimum
temperature in the coldest month.
These extreme weather conditions broadly reflect earlier findings. Maximum rain is lowest in 2010 at
around 400mm and highest in 2012 exceeding 500mm in Dry‐hot and Wet‐cold zones, but with some
considerable variation within the Wet‐cold zone in 2014 (Figure 8). Minimum rain is below 50mm and is
close to zero in the Dry‐cold zone. Whereas this zone had the lowest annual and seasonal temperature
means (Figure 5 and 6), their maximum temperature is higher than in the Wet‐cold zone and their
minimum temperature is at the level of the Dry‐hot and Wet‐hot zone.
17
Figure 8Extremeweatherconditionsinruralcommunesbyclimatezonein2010,2012and2014
a. Maximum rain b. Maximum temperature
c. Minimum rain d. Minimum temperature
Notes: Box plot shows the distribution of commune observations for each climate zone. The boxes illustrate the 25 to 75 percentile with the median value represented by the line in the box. The whiskers indicate the lowest and highest adjacent value. Maximum rainfall is measured by the rainfall level in the wettest month. Minimum rain is measured by the rain level in the driest month. Maximum temperature is measured by the mean of maximum temperature in the hottest month. Minimum temperature is measured by the minimum temperature in the coldest month. Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
4.5Weatherevents
Variables describing the occurrence of weather events, such droughts, floods and storms allow to measure
the impact of weather shocks and natural disasters. A recent study for Vietnam uses monitored data on
riverine floods and cyclones to identify natural disasters and assess their welfare impacts (Thomas et al.,
2010). Notwithstanding the limitations of self‐reported data, which often suffers from subjective
judgements and is highly correlated with welfare outcomes, other studies from Vietnam use self‐reported
data to assess the welfare impacts (Arouri et al., 2015; Bui et al., 2014).
050
01,
000
1,50
0
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
rain
fall
in m
m
2530
3540
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
tem
pe
ratu
re in
C
050
100
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
rain
fall
in m
m
1520
2530
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014
dry_cold dry_hot wet_cold wet_hot
tem
per
atu
re in
C
18
Unfortunately, the monthly weather data from CRU do not allow to identify weather events, such as
floods, droughts, and storms, which often depend on extremes within a short time period. Instead self‐
reported data from the commune surveys are used, which include questions about the five main
emergencies in the last three years.15 Based on this data, those communes that have experienced floods,
droughts and storms within the 12 months before the survey are identified. Although the data also include
information on crop diseases, fires and epidemics, these events are not included as they are supposedly
less directly related to weather conditions.
Overall, more than a fifth of all rural households reports some weather event. The variation between
zones and years is lowest for the occurrence of storms (Figure 9). In line with earlier findings suggesting
that 2010 is the hottest and driest year, the self‐reported data indicates a higher occurrence of droughts
in 2010 – mostly in the Wet‐cold zone. Surprisingly, the share of communes reporting floods is highest in
2010 in the Dry‐hot and Wet‐cold zone, which is in contrast to the rather dry conditions in that year and
the severe flood events observed in the Wet‐cold zone in 2014. The self‐reported data does not
correspond to the flood events experienced by communes in. Auxiliary regressions, however, reveal that
self‐reported floods indeed are related to higher annual and seasonal rainfall levels, number of wet
months and maximum rainfall (Table A.4).
Figure 9 Weathereventsinruralcommunesbyclimatezonein2010,2012and2014
a. Floods b. Droughts c. Storms
Notes: Bars show the average share of communes in each climate zone being affected by the event based on self‐reported data. Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
15 These emergencies are listed by year and month, including crop diseases, storms, floods, droughts, fires and epidemics.
0.0
5.1
.15
.2.2
5
sha
re o
f co
mm
une
s a
ffe
cte
d
dry_cold dry_hot wet_cold wet_hot
201020
1220
1420
1020
1220
1420
1020
1220
1420
1020
1220
14
0.0
5.1
.15
.2.2
5
shar
e of
com
mun
es a
ffe
cte
d
dry_cold dry_hot wet_cold wet_hot
201020
1220
1420
1020
1220
1420
1020
1220
1420
1020
1220
14
0.0
5.1
.15
.2.2
5
sha
re o
f co
mm
unes
aff
ecte
d
dry_cold dry_hot wet_cold wet_hot
201020
1220
1420
1020
1220
1420
1020
1220
1420
1020
1220
14
19
5.Weathersensitivityofincomes
This section shows the impacts of the different sets of weather variables on incomes as estimated by the
regression model described in section 2.3.17 The emphasis is placed on the results from the Panel FE model as
this allows to estimate the effects on income changes over time eliminating any potential omitted variable bias
from time‐invariant factors. All regression models also include a set of control variables which are not the focus
of this discussion (Table A.5). The fitted models indicate the income effects of weather variation (5.1) and how
these weather impacts vary by socioeconomic group (5.2), climate zone (5.3), income activity (5.4) and severity
of conditions (5.5).
5.1Weathervariationrelatestoincomechanges
The results indicate that many weather variables have a significant impact on consumption and income
changes over time. The direction, order of magnitude and statistical significance of the results for consumption
and total incomes are almost identical for the different weather variables (Figure 10 and Table A.6). This finding
suggests that the impact of weather variation on incomes directly translates into consumption changes and
that there is limited smoothing of the consumption effects. Hence, the calculated weather effects on incomes
also provide an estimate of the overall welfare impacts.
The significant results for temperature variation indicate that hotter conditions relate to lower incomes. An
average annual temperature that is 1°C warmer decreases total income by 20 percent – mostly driven by a
10% reduction in in season S1 and S3 (Figure 10). Similarly, an additional hot month lowers income by 6
percent, while an additional cold month increases income by 4 percent. A 1°C increase in temperature in the
hottest month relates to 10 percent income reduction and a 1°C increase in the coldest month to a 5 percent
reduction.
The impacts of rainfall variation suggest that drier conditions can have negative impacts. An additional 100m
of rain in season S3 has an income‐increasing effect of 3 percent, while an additional dry month reduces income
by 6 percent. All other variables are insignificant possibly because wetter conditions can also have negative
impacts when rainfalls become too intensive. Accordingly, having experienced a drought event relates to a 9
percent and a flood event to a 5 percent income reduction. The income effects of rainfall variation are further
disentangled in the next analyses.
As would be expected weather variation is related to some but not all income differences between years. As
per the R2 statistics the models with the measured weather variables explain about 8 to 9 percent of the
variation in total income, whereas the explanatory power of the models with the self‐reported weather events
17 Other specifications of weather conditions are also tested, which do not offer much additional insights. For example, including the quadratic form of annual or seasonal rainfall and temperature levels does yield similar results and makes the interpretation of the results from the Panel regressions difficult. Similarly, using deviations from the long‐term mean instead of absolute rainfall and temperature levels produces almost identical results in the Panel regressions, because differences from the mean are already captured in the Panel FE model.
20
is at only 3 percent (Table A.6). This finding suggests that measured weather data are better suited than self‐
reported data to income welfare changes over time.
Figure 10. Estimatedconsumptionandincomeeffectsofchangeinweatherconditions,2010,2012,and2014
Notes: Figures show the coefficient estimated from the Panel Subsample Fixed Effects regression models (1) –(5). ‘Con’ indicates per‐
capita expenditure and ‘Inc’ per‐capita total income. Whiskers indicate the 95% confidence intervals, and a solid marker is statistically
significant at 90% or higher. The detailed results tables with standard errors are presented in Tables A6.
Source: Author calculation based on based on VHLSS 2010, 2012 & 2014 and CRU data.
To compare the weather impacts on income changes over time with those on income differences between
households, further models are fitted using the Pooled cross‐section data set. Overall, the explanatory power
of the Pooled OLS and Pooled FE model is much higher than in the Panel FE (Table A.7). This result is not
surprising as in the Panel FE the changes in income over time can only be explained by time‐variant factors.
Besides the different sets of weather variables, the data does not allow to sufficiently capture such time‐variant
factors. The estimated income effects are smaller and less significant in the models using the Pooled data,
suggesting that current weather variation can better explain income changes over time than income
differences between households. However, long‐term climate conditions, for example, as measured by the 30
years standard deviation of rainfall and temperature explain differences in living standards between
households (Narloch & Bangalore, 2015).
The difference in results between the Panel FE model capturing all time‐invariant household and commune
factors (Table A.6), the Pooled FE capturing only all time‐invariant commune factors and the Pooled OLS
capturing only observable time‐invariant commune factors (Table A.7) demonstrate that the estimated effects
depend on the extent to which conditioning factors are controlled for. Some considerable differences in the
statistical significance and the magnitude of the estimated effects appear between the Pooled OLS and the
Pooled FE model. Including commune fixed effects instead of observable commune characteristics increases
the explanatory power of the models remarkably (Table A.7). This finding implies that a large extent of the
income variation is due to unobserved differences in commune characteristics. When not including commune
fixed effects, the weather variables may actually capture some of the effects of these unobserved commune
characteristics so that they may be highly biased. This conclusion cautions some of the findings from other
21
work in this field that cannot control for unobservable factors that are highly correlated with weather variables
(Noack et al., 2015; Park et al., 2015).
5.2Weatherimpactsvarybysocioeconomicgroup
To show how income effects vary across different socioeconomic groups, the Panel FE model is estimated for
different groups, including the two lowest (B40) vs three highest (T60) expenditure quintile (Table A.8),
households that moved‐down at least one consumption quintile between years (Down) versus thus that moved
up (Up) as identified in section 3.2 (Table A.9), as well as minority (Min) and female‐headed (Fem) households
(Table A.10). Some interesting differences between these groups appear (Figure 11).
Figure 11. Estimatedincomeeffectsofchangeinweatherconditionsbysocioeconomicgroup,2010,2012,and2014
Notes: Figures show the coefficient estimated from the Panel Subsample Fixed Effects regression models (1) –(5). ‘B40’ indicates
households in the lowest two per‐capita expenditure quintiles in 2012, ‘T60’ indicates households in the upper three per‐capita
expenditure quintiles in 2012, ‘Down’ indicates households that moved down at least one consumption quintile between years, ‘Up’
indicates households that moved up at least one consumption quintile between years, ‘Min’ indicates households from an ethnic
minority, and ‘Fem’ indicates female‐headed households. Whiskers indicate the 95% confidence intervals, and a solid marker is
statistically significant at 90% or higher. The detailed results tables with standard errors are presented in Tables A8‐10.
Source: Author’s calculation based on based on VHLSS 2010, 2012 & 2014 and CRU data.
Rainfall variation can have different impacts on B40 and T60 households. While an additional 100m of monthly
rain reduces the incomes of the B40 by more than 10 percent, it increases the incomes of the Top 60 by ca 10
percent. Similarly, a significant negative impact is estimated for wet months and floods on B40 households and
for dry months and droughts for 60 households. This finding suggest that the livelihoods of poorer households
suffer from extensive rainfalls whereas wealthier households are more negatively affected by lack of rain.
The differences in rainfall effects are less pronounced for other groups. An additional 100m of monthly rain
has a large positive impact for households that moved‐up, but no significant income effects for households
that moved‐down. Wet months have a positive income effect for those households that moved‐up, but a
negative one for those that moved‐down, while dry months have negative impacts for both groups. Female
22
headed households suffer from negative income effects of dry moths, as well as drought and flood events. The
incomes of ethnic minorities are not very sensitive to rainfall variation – possibly because they mostly live the
Northern areas, which are subject to drier conditions and less rainfall variation.
Hotter temperature conditions have negative income effects for all socioeconomic groups. An increase of
average annual temperatures by 1°C relates to an income reduction of about 12 percent for ethnic minorities,
18 percent of B40 households, 22 percent of female‐headed households and 24 percent of T60 households.
Accordingly, more hot months have negative income effects and more cold months have positive income
effects for almost all groups.
5.3Weatherimpactsvarybyclimatezones
To show how weather impacts vary across different climate contexts, their income effects are estimated for
the different zones as identified in section 4 differentiating between Dry‐cold, Dry‐hot (Table A.11) Wet‐cold
and Wet‐hot (Table A.12) zones. As would be expected some differences in the weather impacts between these
climate zones can be observed (Figure 12).
Figure 12. Estimatedincomeeffectsofchangeinweathervariationbyclimatezonesin2010,2012,and2014
Notes: Figures show the coefficient estimated from the Panel Subsample Fixed Effects regression models (1) –(5). ‘Dry‐cold’, ‘dry‐hot’,
‘wet‐cold’, and ‘wet‐hot’ indicate the climate zones as identified in Section 4. Whiskers indicate the 95% confidence intervals, and a
solid marker is statistically significant at 90% or higher. The detailed results tables with standard errors are presented in Tables A11‐
A12.
Source: Author’s calculation based on based on VHLSS 2010, 2012 & 2014 and CRU data.
Some findings indicate that wetter conditions could have positive income effects in drier locations, but
negative ones in wetter locations. However, these effects depend very much on the timing of the weather
impact. For example, more rain in the season S1 is related to large negative income effects in the dry zones,
possibly as soils and livelihoods are not prepared for large quantities of rains in the dry season. More rain in
season S2, however, has a positive impact in dry zones. In the wet zones, the income effect of rain in S2 is
negative implying negative impacts of any additional rainfall in the rainy season in wet locations. Accordingly,
wet months have a positive impact in Dry‐cold zones, but negative one in the Wet‐hot zones. Also floods are
23
related to positive income changes in the Dry‐cold zone, but negative changes in the Wet‐cold zone. Dry
months have a negative income effect in dry zones but no impact in wet zones.
Warmer weather conditions have negative impacts in all contexts with one exception. In the Wet‐cold zones a
temperature increase in season S2 is related to a positive income effect of about 20 percent. Negative income
effects are related to warmer temperatures in season S1 for the wet places and in season 3 for cold places.
There is also some indication that higher temperatures have more severe impacts in hotter locations. An
increase of 1°C in average annual temperature reduces incomes by 15‐17 percent in Wet‐cold and Dry‐cold
zones and by 27‐28 percent in Wet‐hot and Dry‐hot zones.
5.4Weatherimpactsdependonincomeactivity
These estimated income effects are triggered by the weather impacts on different income activities, which is
shown by the results for rice cultivation, staple crops, industrial crops, livestock, forestry, fishing, agricultural
wages, unskilled non‐agricultural wages, skilled agricultural wages and business self‐employment (Tables A.13‐
A.17). The individual income effects can be much larger indicating that a big effect on one activity can be
compensated through other incomes so that the overall income and expenditure effect is more limited (Figure
13).
Figure 13. Estimatedincomeeffectsofchangeinweathervariationbyincomeactivityin2010,2012,and2014
Notes: Figures show the coefficient estimated from the Panel Subsample Fixed Effects regression models (1) –(5). The activity
abbreviations indicate the following activities as defined in section 3: ‘Inc’= total income, ‘Ric’ = rice cultivation, ‘Sta’ = staple crop, ’Ind’
= industrial crops, ‘Liv’ = livestock, ‘For‘ = Forestry, ‘Fis’ = fishing, ‘Wag’ = agricultural wage employment, ‘Wun’ = unskilled non‐
agricultural wage employment, ‘Wsk’ = skilled non‐agricultural wage employment, and ‘Bus’ = business self‐employment. Whiskers
indicate the 95% confidence intervals, and a solid marker is statistically significant at 90% or higher. The detailed results tables with
standard errors are presented in Tables A6, and A13‐A17.
Source: Author’s calculation based on based on VHLSS 2010, 2012 & 2014 and CRU data.
24
Warmer weather conditions have a negative effect on most income activities. Three exceptions are to be
noted. First, the weather impacts for staple crops are not significant or even positive for maximum
temperatures. This finding indicates that these crops – mostly grown in the colder regions – are less sensitive
to warmer temperatures or could even benefit. Second, hotter conditions and even drought events are related
to higher forestry incomes. This finding may indicate that the extraction of forest resources is a copying
mechanism applied to compensate shortfalls from other activities under heat stress. Similarly, fishing incomes
go up under heat extremes expressed by the number of abnormally hot months and the maximum
temperature in the hottest months. These results corresponds to other studies from rural Vietnam showing
that incomes from the extraction of environmental resources tend to go up during weather shocks (Völker and
Waibel, 2010).
Interestingly, weather variables do not only relate to agricultural and other ecosystem‐production related
incomes, but also to supposedly less sensitive activities, such as non‐agricultural wages and business activities.
This result may imply that in rural areas there are strong inter‐sectoral (demand and supply) linkages through
which all income activities are negatively affected when weather shocks hit the agricultural sector. Moreover,
some weather extremes may also affect non‐agricultural activities through their impacts on labor availability
and productivity. Interestingly, all wage‐related activities are negatively affected by higher temperature, but
are less sensitive to rainfall variation. This finding is in line with Park et al. (2015) who find that heat extremes
can have a negative impact on labor productivity.
Some note of caution is needed. The predictive power of the models –especially those with the self‐reported
weather events – become weak for some income activities. This finding mainly applies for income activities
undertaken by a small sub‐sample of households and with limited income variation over time, such as forestry
and fishing. Possibly these activities are less weather sensitive, so that the weather variables combined with
the set of households controls cannot sufficiently explain income variation across time.
This finding may also be due to the dependency of effects on the specific locations within climate zonesand on
specific activities within each income category. To further test this dependency, additional regressions were
run for each income activity at the regional level. The results for the regional models with the highest predictive
power are summarized in Tables A.18‐A.23. For example, in the Mekong River Delta less rain in the dry season
S1 reduces income from summer‐autumn rice, while less rain during season S3 increases income from winter‐
spring rice. Unusually dry months are harmful for incomes from both winter‐spring and summer‐autumn rice
(Tables A.18‐A.19). Coffee incomes in the Central Highlands suffer from high temperatures in S1 and
temperature and rainfall extremes, as well as flood and drought events (Table A.20). In the North West region
forestry incomes decrease in drier years with more cold months, while fishing incomes increase in warmer
years with more hot months (Table A.21). Moreover, weather variation can explain a larger extent of the
variation in wage and business‐related incomes for some regions (Tables A.22‐A.23). These findings suggest
that a further disaggregation by income activities and regions would allow to refine the results.
25
5.5Weatherimpactsdependontheseverityofconditions
The above analyses indicate that more rainfall does not have much of a negative impact on changes in income.
This finding may be surprising in light of the severe livelihood impacts of the flood events caused by the intense
rainfalls in late 2013 in the Central provinces. These results may mainly capture the negative income effects of
too dry conditions in 2010. At the same time, 2012 and 2014 were both relatively wet years so that focusing
on these two years instead of the whole 2010‐14 sample may allow disentangling some of the effects of too
intensive rainfalls. To do so, the above analyses are rerun splitting the Panel sample into a 2010‐12 and a 2012‐
14 sub‐sample (Table A.24). Some very interesting differences between the two time periods appear (Figure
A.14).
Figure 14. Estimatedincomeeffectsofchangeinweathervariationbyperiods,2010‐12versus2012‐14
Notes: Figures show the coefficient estimated from the Panel Subsample Fixed Effects regression models (1) –(5). 2010‐12 is based on
the 2010 and 2012 Panel households, 2012‐14 is based on the 2012 and 2014 Panel households. Whiskers indicate the 95% confidence
intervals, and a solid marker is statistically significant at 90% or higher. The detailed results tables with standard errors are presented
in Table A.24.
Source: Author’s calculation based on based on VHLSS 2010, 2012 & 2014 and CRU data.
Indeed wetter conditions are related to positive income changes between 2010 and 2012, but negative
changes between 2012 and 2014. Having an increase in the annual average of monthly rainfall by 100mm
boosts income by about 20 percent between 2010 and 2012, but reduces income by about 12 percent between
2012 and 2014. An increase in rainfall by 100mm between 2012 and 2014 in the season S3, when most of the
flooding took place has an income reducing effect of 5 percent. Having the same rainfall increase in the
following dry season S1 reduces incomes even by up to 40 percent. Similarly, wet months and floods have a
negative income effect between 2012 and 2014, whereas dry months and droughts have a negative income
effect between 2010 and 2012. For temperature variation the differences between the 2010‐12 and 2012‐14
time period are less pronounced and the results are statistically weaker than in the combined 2010‐14 sample.
These finding suggest that the impacts of weather variation on income changes over time depend on the
severity of conditions in the time periods under consideration. Especially for short‐term panels, such as this
data set, results are sensitive to the weather conditions in the years with observations. While very limited
26
variation between years can hide some of the actual weather impacts, extreme conditions in one of the few
survey years can bias some of the results. To mediate some of these shortcomings, panel data sets that cover
larger time horizons are needed.
5.Conclusions
The results in this paper show that weather variation and income changes over time in rural Vietnam are
related. They also warn against an oversimplification of this relationship as there is a breadth of weather
impacts, which are highly dependent on socioeconomic groups, climate zones, and income activities, and even
the severity of weather conditions. In addition, income effects do not only depend on the intensity of the
rainfall or temperature shock, but also on the timing of the shock and the location‐specific optimal rainfall and
temperature levels.
Despite this complexity the data allow identifying some general patterns. Warmer temperatures and heat
extremes have income‐reducing effects in all climate contexts and for all socioeconomic groups, including
poorer households and ethnic minorities. While most income activities are negatively affected by hotter
conditions, staple crops, forestry and fishing seem to be less sensitive to temperatures. The income effects of
rainfall are more complex. Some findings indicate that more rainfall is beneficial in drier places but harmful in
wetter places. Interestingly, the incomes of poorer households seem to be negatively affected by wetter
conditions, while those of wealthier households are more impacted by drier conditions. This finding implies
that the livelihoods of poor rural people are more vulnerable to severe rainfalls and flooding, which for
example occurred at the end of 2013. The data indeed confirm that an increase in rainfall and wet months
compared to 2012, which was also a wet year, had a negative impact on income growth between 2012 and
2014.
Bringing the variety of income effects from different types of weather variation together with the uncertainty
about how weather conditions will be altered by climate change makes any effort to quantify the income and
poverty impacts of climate change extremely challenging – not even considering that the weather sensitivity
of income activities may change due to structural changes or adaptive responses. While future temperature
increases due to climate change can be predicted with some level of confidence, there is less agreement on
future changes in precipitation patterns. Yet variation between locations and between seasons is likely to
increase (IMHEN and UNDP, 2015; ISPONRE, 2009; MONRE, 2009). Overall, however, it remains difficult to
translate the global climate change scenario from the IPCC (2014) into localized impacts. While some locations
could benefit from more favorable conditions, the overall variability of weather conditions is expected to
increase with abnormal or extreme conditions likely to become more frequent and intense.
Notwithstanding the difficulties to quantify any future climate change impacts on rural incomes, the initial
insights provided by this paper bring important implications for rural development in times of climate change.
The findings demonstrate that high weather variation between years, seasons and locations is already the
norm in rural Vietnam and that the incomes of most people including poor households are currently very
sensitive to this variation. Consequently, rural households are already vulnerable to weather variation. Climate
change could increase these vulnerabilities in many unpredictable ways. Hence it is important to make rural
27
livelihoods more resilient to weather variation by promoting income strategies that are more robust to
uncertain weather conditions (e.g. weather‐resistant crops and production practices) and by enabling
households to be better prepared for weather shocks (e.g. weather‐focused information systems) or to cope
with them ex‐post (e.g. weather‐proofed safety nets). All in all, policies and investments should pay greater
attention not only to short‐lived disasters, but also to more subtle weather variation and gradual changes in
weather conditions.
28
References
Ahmed, S.A., Diffenbaugh, N.S., Hertel, T.W., 2009. Climate volatility deepens poverty vulnerability in developing countries. Environ. Res. Lett. 4, 034004.
Arndt, C., Tarp, F., Thurlow, J., 2012. The Economic Costs of Climate Change: A Multi‐Sector Impact Assessment for Vietnam.
Arouri, M., Nguyen, C., Youssef, A.B., 2015. Natural Disasters, Household Welfare, and Resilience: Evidence from Rural Vietnam. World Dev. 70, 59–77.
Auffhammer, M., Hsiang, S.M., Schlenker, W., Sobel, A., 2013. Using weather data and climate model output in economic analyses of climate change. Rev. Environ. Econ. Policy ret016.
Baez, J.E., Lucchetti, L., Genoni, M.E., Salazar, M., 2015. Gone with the Storm: Rainfall Shocks and Household Well‐Being in Guatemala. World Bank Policy Res. Work. Pap.
Bui, A.T., Dungey, M., Nguyen, C.V., Pham, T.P., 2014. The impact of natural disasters on household income, expenditure, poverty and inequality: evidence from Vietnam. Appl. Econ. 46, 1751–1766.
Burke, M., Dykema, J., Lobell, D., Miguel, E., Satyanath, S., 2011. Incorporating climate uncertainty into estimates of climate change impacts, with applications to US and African agriculture. National Bureau of Economic Research.
Burke, M., Hsiang, S.M., Miguel, E., 2015. Global non‐linear effect of temperature on economic production. Nature advance online publication.
Dell, M., Jones, B.F., Olken, B.A., 2009. Temperature and Income: Reconciling New Cross‐Sectional and Panel Estimates. Am. Econ. Rev. 99, 198.
Dell, M., Jones, B.F., Olken, B.A., 2012. Temperature shocks and economic growth: Evidence from the last half century. Am. Econ. J. Macroecon. 66–95.
Dell, M., Jones, B.F., Olken, B.A., 2014. What Do We Learn from the Weather? The New Climate–Economy Literature. J. Econ. Lit. 52, 740–798.
Deryugina, T., Hsiang, S.M., 2014. Does the environment still matter? Daily temperature and income in the United States. National Bureau of Economic Research.
Deschênes, O., Greenstone, M., 2007. The Economic Impacts of Climate Change: Evidence from Agricultural Output and Random Fluctuations in Weather. Am. Econ. Rev. 97, 354–385.
Deschênes, O., Greenstone, M., 2011. Climate Change, Mortality, and Adaptation: Evidence from Annual Fluctuations in Weather in the US. Am. Econ. J. Appl. Econ. 152–185.
Deschênes, O., Greenstone, M., 2012. The economic impacts of climate change: evidence from agricultural output and random fluctuations in weather: reply. Am. Econ. Rev. 102, 3761–3773.
Feng, S., Krueger, A.B., Oppenheimer, M., 2010. Linkages among climate change, crop yields and Mexico–US cross‐border migration. Proc. Natl. Acad. Sci. 107, 14257–14262.
Fisher, A.C., Hanemann, W.M., Roberts, M.J., Schlenker, W., 2012. The economic impacts of climate change: evidence from agricultural output and random fluctuations in weather: comment. Am. Econ. Rev. 3749–3760.
Gebretsadik, Y., Fant, C., Strzepek, K., 2012. Impact of climate change on irrigation, crops and hydropower in Vietnam. WIDER Working Paper.
Hertel, T.W., Burke, M.B., Lobell, D.B., 2010. The poverty implications of climate‐induced crop yield changes by 2030. Glob. Environ. Change, 20th Anniversary Special Issue 20, 577–585.
Hidalgo, F.D., Naidu, S., Nichter, S., Richardson, N., 2010. Economic determinants of land invasions. Rev. Econ. Stat. 92, 505–523.
Hsiang, S.M., 2010. Temperatures and cyclones strongly associated with economic production in the Caribbean and Central America. Proc. Natl. Acad. Sci. 107, 15367–15372.
IMHEN, UNDP, 2015. Summary for Policy Makers, Viet Nam Special Report on Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation [Trần Thục, Koos Neefjes, Tạ Thị Thanh
29
Hương, Nguyễn Văn Thắng, Mai Trọng Nhuận, Lê Anh Tuấn, Lê Đình Thành, Huỳnh Thị Lan Hương, Võ Thanh Sơn, Nguyễn Thị Hiền Thuận]. Natural Resources and Environment Publishing House, Hanoi,.
ISPONRE, 2009. Viet Nam Assessment Report on Climate Change. Institute of Strategy and Policy on Natural Resources and the Environment, Hanoi, Vietnam.
Jacoby, H.G., Rabassa, M., Skoufias, E., 2014. Distributional Implications of Climate Change in Rural India: A General Equilibrium Approach. Am. J. Agric. Econ. aau084.
Lobell, D.B., Schlenker, W., Costa‐Roberts, J., 2011. Climate trends and global crop production since 1980. Science 333, 616–620.
Mendelsohn, R., Nordhaus, W.D., Shaw, D., 1994. The Impact of Global Warming on Agriculture: A Ricardian Analysis. Am. Econ. Rev. 84, 753–771.
MONRE, 2009. Climate change, sea level rise scenarios for Vietnam. MONRE, Hanoi, Vietnam. Noack, F., Wunder, S., Angelsen, A., Boerner, Jan, 2015. Responses to Weather and Climate: A Cross‐Section
Analysis of Rural Incomes. World Bank Policy Research Working Paper 7478. Park, J., Hallegatte, S., Bangalore, M., Sandhoefner, E., 2015. Households and heat stress: Estimating the
Distributional Consequences of climate change. World Bank Policy Research Working Paper 7479. Rozenberg, J., Hallegatte, S., 2015. The impacts of climate change on poverty in 2030, and the potential from
rapid, inclusive and climate‐smart development. Forthcoming as a World Bank Policy Research Working Paper.
Schlenker, W., Hanemann, W.M., Fisher, A.C., 2006. The impact of global warming on US agriculture: an econometric analysis of optimal growing conditions. Rev. Econ. Stat. 88, 113–125.
Schlenker, W., Lobell, D.B., 2010. Robust negative impacts of climate change on African agriculture. Environ. Res. Lett. 5, 014010.
Schlenker, W., Roberts, M.J., 2009. Nonlinear temperature effects indicate severe damages to US crop yields under climate change. Proc. Natl. Acad. Sci. 106, 15594–15598.
Thomas, T., Christiaensen, L., Do, Q.‐T., Trung, L.D., 2010. Natural disasters and household welfare: evidence from Vietnam. World Bank Policy Res. Work. Pap. Ser. Vol.
Van Hoang, T., Chou, T.Y., Basso, B., Yeh, M.L., Chien, C.Y., 2014. Climate Change Impact on Agricultural Productivity and Environment Influence based on Simulation Model. Int. J. Adv. Remote Sens. GIS 3, pp. 642–659.
Völker, M., Waibel, H., 2010. Do rural households extract more forest products in times of crisis? Evidence from the mountainous uplands of Vietnam. For. Policy Econ. 12, 407–414.
Welch, J.R., Vincent, J.R., Auffhammer, M., Moya, P.F., Dobermann, A., Dawe, D., 2010. Rice yields in tropical/subtropical Asia exhibit large but opposing sensitivities to minimum and maximum temperatures. Proc. Natl. Acad. Sci. 107, 14562–14567.
World Bank, 2010. Vietnam ‐ Economics of adaptation to climate change. World Bank, Washington, DC. World Bank, 2012. Well Begun, Not Yet Done: Vietnam’s Remarkable Progress on Poverty Reduction and the
Emerging Challenges, 2012 Vietnam Poverty Assessment. World Bank, Hanoi. Yang, D., Choi, H., 2007. Are remittances insurance? Evidence from rainfall shocks in the Philippines. World
Bank Econ. Rev. 21, 219–248. Yu, B., Zhu, T., Breisinger, C., Hai, N.M., 2010. Impacts of climate change on agriculture and policy options for
adaptation. International Food Policy Research Institute (IFPRI).
30
Appendix
Figure A.1 Changeinlivingconditionscomparedtofiveyearsagobysocioeconomicgroupin2010,2012and2014
Notes: Unweighted average for households in different groups: Q1‐Q5= Consumption quintiles based on weighted per capita expenditure; All=All rural households; Min=Ethnic minorities; Fem=Female headed households. Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 data.
31
Figure A.2 Relationship between long‐termmean ofmonthly rainfall andmean temperature in ruralcommunes
Notes: Data points present rural communes included in the VHLSS 2010, 2012 and 2014. Lines indicate the mean value.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 & CRU data.
Figure A.3 Inter‐annualdistributionof long‐term rainfalland temperature levels in rural communes bymonthandclimatezone30 years mean of monthly rainfall 30 years mean of temperature
Notes: Boxplots show the distribution of commune observations for each zone. The boxes illustrate the 25 to 75 percentile with the
median value represented by the line in the box. The whiskers indicate the lowest and highest adjacent value.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 & CRU data.
5010
015
020
025
03
0 ye
ars
me
an o
f m
ont
hly
ra
infa
ll in
mm
15 20 25 3030 years mean of monthly temperature in C
050
01,
000
_1 _2 _3 _4 _5 _6 _7 _8 _9_10_11_12 _1 _2 _3 _4 _5 _6 _7 _8 _9_10_11_12 _1 _2 _3 _4 _5 _6 _7 _8 _9_10_11_12 _1 _2 _3 _4 _5 _6 _7 _8 _9_10_11_12
dry_cold dry_hot wet_cold wet_hot
mon
thly
ra
infa
ll in
mm
1020
3040
_1 _2 _3 _4 _5 _6 _7 _8 _9_10_11_12 _1 _2 _3 _4 _5 _6 _7 _8 _9_10_11_12 _1 _2 _3 _4 _5 _6 _7 _8 _9_10_11_12 _1 _2 _3 _4 _5 _6 _7 _8 _9_10_11_12
dry_cold dry_hot wet_cold wet_hot
mon
thly
te
mp
era
ture
in C
32
Figure A.4 Monthlyweatherconditionsinruralcommunesbymonthandregion,2010,2012and2014
Notes: Box plot represent the distribution of commune observations for each climate zone. The boxes illustrate the 25 to 75 percentile with the median value represented by the line in the box. The whiskers indicate the lowest and highest adjacent value. Source: Author’s calculation based on VHLSS, 2010, 2012 & 2014 and CRU data.
a. Monthly rainfall levels
b. Monthly mean temperature levels
050
01,
000
1,50
02,
000
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014_1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12
dry_cold dry_hot wet_cold wet_hot
mon
thly
ra
infa
ll in
mm
Graphs by zone
1020
3040
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014_1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12
dry_cold dry_hot wet_cold wet_hot
mon
thly
tem
pera
ture
in C
Graphs by zone
33
Figure A.5 Deviationofmonthlyweatherconditionsfromlong‐termmeaninruralcommunesbymonthandclimatezone,2010,2012and2014a. Monthly rainfall deviation (positive deviation =wetter, negative deviation = drier than normal)
b. Monthly temperature deviation (positive deviation =warmer, negative deviation = colder than normal)
-6-4
-20
24
6
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014_1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12
dry_cold dry_hot wet_cold wet_hot
dev
iatio
n in
un
its o
f lo
ng-t
erm
sd
Graphs by zone
-6-4
-20
24
6
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014_1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12
dry_cold dry_hot wet_cold wet_hot
dev
iatio
n in
un
its o
f lo
ng-t
erm
sd
Graphs by zone
34
c. Monthly deviation of max temperature (positive deviation = warmer than normal)
d. Monthly deviation of min temperature (positive deviation = colder than normal)
Notes: Box plot represent the distribution of commune observations for each climate zone. The boxes illustrate the 25 to 75 percentile with the median value represented by the line in the box. The whiskers indicate the lowest and highest adjacent value. Source: Author’s calculation based on VHLSS, 2010, 2012 & 2014 and CRU data.
-6-4
-20
24
6
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014_1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12
dry_cold dry_hot wet_cold wet_hotd
evia
tion
in u
nits
of l
ong
-ter
m s
d
Graphs by zone
-6-4
-20
24
6
2010 2012 2014 2010 2012 2014 2010 2012 2014 2010 2012 2014_1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12 _1_2_3_4_5_6_7_8_9_10_11_12
dry_cold dry_hot wet_cold wet_hot
dev
iatio
n in
un
its o
f lo
ng-t
erm
sd
Graphs by zone
35
Table A.1 Shareofhouseholds(in%)engagedinincomeactivitiesandaverageper‐capitaincomevalues(in'000VDNin2010prices)bysocioeconomicgroupin2010,2012and2014
Q1 Q2 Q3 Q4 Q5 All Minority Female
2010
All crop 87% 1,698 86% 2,372 80% 2,733 78% 3,683 64% 3,729 79% 2,843 94% 2,714 63% 1,973
Rice 73% 855 68% 1,149 62% 1,402 58% 1,497 42% 1,455 61% 1,272 80% 1,112 44% 963
Industrial crops 37% 212 38% 529 35% 571 31% 997 23% 1,330 33% 728 45% 560 24% 508
Other staple crops 71% 537 56% 509 49% 495 47% 704 32% 479 51% 544 83% 858 35% 297
Livestock 71% 406 68% 571 61% 712 58% 1,077 46% 1,176 61% 788 82% 680 46% 541
Forestry 62% 365 33% 216 19% 116 14% 99 7% 76 27% 175 81% 553 18% 82
Fishing 31% 203 31% 354 28% 690 24% 395 17% 806 26% 489 37% 167 19% 352
Wage 62% 1,964 70% 3,339 66% 4,079 66% 5,281 59% 7,040 65% 4,340 58% 2,074 69% 5,925
Agricultural wage 31% 734 24% 764 17% 644 11% 485 5% 324 18% 590 27% 649 22% 810
Non‐ag wage ‐ unskilled
30% 621 31% 958 26% 941 23% 813 14% 742 25% 815 24% 442 26% 1,016
Non‐ag wage ‐ skilled 19% 609 36% 1,617 42% 2,494 48% 3,983 50% 5,973 39% 2,934 19% 983 40% 4,099
Business 18% 380 28% 986 37% 1,860 41% 3,288 48% 7,655 34% 2,832 19% 431 31% 2,486
Transfers 92% 558 94% 799 93% 1,198 92% 1,526 91% 3,138 92% 1,443 91% 603 92% 2,502
Other 20% 76 23% 179 22% 203 26% 368 34% 1,353 25% 435 24% 172 26% 601
Total 5,650 8,816 11,591 15,718 24,971 13,346 7,393 14,462
2012
All crop 87% 1,677 82% 2,341 81% 3,357 78% 3,823 62% 5,059 78% 3,251 96% 2,854 66% 2,225
Rice 71% 751 65% 1,208 61% 1,537 58% 1,524 40% 1,320 59% 1,268 80% 1,112 45% 935
Industrial crops 33% 280 31% 532 31% 1,094 32% 1,231 23% 2,635 30% 1,154 41% 767 23% 777
Other staple crops 67% 506 50% 396 47% 475 42% 656 30% 690 47% 545 83% 802 33% 318
Livestock 68% 439 61% 641 56% 791 55% 1,308 42% 1,503 56% 936 81% 695 43% 781
Forestry 60% 378 29% 196 21% 152 13% 113 7% 169 26% 202 78% 581 18% 105
Fishing 30% 126 28% 356 23% 460 22% 536 13% 886 23% 473 32% 125 15% 205
Wage 65% 2,196 71% 4,364 66% 5,249 65% 6,597 59% 8,646 65% 5,410 61% 2,357 68% 6,507
Agricultural wage 28% 751 25% 950 16% 797 11% 614 6% 496 17% 721 25% 696 20% 1,043
Non‐ag wage ‐ unskilled
32% 659 30% 972 21% 895 17% 868 10% 643 22% 807 28% 531 23% 1,018
Non‐ag wage ‐ skilled 21% 787 39% 2,443 45% 3,558 51% 5,115 52% 7,507 42% 3,881 22% 1,129 42% 4,446
Business 17% 383 28% 1,178 31% 2,043 40% 3,578 44% 7,351 32% 2,906 18% 560 27% 2,516
Transfers 94% 713 92% 959 93% 1,262 92% 1,690 91% 2,697 92% 1,464 94% 709 95% 2,182
Other 23% 113 27% 199 27% 275 27% 499 38% 1,130 28% 443 22% 149 31% 507
Total 6,027 10,234 13,589 18,145 27,440 15,084 8,029 15,029
2014
All crop 89% 1,792 85% 2,424 79% 2,886 74% 3,419 62% 4,634 78% 3,031 95% 2,754 66% 2,055
Rice 72% 759 67% 1,215 60% 1,317 53% 1,378 37% 1,343 58% 1,202 78% 1,038 44% 983
Industrial crops 35% 380 34% 524 30% 801 27% 863 24% 2,040 30% 921 42% 742 23% 624
Other staple crops 73% 523 55% 456 46% 438 39% 575 29% 580 49% 514 84% 775 36% 246
Livestock 71% 491 62% 824 59% 951 53% 1,425 43% 1,579 58% 1,054 79% 734 44% 606
Forestry 62% 391 29% 168 20% 134 12% 90 9% 129 26% 182 78% 548 19% 79
Fishing 29% 197 27% 466 25% 545 20% 697 15% 977 23% 576 32% 114 14% 331
Wage 67% 2,595 70% 4,942 70% 6,552 67% 7,497 62% 9,441 67% 6,205 63% 3,092 68% 6,983
Agricultural wage 33% 926 21% 856 16% 874 9% 525 5% 331 17% 702 29% 856 17% 775
Non‐ag wage ‐ unskilled 30%
718 30%
1,200 26%
1,248 22%
1,242 13%
680 24%
1,018 26%
657 25%
1,192
Non‐ag wage ‐ skilled 22% 951 40% 2,886 48% 4,430 52% 5,730 54% 8,430 43% 4,485 22% 1,579 42% 5,017
Business 16% 362 26% 1,392 35% 2,765 39% 3,928 44% 7,999 32% 3,289 17% 633 30% 3,327
Transfers 91% 758 91% 1,115 93% 1,470 92% 1,781 91% 2,844 92% 1,593 89% 716 94% 2,775
Other 27% 124 28% 180 24% 233 29% 430 39% 1,421 29% 478 25% 204 35% 621
Total 6,710 11,511 15,537 19,267 29,025 16,408 8,795 16,779
Notes: Weighted average calculated by group. Q1‐Q5= Consumption quintiles based on weighted per capita expenditure; All=All rural households;
Minority=Ethnic minorities; Female=Female headed households.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014.
36
Table A.2 Shareofhouseholds(in%)engagedinincomeactivitiesandaverageper‐capitaincomevalues(in'000VDNin2010prices)byregionin2010,2012,and2014
North East North West Red River North C Coast South C Coast C Highlands South East Mekong
2010
All crops 91% 2,402 93% 2,489 86% 2,190 85% 2,017 48% 1,895 88% 5,803 77% 4,182 64% 3,412
Rice 79% 1,163 76% 964 77% 1,436 73% 1,103 14% 960 39% 491 68% 380 37% 2,267
Industrial crops 47% 259 34% 232 23% 89 40% 408 27% 314 60% 4,134 27% 2,778 22% 328
Other staple crops 85% 715 91% 1,157 54% 428 72% 348 17% 466 62% 955 44% 558 19% 371
Livestock 87% 1,083 85% 729 62% 948 77% 783 30% 853 59% 690 60% 606 39% 546
Forestry 66% 533 89% 599 1% 20 44% 319 9% 158 36% 150 31% 64 15% 73
Fishing 27% 158 39% 197 18% 206 21% 157 6% 451 21% 98 11% 396 47% 1,356
Wage 52% 2,990 53% 2,383 62% 4,878 56% 2,752 70% 4,574 69% 3,093 66% 8,378 61% 3,873
Agricultural wage 5% 140 14% 294 4% 93 13% 397 27% 602 42% 1,140 19% 1,580 26% 858
Non‐ag wage ‐ unskilled 26% 721 27% 528 23% 866 22% 604 21% 810 21% 372 25% 1,452 25% 879
Non‐ag wage ‐ skilled 28% 2,129 23% 1,562 48% 3,919 33% 1,751 41% 3,162 21% 1,581 40% 5,346 28% 2,135
Business 29% 1,536 25% 597 37% 3,822 29% 1,476 31% 2,790 25% 1,334 36% 4,287 35% 3,173
Transfers 88% 1,270 91% 701 95% 2,394 93% 2,053 96% 1,697 96% 710 93% 2,390 92% 1,973
Other 21% 204 46% 216 30% 470 18% 200 13% 341 15% 274 30% 573 30% 766
Total 10,178 7,912 14,929 9,757 12,759 12,151 20,876 15,170
2012
All crops 92% 2,377 96% 2,620 81% 2,156 87% 2,059 48% 2,168 88% 7,922 78% 6,251 62% 3,600
Rice 79% 1,107 75% 905 72% 1,409 75% 1,041 14% 916 40% 507 69% 494 35% 2,334
Industrial crops 42% 313 29% 401 14% 90 39% 524 27% 472 60% 5,923 26% 4,738 23% 349
Other staple crops 83% 691 92% 1,216 47% 411 73% 292 13% 518 60% 1,281 43% 538 15% 374
Livestock 86% 1,354 85% 809 53% 1,052 75% 1,105 26% 706 56% 565 50% 838 38% 558
Forestry 68% 591 89% 674 1% 137 41% 270 8% 387 32% 147 34% 30 13% 38
Fishing 22% 233 34% 183 16% 269 21% 228 5% 468 21% 55 7% 279 44% 1,464
Wage 57% 3,633 57% 2,414 58% 5,974 55% 3,300 70% 5,621 60% 3,091 66% 9,715 60% 4,377
Agricultural wage 7% 271 13% 240 3% 117 12% 479 26% 939 39% 1,240 22% 1,879 24% 1,007
Non‐ag wage ‐ unskilled 27% 868 33% 593 17% 808 19% 636 16% 832 15% 388 24% 1,073 25% 949
Non‐ag wage ‐ skilled 31% 2,493 23% 1,581 49% 5,050 34% 2,186 46% 3,850 18% 1,463 42% 6,762 28% 2,422
Business 27% 1,670 23% 691 33% 4,012 30% 2,024 25% 2,842 22% 1,582 32% 3,386 31% 3,111
Transfers 88% 1,361 97% 1,042 96% 2,902 91% 1,976 97% 1,683 94% 840 93% 1,812 87% 1,974
Other 26% 238 20% 158 30% 576 30% 418 22% 493 11% 143 25% 683 39% 722
Total 11,456 8,590 17,079 11,381 14,369 14,344 22,994 15,844
2014
All crops 13% 2,617 12% 2,457 10% 2,190 10% 1,828 7% 2,031 6% 7,846 3% 4,551 13% 3,943
Rice 4% 1,107 4% 778 1% 1,366 4% 921 2% 821 32% 437 10% 497 2% 2,359
Industrial crops 10% 420 14% 485 3% 95 3% 425 4% 392 8% 6,378 3% 2,805 2% 422
Other staple crops 2% 747 1% 1,019 2% 502 1% 320 1% 571 1% 739 2% 557 3% 321
Livestock 7% 1,517 11% 1,096 0% 1,119 4% 1,105 3% 1,073 2% 649 0% 1,056 0% 862
Forestry 2% 583 2% 606 2% 4 2% 301 2% 329 1% 136 1% 63 8% 43
Fishing 3% 238 5% 235 1% 346 5% 246 8% 813 15% 58 13% 447 9% 1,614
Wage 8% 4,216 7% 3,611 7% 7,089 6% 4,307 9% 6,324 2% 3,805 7% 9,504 9% 5,323
Agricultural wage 16% 249 14% 337 29% 69 19% 549 22% 943 10% 1,215 26% 1,853 15% 1,040
Non‐ag wage ‐ unskilled 10% 847 6% 555 17% 1,190 13% 709 14% 1,226 8% 273 15% 1,306 14% 1,277
Non‐ag wage ‐ skilled 11% 3,120 10% 2,720 19% 5,830 20% 3,049 16% 4,154 8% 2,317 11% 6,345 14% 3,006
Business 1% 2,004 1% 1,400 3% 3,967 3% 2,537 4% 2,846 2% 2,097 3% 4,551 4% 3,391
Transfers 1,385 612 3,141 2,442 2,014 827 1,761 2,424
Other 234 122 552 361 585 309 612 983
Total 12,793 10,138 18,408 13,127 16,015 15,727 22,544 18,583
Notes: Unweighted averages calculated by region.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014.
37
Table A.3 Shareofcommunesindifferentclimatezonesbyregion
North East North West
Red River Delta
North Central Coast
South Central Coast
Central High‐lands
South‐East South
Mekong River Delta
Total
Dry & Cold 96.4% 98.6% 100% 5.7% 1.5% 26.7% 0.2% 0% 41.2%
Dry & Hot 3.6% 0% 0% 0% 34.7% 9.2% 23.8% 3.3% 8.4%
Wet & Cold 0% 1.4% 0% 65.7% 28.9% 62.2% 6.7% 0% 14.6%
Wet & Hot 0% 0% 0% 28.6% 34.9% 1.9% 69.2% 96.7% 35.9%
Notes: Share calculated as unweighted average.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
38
Table A.4 Regressionresultsexplainingself‐reportedweathereventsbymeasuredweatherconditions
Flood Drought Storm
(1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4)
Annual rain 0.0132*** 0.00736*** 0.00803***
(0.00123) (0.00166) (0.00104) Annual temperature ‐0.153*** ‐0.315*** ‐0.0222
(0.0295) (0.0393) (0.0221) Seasonal rain S1 0.0142*** 0.00860*** 0.00491***
(0.00159) (0.00187) (0.00138) Seasonal rain S2 ‐0.00294*** 0.00250* ‐0.00129
(0.00106) (0.00134) (0.000888) Seasonal rain S3 0.00204*** 0.00137* 0.00154***
(0.000501) (0.000744) (0.000417) Seasonal temperature S1 0.243*** ‐0.0960 0.421***
(0.0630) (0.0762) (0.0513) Seasonal temperature S2 0.180*** ‐0.0597 0.105**
(0.0671) (0.0797) (0.0513) Seasonal temperature S3 ‐0.513*** ‐0.106 ‐0.649***
(0.104) (0.124) (0.0831) Wet months 0.699*** 0.284*** 0.317***
(0.0646) (0.0795) (0.0517)
Dry months 0.389*** 0.164* 0.255***
(0.0864) (0.0995) (0.0710)
Hot months ‐0.197*** ‐0.0288 ‐0.0193
(0.0610) (0.0631) (0.0502)
Cold months 0.0849 ‐0.319*** ‐0.109**
(0.0540) (0.0868) (0.0459)
Max rain 0.000591*** 0.00106*** 0.00111***
(0.000189) (0.000252) (0.000162)
Minimum rain 0.0268*** ‐0.0000298 ‐0.00670*
(0.00443) (0.00609) (0.00404)
Max temperature ‐0.0580 ‐0.280*** ‐0.218***
(0.0422) (0.0551) (0.0387)
Min temperature ‐0.0991*** ‐0.0726*** 0.0516***
(0.0133) (0.0183) (0.0119)
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Cons ‐1.077 ‐1.924 ‐3.974*** 0.0461 3.808*** 2.128 ‐3.625*** 6.931*** ‐3.264*** 0.210 ‐3.175*** 3.335***
(0.750) (1.329) (0.205) (1.334) (0.936) (1.535) (0.220) (1.695) (0.575) (0.991) (0.161) (1.186)
N 9235 9235 9235 9235 9235 9235 9235 9235 9235 9235 9235 9235
Notes: Table present coefficients estimated by mixed effects logistic regression with commune fixed effects. * 0.10 ** 0.05 *** 0.01 significance level. Values in parentheses indicate standard errors.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
39
Table A.5 Summaryofvariablesusedinregressionanalysesforpooledandpanelmodels,2010,2012and2014
Pooled Panel
2010 2012 2014 2010 2012 2014
N=6749 N=6696 N=6618 N=3083 N=4526 N=2841
Outcome variables
Consumption Total per capita expenditure in '000 VDN in 2010 prices 13372 14793 15881 13184 14612 15673
Total Income Total per‐capita income in '000 VDN in 2010prices 13569 15050 16544 13247 14958 16672
Rice Per‐capita income (production value ‐ costs) from rice cultivation in '000 VDN in 2010 prices 1290 1294 1261 1294 1291 1302
Other staples Per‐capita income (production value ‐ costs) from (non‐rice) staple cultivation in '000 VDN in 2010 prices 797 1167 1008 883 1137 983
Industrial crops Per‐capita income (production value ‐ costs) from industrial crop cultivation in '000 VDN in 2010 prices 542 556 532 554 561 559
Livestock Per‐capita income (production value ‐ costs) from livestock holdings activities in '000 VDN in 2010 prices 788 900 1077 792 901 1089
Forestry Per‐capita income (production value ‐ costs) from forestry activities in '000 VDN in 2010 prices 200 239 211 219 257 221
Fishing Per‐capita income (production value ‐ costs) from fishing activities in '000 VDN in 2010 prices 474 516 616 452 487 555
Agric wage Per‐capita income from agricultural wage employment in '000 VDN in 2010 prices 587 713 713 563 713 709
Non‐ag wage unskilled Per‐capita income from non‐agricultural wage employment in low‐skill occupations in '000 VDN in 2010 prices 825 818 1017 788 807 987
Non‐ag wage skilled Per‐capita income from non‐agricultural wage employment in high‐skill occupations in '000 VDN in 2010 prices 2821 3409 3975 2588 3354 3783
Business Per‐capita income (production value ‐ costs) from non‐agricultural self‐employment in '000 VDN in 2010 prices 2687 2721 3072 2611 2662 3227
Weather variables
Annual rain annual average of monthly rainfall in survey year in mm 142 182 168 141 184 169
Annual temperature annual average of monthly mean temperature in survey year in C 26 25 25 26 25 25
Seasonal rain Q1 seasonal average of monthly rainfall in December‐March in mm 53 51 32 53 51 35
Seasonal rain Q2 seasonal average of monthly rainfall in April‐July in mm 219 270 228 217 272 231
Seasonal rain Q3 seasonal average of monthly rainfall in August‐November in mm 153 226 243 153 228 241
Seasonal temperature Q1 seasonal average of monthly mean temperature in December‐March in C 24 23 23 24 23 23
Seasonal temperature Q2 seasonal average of monthly mean temperature in April‐July in C 29 28 28 29 28 28
Seasonal temperature Q3 seasonal average of monthly mean temperature in August‐November in C 25 25 24 25 25 24
Wet months Number of months with rainfall that exceeds long‐term mean by more than 1.5 sd 0.85 1.50 1.30 0.85 1.62 1.28
Dry months Number of months with rainfall that falls below long‐term mean by more than 1.5 sd 0.79 0.22 0.27 0.78 0.27 0.20
Hot months Number of months with maximum temperature that exceeds long‐term mean by more than 1.5 sd 1.98 0.25 0.22 1.98 0.28 0.44
Cold months Number of months with minimum temperature that falls below long‐term mean by more than 1.5 0.07 0.37 1.28 0.07 0.27 1.23
Minimum rain Rainfall in mm in driest month 372 490 458 372 503 458
Maximum rain Rainfall in mm in wettest month 11 19 11 11 19 11
Minimum temperature Mean of minimum temperature in C in the coldest month 33 33 33 33 33 33
Maximum temperature Mean of maximum temperature in C in the hottest month 18 16 15 18 16 15 Flood Dummy = 1 if community reported a flood 0.122 0.099 0.113 0.121 0.103 0.099
Drought Dummy = 1 if community reported a drought 0.118 0.038 0.050 0.126 0.043 0.058
Storm Dummy = 1 if community reported a storm 0.131 0.137 0.151 0.123 0.139 0.139
Commune controls
Long‐term rainfall mean Mean of monthly rainfall in last 30 years 155.9 155.6 157.6 ‐ ‐ ‐
Long‐term temperature mean Mean of monthly temperature mean in last 30 years 25.3 25.3 25.3 ‐ ‐ ‐
Long‐term rainfall variability Standard deviation of monthly rainfall in last 30 years 55.9 55.4 56.7 ‐ ‐ ‐
40
Long‐term temperature variability Standard deviation of monthly temperature mean in last 30 years 0.7 0.7 0.7 ‐ ‐ ‐
Gravel Topsoil with gravel content measured as % of volume 7.0 7.0 7.0 ‐ ‐ ‐
Sand Fraction of topsoil with sand content measured as % of weight 36.8 37.0 37.0 ‐ ‐ ‐
Clay Fraction of topsoil with clay content measured as % of weight 35.6 35.5 35.6 ‐ ‐ ‐
Tree cover % of community area with tree cover 16.7 16.8 16.9 ‐ ‐ ‐
Slope Median slope category in community: 1=least steep – 8=most steep 3.5 3.5 3.5 ‐ ‐ ‐
Distance city Distance from community to next main city 33.5 33.6 33.7 ‐ ‐ ‐
Distance road Distance from community to next road 3.7 3.7 3.7 ‐ ‐ ‐
Household controls
Area agriculture Area for agricultural activities household has access to 4805 5041 4956 5058 5205 5087
Area forest Forest area household has access to 2460 1997 1845 1669 2102 2064
Area water surface Water surface area household has access to 444 432 463 359 355 419
Workforce Share of household memebrs involved in income generating activitiies as % of total household size 0.9 0.8 0.9 0.9 0.8 0.9
Education Average number of school years of household members 6.0 6.1 6.2 6.0 6.1 6.1
Women Dummy = 1 if household head is female 0.2 0.2 0.2 ‐ ‐ ‐
Minority Dummy = 1 if household belongs to ethnic minority 0.2 0.2 0.2 ‐ ‐ ‐
Notes: Unweighted mean values for sample households.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
41
Table A.6 Estimatedimpactsofweathervariationonchangesinper‐capitaexpenditureandtotalincomes:allPanelhouseholds
Ln per‐capita expenditure Ln per‐capita total income (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.000249 0.000127
(0.000211) (0.000275) Annual temperature ‐0.197*** ‐0.216***
(0.0151) (0.0195) Seasonal rain S1 ‐0.000345 ‐0.000209
(0.000336) (0.000398) Seasonal rain S2 ‐0.000204* ‐0.000249
(0.000114) (0.000162) Seasonal rain S3 0.000312*** 0.000285**
(0.0000991) (0.000128) Seasonal temperature S1 ‐0.0879*** ‐0.0700***
(0.0104) (0.0144) Seasonal temperature S2 0.0479 0.00367
(0.0347) (0.0481) Seasonal temperature S3 ‐0.0717*** ‐0.101***
(0.0136) (0.0180) Wet months ‐0.00209 ‐0.00490
(0.00551) (0.00721) Dry months ‐0.0505*** ‐0.0612***
(0.00791) (0.0103) Hot months ‐0.0369*** ‐0.0416***
(0.00418) (0.00576) Cold months 0.0274*** 0.0390***
(0.00480) (0.00656) Max rain 0.0000800** 0.0000563
(0.0000324) (0.0000356)
Minimum rain 0.00144** ‐0.000187
(0.000587) (0.000784)
Max temperature ‐0.0621*** ‐0.104***
(0.0139) (0.0185)
Min temperature ‐0.0437*** ‐0.0459***
(0.00379) (0.00513)
Flood ‐0.0381*** ‐0.0472**
(0.0141) (0.0193)
Droughts ‐0.0589*** ‐0.0870***
(0.0169) (0.0231)
Storm 0.0184 ‐0.00467
(0.0123) (0.0168)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 10281 10281 10281 10281 10445 10274 10274 10274 10274 10438
R2 0.090 0.097 0.071 0.089 0.025 0.086 0.089 0.082 0.091 0.048
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
42
Table A.7 Estimatedimpactsofweathervariationondifferenceinper‐capitatotalincomes:PooledCross‐SectionOLSandFixedEffectsmodel
Ln per‐capita total income ‐ Pooled OLS Ln per‐capita total income ‐ Pooled FE (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.00236*** 0.000191
(0.000319) (0.000322) Annual temperature ‐0.0322 ‐0.0220
(0.0271) (0.0287) Seasonal rain S1 0.000826** ‐0.000669
(0.000353) (0.000410) Seasonal rain S2 0.000381*** 0.0000931
(0.000141) (0.000166) Seasonal rain S3 0.000932*** ‐0.0000946
(0.000115) (0.000147) Seasonal temperature S1 ‐0.0208* 0.00554
(0.0112) (0.0151) Seasonal temperature S2 ‐0.0727*** 0.0290
(0.0196) (0.0429) Seasonal temperature S3 ‐0.0170 ‐0.0282
(0.0182) (0.0209) Wet months ‐0.0149** ‐0.0129*
(0.00711) (0.00702) Dry months ‐0.00775 ‐0.0353***
(0.00931) (0.00949) Hot months ‐0.00356 ‐0.00124
(0.00697) (0.00718) Cold months ‐0.00414 0.0145**
(0.00644) (0.00638) Maximum rain 0.000197*** 0.0000218
(0.0000353) (0.0000354)
Minimum rain ‐0.00112 ‐0.000353
(0.000860) (0.000911)
Maximum temperature 0.0352** 0.00932
(0.0140) (0.0214)
Minimum temperature 0.0190*** ‐0.0131**
(0.00327) (0.00607)
Flood ‐0.0701*** ‐0.0364*
(0.0179) (0.0186)
Drought ‐0.0769*** ‐0.0428*
(0.0212) (0.0223)
Storm 0.0181 ‐0.00163
(0.0158) (0.0148)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Commune Controls Yes Yes Yes Yes Yes Commune fixed effects Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 19615 19615 19615 19615 19609 19747 19747 19747 19747 20044
R2 0.305 0.307 0.303 0.308 0.305 0.498 0.498 0.499 0.498 0.500
Notes: Table presents coefficients estimated from Pooled cross‐section Ordinary Least Squares (Pooled OLS) model and Pooled Cross‐Section Fixed
Effects (Pooled FE) model. * 0.10 ** 0.05 *** 0.01 significance level. Values in parentheses indicate standard errors corrected for cluster correlation
at commune‐level. Regressions include commune and household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
43
Table A.8 Estimatedimpactsofweathervariationonchangesinper‐capitatotalincome:Bottom40andTop60households
Ln per‐capita total income Bottom 40 Ln per‐capita total income Top 60 (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain ‐0.00129*** 0.000984***
(0.000412) (0.000361) Annual temperature ‐0.181*** ‐0.244***
(0.0286) (0.0251) Seasonal rain S1 ‐0.00186*** 0.000931*
(0.000583) (0.000507) Seasonal rain S2 ‐0.000814*** 0.000133
(0.000233) (0.000211) Seasonal rain S3 ‐0.000131 0.000500***
(0.000177) (0.000169) Seasonal temperature S1 ‐0.0619*** ‐0.0644***
(0.0193) (0.0192) Seasonal temperature S2 0.126* ‐0.154**
(0.0657) (0.0650) Seasonal temperature S3 ‐0.121*** ‐0.0852***
(0.0267) (0.0228) Wet months ‐0.0277** 0.00962
(0.0112) (0.00883) Dry months ‐0.00412 ‐0.0975***
(0.0150) (0.0130) Hot months ‐0.0260*** ‐0.0535***
(0.00813) (0.00756) Cold months 0.0491*** 0.0322***
(0.0103) (0.00807) Maximum rain 0.0000584 0.0000453
(0.0000480) (0.0000490)
Minimum rain ‐0.00392*** 0.00235**
(0.00112) (0.00102)
Maximum temperature ‐0.0245 ‐0.167***
(0.0264) (0.0236)
Minimum temperature ‐0.0460*** ‐0.0464***
(0.00766) (0.00644)
Flood ‐0.0737*** ‐0.0241
(0.0269) (0.0269)
Drought ‐0.0137 ‐0.148***
(0.0319) (0.0326)
Storm 0.00559 ‐0.0143
(0.0246) (0.0226)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 4040 4040 4040 4040 4108 6234 6234 6234 6234 6330
R2 0.059 0.078 0.059 0.073 0.042 0.112 0.114 0.110 0.114 0.055
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
44
Table A.9 Estimatedimpactsofweathervariationonchangesinper‐capitatotalincomes:Moved‐DownandMoved‐Uphouseholds
Ln per‐capita total income: Moved‐down households Ln per‐capita total income: Moved‐up households (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain ‐0.000130 0.00281***
(0.000416) (0.000558) Annual temperature ‐0.0766*** ‐0.273***
(0.0280) (0.0320) Seasonal rain S1 ‐0.0000988 0.00226***
(0.000713) (0.000714) Seasonal rain S2 ‐0.000200 0.000109
(0.000267) (0.000349) Seasonal rain S3 0.0000504 0.00139***
(0.000205) (0.000230) Seasonal temperature S1 ‐0.00754 ‐0.0987***
(0.0215) (0.0248) Seasonal temperature S2 0.0264 0.0180
(0.0777) (0.0899) Seasonal temperature S3 ‐0.0723*** ‐0.147***
(0.0253) (0.0266) Wet months ‐0.0187* 0.0443***
(0.0111) (0.0141) Dry months ‐0.0317* ‐0.0769***
(0.0174) (0.0201) Hot months ‐0.0130 ‐0.0885***
(0.00855) (0.00916) Cold months 0.0103 0.0389***
(0.00912) (0.0120) Maximum rain ‐0.0000405 0.000253***
(0.0000531) (0.0000745)
Minimum rain 0.00182 0.00151
(0.00135) (0.00136)
Maximum temperature ‐0.0193 ‐0.226***
(0.0286) (0.0293)
Minimum temperature ‐0.0178** ‐0.0620***
(0.00787) (0.00840)
Flood ‐0.0486 ‐0.0569*
(0.0306) (0.0343)
Drought ‐0.00504 ‐0.161***
(0.0384) (0.0416)
Storm 0.00396 0.0286
(0.0267) (0.0312)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 3533 3533 3533 3533 3588 2949 2949 2949 2949 2998
R2 0.051 0.053 0.052 0.052 0.047 0.210 0.217 0.181 0.209 0.050
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
45
Table A.10 Estimatedimpactsofweathervariationonchangesinper‐capitatotalincomes:EthnicminoritiesandFemale‐headedhouseholds
Ln per‐capita total income: Ethnic minorities Ln per‐capita total income: Female‐headed households (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain ‐0.000629 0.000514
(0.000594) (0.000603) Annual temperature ‐0.127*** ‐0.222***
(0.0359) (0.0452) Seasonal rain S1 ‐0.00140 ‐0.00165*
(0.000974) (0.000927) Seasonal rain S2 ‐0.000412 0.000321
(0.000315) (0.000390) Seasonal rain S3 0.0000124 0.0000382
(0.000297) (0.000305) Seasonal temperature S1 ‐0.0349 ‐0.0269
(0.0248) (0.0319) Seasonal temperature S2 ‐0.0893 ‐0.0785
(0.0873) (0.115) Seasonal temperature S3 ‐0.0281 ‐0.135***
(0.0358) (0.0442) Wet months ‐0.0150 ‐0.0118
(0.0155) (0.0158) Dry months 0.0237 ‐0.0574**
(0.0212) (0.0227) Hot months ‐0.0354*** ‐0.0509***
(0.0103) (0.0128) Cold months 0.00822 0.0359**
(0.0169) (0.0152) Maximum rain 0.0000539 0.0000480
(0.0000837) (0.0000861)
Minimum rain ‐0.000538 ‐0.000628
(0.00161) (0.00188)
Maximum temperature 0.00275 ‐0.110***
(0.0369) (0.0401)
Minimum temperature ‐0.0333*** ‐0.0554***
(0.0108) (0.0120)
Flood ‐0.0480 ‐0.116**
(0.0354) (0.0493)
Drought ‐0.0334 ‐0.230***
(0.0352) (0.0551)
Storm ‐0.00336 ‐0.00828
(0.0302) (0.0395)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 2246 2246 2246 2246 2294 2145 2145 2145 2145 2170
R2 0.082 0.090 0.080 0.083 0.069 0.074 0.081 0.068 0.082 0.047
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
46
Table A.11 Estimatedimpactsofweathervariationonchangesinper‐capitatotalincomes:Dry‐coldandDry‐hotzones
Ln per‐capita total income: Dry‐cold zone Ln per‐capita total income: Dry‐hot zone (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.000942 ‐0.00152*
(0.000592) (0.000823) Annual temperature ‐0.175*** ‐0.282***
(0.0354) (0.0517) Seasonal rain S1 ‐0.00566*** ‐0.0163***
(0.00193) (0.00469) Seasonal rain S2 0.00133*** 0.00300***
(0.000350) (0.000916) Seasonal rain S3 0.000818 ‐0.00261***
(0.000687) (0.000558) Seasonal temperature S1 0.00925 ‐0.0131
(0.0214) (0.0543) Seasonal temperature S2 ‐0.0776 ‐0.788***
(0.0908) (0.216) Seasonal temperature S3 ‐0.0727** 0.0486
(0.0334) (0.0725) Wet months 0.0334** ‐0.0149
(0.0141) (0.0193) Dry months ‐0.107*** ‐0.103***
(0.0193) (0.0362) Hot months ‐0.0206* ‐0.0567***
(0.0111) (0.0148) Cold months 0.0428*** 0.0132
(0.00947) (0.0270) Maximum rain 0.000247 ‐0.000560***
(0.000253) (0.000167)
Minimum rain 0.0000430 0.00552**
(0.00153) (0.00265)
Maximum temperature ‐0.0211 ‐0.254***
(0.0420) (0.0826)
Minimum temperature ‐0.0570*** ‐0.0396**
(0.00865) (0.0169)
Flood 0.0957** ‐0.0339
(0.0425) (0.0645)
Drought ‐0.0857* ‐0.0526
(0.0511) (0.0745)
Storm 0.0150 ‐0.0733
(0.0397) (0.0584)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 3162 3162 3162 3162 3162 765 765 765 765 765
R2 0.099 0.118 0.102 0.112 0.047 0.121 0.162 0.137 0.126 0.056
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
47
Table A.12 Estimatedimpactsofweathervariationonchangesinper‐capitatotalincomes:Wet‐coldandWet‐hotzones
Ln per‐capita total income: Wet‐cold zone Ln per‐capita total income: Wet‐hot zone (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain ‐0.00000514 0.000104
(0.000540) (0.000408) Annual temperature ‐0.158*** ‐0.273***
(0.0418) (0.0354) Seasonal rain S1 0.00135 0.0000126
(0.000979) (0.000488) Seasonal rain S2 ‐0.00112*** ‐0.00102***
(0.000321) (0.000372) Seasonal rain S3 0.000593** 0.000226
(0.000245) (0.000168) Seasonal temperature S1 ‐0.122*** ‐0.171***
(0.0322) (0.0283) Seasonal temperature S2 0.216* ‐0.0781
(0.115) (0.0774) Seasonal temperature S3 ‐0.0929** 0.0259
(0.0433) (0.0439) Wet months 0.0247 ‐0.0540***
(0.0180) (0.0119) Dry months 0.0126 ‐0.0257
(0.0268) (0.0160) Hot months ‐0.0613*** ‐0.0605***
(0.0162) (0.00898) Cold months 0.0617*** 0.0199*
(0.0170) (0.0117) Maximum rain 0.0000757 0.0000663
(0.0000695) (0.0000430)
Minimum rain ‐0.00226 0.000261
(0.00172) (0.00124)
Maximum temperature ‐0.165*** ‐0.122***
(0.0505) (0.0238)
Minimum temperature ‐0.0122 ‐0.0629***
(0.0129) (0.00974)
Flood ‐0.0883** ‐0.0910***
(0.0421) (0.0271)
Drought ‐0.0357 ‐0.128***
(0.0407) (0.0378)
Storm ‐0.0352 0.0151
(0.0354) (0.0233)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 1941 1941 1941 1941 1941 4406 4406 4406 4406 4570
R2 0.039 0.060 0.060 0.049 0.023 0.106 0.114 0.104 0.110 0.078
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
48
Table A.13 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromricecultivationandstaplecrops
Ln per‐capita income from rice cultivation Ln per‐capita income from staple crops (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain ‐0.00144*** ‐0.00211**
(0.000462) (0.000895) Annual temperature ‐0.0987** ‐0.0849
(0.0399) (0.0575) Seasonal rain S1 0.00101 ‐0.00144
(0.000784) (0.00143) Seasonal rain S2 ‐0.000517* ‐0.000480
(0.000279) (0.000426) Seasonal rain S3 ‐0.000366* ‐0.000927*
(0.000204) (0.000524) Seasonal temperature S1 ‐0.0683** ‐0.0390
(0.0285) (0.0379) Seasonal temperature S2 ‐0.206** ‐0.0519
(0.0969) (0.132) Seasonal temperature S3 0.0821** ‐0.000881
(0.0347) (0.0521) Wet months ‐0.0331** ‐0.0502**
(0.0138) (0.0243) Dry months 0.0135 0.00837
(0.0185) (0.0341) Hot months ‐0.0248* ‐0.0188
(0.0128) (0.0183) Cold months 0.000849 0.00231
(0.0103) (0.0203) Maximum rain ‐0.000129** ‐0.0000319
(0.0000571) (0.000106)
Minimum rain 0.00378*** 0.0000221
(0.00135) (0.00227)
Maximum temperature 0.0343 0.158**
(0.0403) (0.0674)
Minimum temperature ‐0.00480 ‐0.0247
(0.00923) (0.0165)
Flood ‐0.0518* ‐0.187***
(0.0296) (0.0570)
Drought 0.0513 0.0486
(0.0369) (0.0618)
Storm 0.0139 ‐0.0174
(0.0287) (0.0546)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 5926 5926 5926 5926 6000 5102 5102 5102 5102 5181
R2 0.056 0.065 0.055 0.055 0.053 0.018 0.018 0.017 0.019 0.020
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
49
Table A.14 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromindustrializedcropsandlivestock
Ln per‐capita income from industrialized crops Ln per‐capita income from livestock (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.00145 ‐0.00125
(0.00115) (0.000794) Annual temperature ‐0.205*** ‐0.266***
(0.0790) (0.0553) Seasonal rain S1 ‐0.000500 ‐0.00140
(0.00147) (0.00108) Seasonal rain S2 0.000454 ‐0.000330
(0.000611) (0.000406) Seasonal rain S3 0.000414 ‐0.000631
(0.000492) (0.000417) Seasonal temperature S1 ‐0.118** ‐0.130***
(0.0532) (0.0414) Seasonal temperature S2 ‐0.224 0.0588
(0.174) (0.129) Seasonal temperature S3 0.0739 ‐0.0919*
(0.0677) (0.0495) Wet months ‐0.0798*** ‐0.0302
(0.0281) (0.0220) Dry months ‐0.0585 ‐0.0908***
(0.0430) (0.0321) Hot months ‐0.107*** ‐0.0262
(0.0221) (0.0178) Cold months 0.0144 0.0377**
(0.0283) (0.0184) Maximum rain 0.000185 ‐0.000141
(0.000129) (0.0000993)
Minimum rain ‐0.000496 0.00318
(0.00297) (0.00216)
Maximum temperature ‐0.128* ‐0.112**
(0.0666) (0.0532)
Minimum temperature ‐0.0291 ‐0.0303**
(0.0204) (0.0145)
Flood ‐0.103 ‐0.0445
(0.0738) (0.0553)
Drought ‐0.134* ‐0.129**
(0.0753) (0.0618)
Storm 0.0124 0.106**
(0.0634) (0.0509)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 3200 3200 3200 3200 3282 6012 6012 6012 6012 6103
R2 0.034 0.038 0.048 0.028 0.019 0.018 0.019 0.015 0.016 0.009
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
50
Table A.15 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromforestryandfishing
Ln per‐capita income from forestry Ln per‐capita income from fishing (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.00115 ‐0.00299**
(0.00108) (0.00151) Annual temperature 0.134** ‐0.103
(0.0637) (0.0940) Seasonal rain S1 0.000530 ‐0.00146
(0.00161) (0.00219) Seasonal rain S2 0.000788 ‐0.000598
(0.000507) (0.000681) Seasonal rain S3 0.0000975 ‐0.00137
(0.000575) (0.000896) Seasonal temperature S1 0.0600 ‐0.110*
(0.0463) (0.0664) Seasonal temperature S2 0.360** 0.200
(0.141) (0.237) Seasonal temperature S3 ‐0.113* ‐0.0176
(0.0625) (0.0973) Wet months 0.0214 0.00739
(0.0278) (0.0355) Dry months ‐0.0590 0.0131
(0.0372) (0.0481) Hot months 0.0308 0.0568*
(0.0192) (0.0298) Cold months ‐0.0568* 0.00459
(0.0292) (0.0299) Maximum rain ‐0.0000860 ‐0.000639
(0.000148) (0.000433)
Minimum rain ‐0.00145 0.00763*
(0.00282) (0.00452)
Maximum temperature 0.00682 0.199**
(0.0636) (0.0881)
Minimum temperature 0.0164 ‐0.0631***
(0.0189) (0.0234)
Flood ‐0.0585 0.0469
(0.0579) (0.100)
Drought 0.113* 0.0417
(0.0607) (0.120)
Storm ‐0.0146 0.0108
(0.0554) (0.0801)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 3069 3069 3069 3069 3139 2389 2389 2389 2389 2427
R2 0.004 0.014 0.007 0.003 0.003 0.008 0.013 0.010 0.024 0.004
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
51
Table A.16 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromagriculturalwagesandunskillednon‐agriculturalwages
Ln per‐capita income from agricultural wages Ln per‐capita income from unskilled non‐agricultural wages (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.000290 ‐0.00168
(0.00111) (0.00150) Annual temperature ‐0.345*** ‐0.453***
(0.0971) (0.112) Seasonal rain S1 0.00176 ‐0.00670***
(0.00194) (0.00205) Seasonal rain S2 ‐0.00164 0.00170*
(0.00113) (0.000943) Seasonal rain S3 0.000702 ‐0.00239***
(0.000503) (0.000693) Seasonal temperature S1 ‐0.206*** ‐0.118
(0.0721) (0.0827) Seasonal temperature S2 0.210 ‐0.258
(0.262) (0.290) Seasonal temperature S3 ‐0.128 ‐0.112
(0.109) (0.0898) Wet months ‐0.0385 ‐0.0113
(0.0327) (0.0421) Dry months 0.0864* ‐0.181***
(0.0511) (0.0542) Hot months ‐0.0762*** ‐0.0490
(0.0260) (0.0324) Cold months 0.0249 ‐0.0448
(0.0301) (0.0377) Maximum rain 0.000294** ‐0.000296*
(0.000121) (0.000179)
Minimum rain ‐0.00146 0.00695*
(0.00389) (0.00408)
Maximum temperature ‐0.0973 ‐0.0992
(0.0768) (0.103)
Minimum temperature ‐0.0707** ‐0.0998***
(0.0301) (0.0273)
Flood ‐0.0658 0.0240
(0.0810) (0.0998)
Drought ‐0.0618 ‐0.259**
(0.105) (0.126)
Storm 0.00501 0.0278
(0.0732) (0.0937)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 1680 1680 1680 1680 1721 2340 2340 2340 2340 2376
R2 0.038 0.045 0.028 0.032 0.011 0.047 0.067 0.039 0.054 0.019
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
52
Table A.17 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromskillednon‐agriculturalwagesandbusinessself‐employment
Ln per‐capita income from skilled non‐agricultural wages Ln per‐capita income from business self‐employment (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.000678 0.000835
(0.000656) (0.000809) Annual temperature ‐0.336*** ‐0.136*
(0.0450) (0.0699) Seasonal rain S1 0.000650 ‐0.000583
(0.000972) (0.00149) Seasonal rain S2 ‐0.000642* 0.000131
(0.000353) (0.000509) Seasonal rain S3 0.000936*** 0.000390
(0.000326) (0.000396) Seasonal temperature S1 ‐0.0938*** ‐0.0237
(0.0345) (0.0474) Seasonal temperature S2 0.0235 0.0897
(0.115) (0.159) Seasonal temperature S3 ‐0.200*** ‐0.128**
(0.0396) (0.0599) Wet months 0.0439** 0.0147
(0.0174) (0.0239) Dry months ‐0.107*** ‐0.0581
(0.0246) (0.0385) Hot months ‐0.0530*** ‐0.0197
(0.0123) (0.0216) Cold months 0.0776*** 0.0379**
(0.0123) (0.0188) Maximum rain 0.000212** 0.000156
(0.000104) (0.000115)
Minimum rain ‐0.00197 ‐0.00544**
(0.00174) (0.00269)
Maximum temperature ‐0.157*** ‐0.180***
(0.0427) (0.0642)
Minimum temperature ‐0.0901*** ‐0.0483***
(0.0112) (0.0170)
Flood 0.00402 0.0427
(0.0501) (0.0671)
Drought ‐0.148** ‐0.158*
(0.0635) (0.0914)
Storm 0.0250 ‐0.0432
(0.0444) (0.0599)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 3676 3676 3676 3676 3719 3036 3036 3036 3036 3071
R2 0.099 0.115 0.101 0.127 0.026 0.017 0.020 0.016 0.036 0.010
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
53
Table A.18 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromricecultivationandwinter‐springricecultivationintheMekongRiverDelta
Ln per‐capita income from rice cultivation in the Mekong River Delta Ln per‐capita income from winter‐spring rice in the Mekong River Delta (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.00142 ‐0.00315
(0.00246) (0.00256) Annual temperature ‐0.177 ‐0.0549
(0.140) (0.144) Seasonal rain S1 0.00605* 0.00572
(0.00357) (0.00430) Seasonal rain S2 0.000272 0.000595
(0.00303) (0.00366) Seasonal rain S3 ‐0.00168 ‐0.00314**
(0.00140) (0.00149) Seasonal temperature S1 ‐0.105 0.0182
(0.127) (0.124) Seasonal temperature S2 ‐0.347 ‐0.211
(0.529) (0.576) Seasonal temperature S3 ‐0.00820 ‐0.171
(0.269) (0.266) Wet months 0.0130 ‐0.0130
(0.0329) (0.0369) Dry months ‐0.142** ‐0.135*
(0.0640) (0.0707) Hot months ‐0.0539 ‐0.0238
(0.0334) (0.0309) Cold months ‐0.0705 ‐0.107**
(0.0458) (0.0506) Maximum rain ‐0.000338 ‐0.000950
(0.000758) (0.000770)
Minimum rain ‐0.00383 0.000288
(0.0143) (0.0206)
Maximum temperature ‐0.179 ‐0.0536
(0.127) (0.136)
Minimum temperature 0.00130 0.0269
(0.0729) (0.0864)
Flood ‐0.00687 ‐0.0495
(0.111) (0.108)
Drought 0.0391 0.529
(0.311) (0.347)
Storm 0.118 0.00796
(0.0772) (0.0787)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 816 816 816 816 824 674 674 674 674 682
R2 0.123 0.133 0.137 0.125 0.123 0.161 0.175 0.175 0.161 0.163
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
54
Table A.19 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromsummer‐autumnricecultivationandautumn‐winterricecultivationintheMekongRiverDelta
Ln per‐capita income from summer‐autumn rice in the Mekong Ln per‐capita income from autumn‐winter rice in the Mekong (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.00233 ‐0.00256
(0.00278) (0.00291) Annual temperature ‐0.136 ‐0.287
(0.158) (0.210) Seasonal rain S1 0.0112*** ‐0.000820
(0.00412) (0.00484) Seasonal rain S2 0.00209 0.00180
(0.00345) (0.00339) Seasonal rain S3 ‐0.00252 ‐0.00102
(0.00159) (0.00181) Seasonal temperature S1 ‐0.102 0.0963
(0.140) (0.158) Seasonal temperature S2 ‐0.401 0.000533
(0.619) (1.018) Seasonal temperature S3 ‐0.0386 ‐0.549
(0.298) (0.400) Wet months 0.0249 ‐0.0371
(0.0373) (0.0373) Dry months ‐0.174** 0.00122
(0.0734) (0.0758) Hot months ‐0.0553 ‐0.0262
(0.0377) (0.0531) Cold months ‐0.0581 ‐0.00746
(0.0540) (0.0617) Maximum rain ‐0.000983 ‐0.000814
(0.000733) (0.000779)
Minimum rain 0.0240 ‐0.00712
(0.0159) (0.0186)
Maximum temperature ‐0.266** ‐0.106
(0.134) (0.176)
Minimum temperature ‐0.0521 ‐0.0194
(0.0777) (0.0964)
Flood ‐0.0139 0.0262
(0.114) (0.158)
Drought 0.813** 0.245
(0.359) (0.349)
Storm 0.0416 0.137
(0.0800) (0.115)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 698 698 698 698 706 509 509 509 509 513
R2 0.132 0.159 0.153 0.141 0.143 0.151 0.159 0.142 0.151 0.149
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
55
Table A.20 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfrommaizecultivationintheNorthWestandcoffeecultivationintheCentralHighlands
Ln per‐capita income from maize cultivation in the North West Ln per‐capita income from coffee cultivation in the Central Highlands (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.000723 0.00202
(0.00203) (0.00288) Annual temperature 0.103 ‐0.776***
(0.141) (0.264) Seasonal rain S1 ‐0.00249 ‐0.0428
(0.00793) (0.0278) Seasonal rain S2 0.000630 ‐0.00375
(0.00170) (0.00722) Seasonal rain S3 ‐0.00448 ‐0.00121
(0.00428) (0.00158) Seasonal temperature S1 ‐0.245 ‐0.463*
(0.168) (0.242) Seasonal temperature S2 0.505 ‐0.245
(0.597) (0.888) Seasonal temperature S3 ‐0.0693 0.147
(0.168) (0.302) Wet months 0.0586 0.0790
(0.0912) (0.0671) Dry months 0.0565 ‐0.228
(0.0877) (0.163) Hot months ‐0.00296 ‐0.126**
(0.0380) (0.0577) Cold months ‐0.220** 0.00167
(0.102) (0.186) Maximum rain 0.000366 ‐0.000841*
(0.000882) (0.000493)
Minimum rain 0.0152 0.0403*
(0.0114) (0.0203)
Maximum temperature 0.181 ‐0.681***
(0.129) (0.176)
Minimum temperature 0.0195 0.0529
(0.0490) (0.118)
Flood 0.0443 ‐0.530**
(0.178) (0.253)
Drought 0.0962 ‐0.424**
(0.147) (0.165)
Storm ‐0.368** ‐0.243*
(0.161) (0.130)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 357 357 357 357 365 306 306 306 306 338
R2 0.107 0.146 0.149 0.141 0.131 0.222 0.245 0.220 0.198 0.154
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
56
Table A.21 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromforestryandfishingintheNorthWest
Ln per‐capita income from forestry in the North West Ln per‐capita income from fishing in the North West (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.00334* ‐0.00182
(0.00168) (0.00286) Annual temperature 0.106 0.450**
(0.0835) (0.216) Seasonal rain S1 0.00187 0.0121
(0.00501) (0.0111) Seasonal rain S2 0.000898 ‐0.00300
(0.000931) (0.00210) Seasonal rain S3 ‐0.00191 0.00439
(0.00205) (0.00657) Seasonal temperature S1 0.0360 0.00297
(0.0993) (0.205) Seasonal temperature S2 0.350 0.909
(0.404) (0.916) Seasonal temperature S3 ‐0.264** 0.0605
(0.115) (0.254) Wet months 0.0431 ‐0.0287
(0.0710) (0.120) Dry months ‐0.0968 0.0616
(0.0605) (0.105) Hot months 0.0199 0.155**
(0.0289) (0.0710) Cold months ‐0.0794** 0.0500
(0.0358) (0.130) Maximum rain ‐0.000531 ‐0.00145
(0.000715) (0.00134)
Minimum rain 0.0209** 0.0102
(0.00965) (0.0219)
Maximum temperature 0.0135 0.358
(0.104) (0.258)
Minimum temperature 0.0167 ‐0.00965
(0.0340) (0.0785)
Flood ‐0.131 ‐0.539*
(0.147) (0.300)
Drought 0.0867 0.306
(0.116) (0.234)
Storm 0.0367 0.215
(0.126) (0.314)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 500 500 500 500 513 205 205 205 205 216
R2 0.037 0.068 0.044 0.048 0.025 0.209 0.233 0.258 0.215 0.161
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
57
Table A.22 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromagriculturalwagesintheNorthEastandunskillednon‐agriculturalwagesintheSouthCentralCoast
Ln per‐capita income from agricultural wages in the North East Ln per‐capita income from unskilled non‐ag wages in the South C Coast (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.000245 ‐0.00793***
(0.00336) (0.00283) Annual temperature ‐0.279 ‐0.990***
(0.173) (0.353) Seasonal rain S1 0.0178** 0.00204
(0.00867) (0.0120) Seasonal rain S2 ‐0.000778 0.00197
(0.00306) (0.00810) Seasonal rain S3 0.00659* ‐0.00292
(0.00383) (0.00232) Seasonal temperature S1 0.299 ‐0.487
(0.184) (0.294) Seasonal temperature S2 ‐0.310 ‐0.185
(0.761) (1.724) Seasonal temperature S3 0.0886 ‐0.119
(0.303) (0.438) Wet months ‐0.139 0.102
(0.121) (0.163) Dry months ‐0.108 ‐0.249
(0.183) (0.201) Hot months ‐0.0178 0.0996
(0.0571) (0.142) Cold months 0.102** 0.00648
(0.0468) (0.319) Maximum rain ‐0.00186 ‐0.000763*
(0.00118) (0.000440)
Minimum rain ‐0.0187 0.0269*
(0.0117) (0.0137)
Maximum temperature ‐0.296 0.631
(0.322) (0.420)
Minimum temperature ‐0.0441 0.126
(0.0593) (0.202)
Flood 0.598* ‐0.111
(0.338) (0.190)
Drought 0.0595
(0.293)
Storm ‐0.0655 0.0895
(0.372) (0.242)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 109 109 109 109 112 209 209 209 209 218
R2 0.304 0.439 0.370 0.441 0.455 0.269 0.285 0.161 0.239 0.140
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
58
Table A.23 Estimatedimpactsofweathervariationonchangesinper‐capitaincomesfromskillednon‐agriculturalwagesandbusinessself‐employmentintheSouthEastregion
Ln per‐capita income from skilled non‐ag wages in the South East Ln per‐capita income from businesses in the South East (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.00235 0.00237
(0.00199) (0.00330) Annual temperature ‐0.455*** ‐0.519*
(0.110) (0.283) Seasonal rain S1 0.0248 0.0178
(0.0189) (0.0562) Seasonal rain S2 ‐0.00724* ‐0.00219
(0.00370) (0.00550) Seasonal rain S3 0.00202* 0.00124
(0.00115) (0.00261) Seasonal temperature S1 ‐0.611*** ‐0.322
(0.165) (0.280) Seasonal temperature S2 0.285 0.0110
(0.432) (0.926) Seasonal temperature S3 0.201 ‐0.131
(0.225) (0.370) Wet months ‐0.0148 ‐0.158
(0.0595) (0.105) Dry months ‐0.0692 0.108
(0.0630) (0.107) Hot months ‐0.0716** ‐0.154***
(0.0330) (0.0587) Cold months ‐0.0196 0.170
(0.0543) (0.124) Maximum rain ‐0.0000867 0.000499
(0.000732) (0.000964)
Minimum rain ‐0.0203 ‐0.0494
(0.0373) (0.0704)
Maximum temperature ‐0.313** ‐0.185
(0.151) (0.251)
Minimum temperature ‐0.0686 ‐0.0963
(0.0549) (0.0889)
Flood 0.0645 0.202
(0.177) (0.304)
Drought ‐0.875*** ‐0.895**
(0.176) (0.343)
Storm 0.00292 ‐0.337**
(0.0861) (0.144)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 460 460 460 460 467 282 282 282 282 285
R2 0.204 0.244 0.192 0.203 0.209 0.156 0.159 0.153 0.149 0.164
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.
59
Table A.24 Estimatedimpactsofweathervariationonchangesinper‐capitatotalincomes:2010‐2012and2012‐2014
Ln per‐capita total income 2010‐2012 Ln per‐capita total income 2012‐2014 (1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
Annual rain 0.00200*** ‐0.00125**
(0.000422) (0.000521) Annual temperature ‐0.0721*** ‐0.163***
(0.0273) (0.0581) Seasonal rain S1 0.000910** ‐0.00408***
(0.000438) (0.00101) Seasonal rain S2 0.000461* 0.000181
(0.000273) (0.000284) Seasonal rain S3 0.000728*** ‐0.000564**
(0.000172) (0.000239) Seasonal temperature S1 ‐0.0368 ‐0.00193
(0.0248) (0.0221) Seasonal temperature S2 ‐0.0551 ‐0.0271
(0.0695) (0.0727) Seasonal temperature S3 0.000994 ‐0.124***
(0.0374) (0.0309) Wet months 0.0149 ‐0.0416***
(0.0109) (0.00967) Dry months ‐0.0628*** ‐0.0283
(0.0136) (0.0189) Hot months ‐0.0385*** 0.0400**
(0.00776) (0.0170) Cold months ‐0.00785 0.0232***
(0.0140) (0.00807) Maximum rain 0.000198*** ‐0.0000469
(0.0000658) (0.0000477)
Minimum rain 0.000847 ‐0.000338
(0.00124) (0.00185)
Maximum temperature ‐0.110*** ‐0.0208
(0.0214) (0.0303)
Minimum temperature ‐0.0299*** ‐0.0561***
(0.0104) (0.0114)
Flood ‐0.0427 ‐0.0701***
(0.0268) (0.0260)
Drought ‐0.117*** ‐0.00465
(0.0293) (0.0373)
Storm 0.0000198 ‐0.0128
(0.0235) (0.0233)
Household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 7480 7480 7480 7480 7599 7247 7247 7247 7247 7360
R2 0.091 0.092 0.086 0.093 0.048 0.057 0.073 0.063 0.072 0.050
Notes: Table presents coefficients estimated from Panel sub‐sample Fixed Effects model. * 0.10 ** 0.05 *** 0.01 significance level. Values in
parentheses indicate standard errors, which are corrected for cluster correlation at commune‐level in regressions (1) – (4). Regressions include
household controls as presented in Table A.5.
Source: Author’s calculation based on VHLSS 2010, 2012 & 2014 and CRU data.