As Pakistan is currently facing a severe shortage of irrigation water, this paper analyzes the determinants of water scarcity and its impact on the yield of cereal crops (wheat, maize and rice), household income, food security and poverty levels by employing the propensity-score-matching approach. This study is based on a comprehensive set of cross-sectional data collected from 950 farmers from all four major provinces in Pakistan. The empirical analysis indicated that farmers with a water-scarcity problem have lower yield and household income, and are food insecure. Poverty levels were higher: in the range of 7–12% for a household facing a water-scarcity problem. The policy implications of the study are that the public and private sector in Pakistan needs to invest in irrigation water management to maintain the productivity of cereal crops which is important for household food security and poverty reduction.

INTRODUCTION

With increasing climate variability and the rapid melting of glaciers, water resource scarcity is becoming a major constraint on agricultural production in South Asia (Cruz et al. 2007). Given that water is one of the critical inputs for agriculture, climate variability and glacial retreat might threaten the food security of the region. Increasing temperatures will also accentuate the demand for irrigation water in the future. Recent studies predict that there will be at least a 10% increase in irrigation water demand with a 1 °C rise in temperature in arid and semi-arid regions of Asia (Sivakumar & Stefanski 2011). Furthermore, water availability is expected to decline; whereas global agricultural water demand is estimated to increase by approximately 19% by 2050. In Pakistan almost 90% of the total fresh water withdrawal is used for agricultural production (World Bank 2013), resulting in the depletion of aquifers in the country's main food producing states. With an increase in the global temperature and the melting of the Himalayan glaciers, the severity of the water shortage in South Asia is likely to increase exponentially. Water scarcity, combined with the rising demand for food due to population growth, will create increasing pressure to produce more with less irrigation water (Qureshi et al. 2003).

Water scarcity is defined from the perspective of individual water users who lack secure access to safe and affordable water to consistently satisfy their need for food production, drinking, washing, or livelihoods. Water scarcity is first and foremost a poverty issue. About 1.2 billion people live in areas of water scarcity and one in three people in the world face water shortages (Molden et al. 2007).

Irrigation consumes about 70% of the world's available water. There is an urgent need for the new strategies to improve the productivity of water in both irrigated and rain-fed agriculture and ensure access to water and technologies by the poor (Barker & Koppen 1999). In Pakistan about 80% of the agricultural areas are irrigated by a canal irrigation system. In the past the main source of agricultural growth was the public-sector-funded canal irrigation and the private-sector-funded tube well irrigation systems (Shah 1993).

Water scarcity is an emerging challenge in Pakistan (Government of Pakistan 2013). Pakistan's water storage capacity stands at merely 30 days as opposed to a minimum requirement of 120 days. In Pakistan per capita water availability at the time of independence was 5,600 cubic meters against the current measure of 1,000 cubic meters and the water shortage is expected to rise to 31% of the people's needs by 2025.

Pakistan loses 13 million cubic feet of water every year into the sea. During periods of reduced river flow, seawater encroachment damages land, extending up to 100 kilometers upstream and impacting cultivable land. This is a dangerous trend for a country that uses nearly 90% of its water resources for agriculture and depends on the agriculture sector to remain buoyant (Ashraf 2013). The severity of the water crisis cannot be ignored while pursuing economic development as it serves as the backbone of the economy. Water scarcity is increasing rapidly due to increasing populations coupled with global warming and climate change, thereby necessitating serious interventions from policy makers.

The current energy crisis, especially the severe load-shedding in rural areas, has broadened the irrigation water scarcity issue. Due to climate changes the canal water is decreasing continuously and due to load-shedding the farmers cannot run the tube wells. In addition, small landholders (>80%) cannot afford the diesel tube wells in Pakistan.

The purpose of the current paper is to investigate the impact of the water scarcity on the cereal crop productivity (wheat, maize and rice), household income and poverty levels in Pakistan.

CONCEPTUAL FRAMEWORK

It is assumed that a rural household's utility due to the availability of water U(WA) is higher compared with the non-availability of water or when there is a water scarcity U(WNA): 
formula
1
The availability of water leads to higher crop yields (Ci), household income (Ii), food security (Fi) and lower levels of poverty (Pi): 
formula
2
In Equation (2) it is assumed that household utility levels are higher due to the availability of water compared with the non-availability of water. The impact of water availability is estimated by employing the propensity-score-matching (PSM) approach.

The method of matching has achieved popularity more recently as a tool of evaluation. It assumes that selection can be explained purely in terms of observable characteristics. Propensity score matching can be implemented with both cross-sectional and longitudinal datasets. Matching deals with the selection process by constructing a comparison group of individuals with observable characteristics similar to those of the treated. Applying the method is, in principle, simple. For every individual in the treatment group, a matching individual is found from among the non-treatment group. The choice of match is dictated by observable characteristics. What is required is to match each treatment-group individual with individuals sharing similar characteristics. The mean effect of treatment can then be calculated as the average difference in outcomes between the treated and non-treated.

The matching method is a non-parametric approach and is more general in the sense that no particular specification has to be assumed. The main purpose of the matching is to re-establish the conditions of an experiment when no such data are available.

It follows that the expected treatment effect for the treated population is of primary significance. This effect may be given as: 
formula
3
where is the average treatment effect for the treated (ATT), denotes the value of the outcome for farmers without water scarcity and is the value of the same variable for farmers facing water scarcity. As noted above, a major problem is that we do not observe . Although the difference can be estimated, it is a potentially biased estimator.
In the absence of experimental data, the PSM model can be employed to account for this sample selection bias (Dehejia & Wahba 2002). The PSM is defined as the conditional probability that a farmer adopts the new technology, given pre-adoption characteristics (Rosenbaum & Rubin 1983). To create the condition of a randomized experiment, the PSM employs the unconfoundedness assumption also known as conditional independence assumption, which implies that once Z is controlled for, technology adoption is random and uncorrelated with the outcome variables. (As pointed out by Imbens & Wooldridge (2009), unconfoundedness implies that we have a sufficiently rich set of predictors for the adoption indicator, contained in the vector of covariates, such that adjusting for differences in these covariates leads to valid estimates of causal effects.) The PSM can be expressed as: 
formula
4
where is the indicator for farmers without water scarcity and Z is the vector of pre-non-scarcity characteristics. The conditional distribution of Z, given is similar in both groups of non-water-scarce households and water-scarce households.

Unlike the parametric methods, propensity score matching requires no assumption about the functional form in specifying the relationship between outcome and predictors of outcome. The drawback of the approach is the strong assumption of unconfoundedness. As argued by Smith & Todd (2005), there may be systematic differences between adopters' and non-adopters' outcomes even after conditioning because selection is based on unmeasured characteristics. However, Jalan & Ravallion (2003) point out that the assumption is no more restrictive than those of an IV approach employed in cross-sectional data analysis. In a study by Michalopoulos et al. (2004) to assess which non-experimental method provides the most accurate estimates in the absence of random assignment, they conclude that propensity score methods provided a specification check that tended to eliminate biases that were larger than average. On the other hand, a fixed effects model did not consistently improve the results.

In practice, the choice of matching method often appears to make little difference (Smith & Todd 2005). In small samples the choice of matching approach can be important (Heckman et al. 1997). However, there appears to be little formal guidance in the choice of optimal method. The choice should be guided in part by what the distribution of scores in the comparison and treatment samples looks like. For example, if some treated persons have lots of close neighbors and others only have one, one would favor kernel matching or caliper matching over multiple nearest-neighbor matching (NNM) because either of the latter would result in many poor matches. Taking another example, if the comparison and treatment samples are of roughly equal size, then single NNM makes more sense than it does when the comparison sample is much larger than the treatment sample because, in the latter case, single NNM would result in discarding useful information. Pragmatically, it seems sensible to try a number of approaches because, as noted earlier, the performance of different matching estimators varies case by case and depends largely on the data structure at hand (Zhao 2000). Should they give similar results, the choice may be unimportant. Should the results differ, further investigation may be needed in order to reveal more about the source of the disparity. This serves to reinforce the belief that matching should be implemented in a thoughtful way and not treated as black box. More, specifically, judgment and consideration is required at each stage of the process.

The four commonly used matching algorithms are NNM, caliper and radius matching, kernel and local linear matching and stratification matching. In this paper, nearest neighbor matching and kernel-based matching (KBM) methods are employed.

DATA AND DESCRIPTION OF VARIABLES

Using a structured questionnaire, data was collected through a field survey of 950 farmers in Pakistan. In the first stage, all four provinces of Pakistan were selected. In the second stage, about 350 farmers in the Punjab province, 250 in KPK and Sindh provinces and 100 farmers in Baluchistan province were randomly selected. Socioeconomic, farm and household information were collected. A large number of questions were included regarding the water availability and shortage of water.

The description of variables is presented in Table 1. The mean age of the farmers was 43 years and the mean education level was about 9 years of schooling, indicating that farmers in the study area are quite educated. The experience of the farming community was about 14 years. About 92% were local farmers and the other 8% were migrant farmers. About 74% of the farmers owned the land with the remaining 26% being tenant farmers.

Table 1

Data and description of variables

VariableDescriptionMeanStd dev.
Age of farmers Age of the farmer in number of years 43 11.6 
Years of schooling of the household head Education of the farmer in number of years 8.71 5.42 
Farming experience Farming experience in number of years 14.44 7.83 
Local resident 1 if the farmer is local, 0 migrant 0.92 0.41 
Water scarcity Land ownership 1 if the household has faced irrigation water scarcity, 0 otherwise 0.57 0.32 
1 if the farmer is owner of land, 0 otherwise 0.74 0.55 
Male head 1 if the farmer is head, 0 otherwise 0.89 0.39 
Good soil quality 1 if the soil is of good quality, 0 otherwise 0.65 0.48 
Fragmented land 1 if the land is fragmented, 0 otherwise 0.67 0.35 
Slope: plain land 1 if the slope is same, 0 otherwise 0.85 0.51 
Land leveling 1 if the farmer has practiced land leveling, 0 otherwise 0.23 0.43 
Legumes crop rotation 1 if the farmer has included legumes in crop rotation and 0 otherwise 0.08 0.15 
Land size (hectare) Land owned by the farmer in number of hectares 2.65 1.42 
Male head 1 if the head is male, 0 otherwise 0.96 0.75 
Household size Number of family members in the household 10.13 5.12 
Joint family 1 if living in joint family, 0 otherwise 0.68 0.34 
Access to metal road 1 if the household has access to metal road, 0 otherwise 0.48 0.57 
Owns a tractor 1 if the farmer owns a tractor, 0 otherwise 0.09 0.13 
Owns a tube well 1 if the farmer owns a tube well, 0 otherwise 0.07 0.27 
Owns a car 1 if the farmer owns a car, 0 otherwise 0.19 0.24 
Livestock assets Number of livestock owned by the farmer 7.35 5.03 
Seed source 1 if home seed is used, 0 otherwise 0.66 0.32 
Access to credit 1 if the household has access to credit facility, 0 otherwise 0.06 0.28 
Access to extension 1 if the farmer has contact with extension services, 0 otherwise 0.26 0.15 
Income (Pak Rupee) Per month household income from all the sources 42,165 12,560 
Expenditure (Pak Rupee) Per month household expenditure in rupees 36,906 25,783 
Membership in organizations 1 if farmer is member of any organization, 0 otherwise 0.16 0.12 
Punjab Province 1 if the farmer is from Punjab, 0 otherwise 0.35 0.20 
Sindh Province 1 if the farmer is from Sindh, 0 otherwise 0.25 0.18 
KPK Province 1 if the farmer is from KPK, 0 otherwise 0.25 0.18 
Baluchistan Province 1 if the farmer is from Baluchistan, 0 otherwise 0.10 0.06 
VariableDescriptionMeanStd dev.
Age of farmers Age of the farmer in number of years 43 11.6 
Years of schooling of the household head Education of the farmer in number of years 8.71 5.42 
Farming experience Farming experience in number of years 14.44 7.83 
Local resident 1 if the farmer is local, 0 migrant 0.92 0.41 
Water scarcity Land ownership 1 if the household has faced irrigation water scarcity, 0 otherwise 0.57 0.32 
1 if the farmer is owner of land, 0 otherwise 0.74 0.55 
Male head 1 if the farmer is head, 0 otherwise 0.89 0.39 
Good soil quality 1 if the soil is of good quality, 0 otherwise 0.65 0.48 
Fragmented land 1 if the land is fragmented, 0 otherwise 0.67 0.35 
Slope: plain land 1 if the slope is same, 0 otherwise 0.85 0.51 
Land leveling 1 if the farmer has practiced land leveling, 0 otherwise 0.23 0.43 
Legumes crop rotation 1 if the farmer has included legumes in crop rotation and 0 otherwise 0.08 0.15 
Land size (hectare) Land owned by the farmer in number of hectares 2.65 1.42 
Male head 1 if the head is male, 0 otherwise 0.96 0.75 
Household size Number of family members in the household 10.13 5.12 
Joint family 1 if living in joint family, 0 otherwise 0.68 0.34 
Access to metal road 1 if the household has access to metal road, 0 otherwise 0.48 0.57 
Owns a tractor 1 if the farmer owns a tractor, 0 otherwise 0.09 0.13 
Owns a tube well 1 if the farmer owns a tube well, 0 otherwise 0.07 0.27 
Owns a car 1 if the farmer owns a car, 0 otherwise 0.19 0.24 
Livestock assets Number of livestock owned by the farmer 7.35 5.03 
Seed source 1 if home seed is used, 0 otherwise 0.66 0.32 
Access to credit 1 if the household has access to credit facility, 0 otherwise 0.06 0.28 
Access to extension 1 if the farmer has contact with extension services, 0 otherwise 0.26 0.15 
Income (Pak Rupee) Per month household income from all the sources 42,165 12,560 
Expenditure (Pak Rupee) Per month household expenditure in rupees 36,906 25,783 
Membership in organizations 1 if farmer is member of any organization, 0 otherwise 0.16 0.12 
Punjab Province 1 if the farmer is from Punjab, 0 otherwise 0.35 0.20 
Sindh Province 1 if the farmer is from Sindh, 0 otherwise 0.25 0.18 
KPK Province 1 if the farmer is from KPK, 0 otherwise 0.25 0.18 
Baluchistan Province 1 if the farmer is from Baluchistan, 0 otherwise 0.10 0.06 

About 89% of the respondents were the head of the households and the rest were related to the household head. Approximately 65% of the farmers have good quality soil while the rest have poor quality soil. About two-thirds of the farmers (67%) have fragmented land while 33% of the farmers have a consolidated land parcel. An overwhelming majority (85%) of the farmers have level land and the rest farm steep slopes. About 23% of the farmers have practiced land leveling while the majority of farmers (77%) have not. Only 8% of the farmers have cultivated legumes to increase the soil fertility, while the great majority (92%) do not. The mean land holding of the farmers was about 2.65 hectares. In the majority of cases (96%), the farmers are males and are heads of the households, while the rest are tenants. The average family size was about 10 family members per household. The majority of rural households (68%) are living in the joint family system. About 48% of rural households have access to a metal road, while 52% do not. Few households have a tractor or own a tube well or a car. The average livestock ownership is about seven per household. About two-thirds of the farmers use home seed while the rest use seed from other sources, including fellow farmers and dealers, etc. Only 6% of the farmers have access to a credit facility, while 94% do not. About 26% of the farmers have access to extension services, while 74% do not. The mean annual income of a household is 42,165 rupees and the average household expenditure is about 36,905 rupees.

Changes in the climate can have a significant impact on crop yields. A great majority (94%) of the respondents expressed the view that they have observed a change in climatic conditions over time; only 6% were of the view that they have not observed any change in the climatic conditions over time. Similarly, more than 90% of the respondents identified a change in rainfall, temperature, rainfall timing and monsoon over time (10 years), while less than 10% stated that they had not observed any change in rainfall, temperature, rainfall timing and the monsoon in the last 10 years. The data in Table 1 reveal that more than half (57%) of the sampled farmers had adjusted the sowing time of wheat according to the climatic conditions while the rest (43%) of the sample respondents had not adjusted the wheat sowing time according to the change in the climatic conditions.

Regarding the adoption of heat/stress-tolerant varieties, 91% of the respondents stated that they had not adopted heat/stress-tolerant varieties, while the remaining 9% of the respondents have adopted heat/stress-tolerant varieties. Similarly, 93% of the respondents indicated that they had not adopted new crops/did not plant some crops due to climatic conditions, i.e. only 7% of the respondents have adopted new crops/did not plant some crops in response to climatic conditions.

IRRIGATION SOURCES USED BY THE FARMERS

In the study area, the most popular source of irrigation water are the canals (72%) followed by tube wells (24%) and wells (3%), with only 1% obtained from other sources.

In the study area, the canal water is not sufficient to meet the irrigation requirements as 64% of the farmers face a water-scarcity problem during the season. The farmers turn to alternate source of irrigation water, i.e. tube wells; however, due to load-shedding problems, 74% of the farmers are not able to get the full benefit from the tube wells. Around 87% of the tube wells are powered by electricity while about 18% of the farmers use diesel tube wells for irrigation purposes which is very costly and cannot be afforded by the majority of farmers in Pakistan who are smallholder farmers.

EMPIRICAL RESULTS

Determinants of water scarcity

The matching process is preceded by specification of the propensity scores for the treatment variable. A logit model was employed to predict the probability of irrigation water scarcity among the farming community in Pakistan. The dependent variable is a dummy, i.e. 1 if the farmer faces water scarcity, and 0 otherwise. A set of independent variables was also included in the model (refer to Table 2). The age coefficient is positive and significant at the 1% level of significance, indicating that mostly the older farmers face water-scarcity problems. The education is negative and significant at the 5% level of significance, indicating that more less-educated farmers are facing water-scarcity problems. Similarly, the household head coefficient is negative and significant at a 10% level of significance, indicating that if the farmer himself is head of the household then he faces fewer water-scarcity problems.

Table 2

Determinants of the water scarcity (logit estimates)

VariableCoefficientt-value
Age of farmer 0.04*** 2.81 
Education of famer −0.01** 2.05 
Land ownershipa,b −0.02** −1.98 
If farmer is household heada,c −0.01* −1.67 
Land fragmentationa,d 0.01** 2.19 
Slopea,e 0.01 1.45 
Land levelinga,f −0.01** 2.10 
Land holding (hectare) −0.02** −2.30 
Family size 0.03*** 2.76 
Joint familya,g −0.01** −2.54 
Metal roada,h 0.01 1.33 
Tractora,i 0.01 1.41 
Tube wella,j −0.01*** 2.75 
Cara,k −0.01** 2.37 
Livestock 0.02*** 2.58 
Credita,l −0.01** −2.06 
Extensiona,m −0.03** −2.15 
Membershipa,n −0.01*** −2.27 
Punjaba,o −0.02** −2.10 
Sindha,o −0.05* 1.83 
KPKa,o 0.01 1.12 
Value of R-square 0.21 
LR-chi square 198.34 
Prob > chi square 0.000 
Number of observations 950 
VariableCoefficientt-value
Age of farmer 0.04*** 2.81 
Education of famer −0.01** 2.05 
Land ownershipa,b −0.02** −1.98 
If farmer is household heada,c −0.01* −1.67 
Land fragmentationa,d 0.01** 2.19 
Slopea,e 0.01 1.45 
Land levelinga,f −0.01** 2.10 
Land holding (hectare) −0.02** −2.30 
Family size 0.03*** 2.76 
Joint familya,g −0.01** −2.54 
Metal roada,h 0.01 1.33 
Tractora,i 0.01 1.41 
Tube wella,j −0.01*** 2.75 
Cara,k −0.01** 2.37 
Livestock 0.02*** 2.58 
Credita,l −0.01** −2.06 
Extensiona,m −0.03** −2.15 
Membershipa,n −0.01*** −2.27 
Punjaba,o −0.02** −2.10 
Sindha,o −0.05* 1.83 
KPKa,o 0.01 1.12 
Value of R-square 0.21 
LR-chi square 198.34 
Prob > chi square 0.000 
Number of observations 950 

Standard errors corrected in parentheses; the district dummy has been used to control for the locational effect.

***, **, and * indicate significance at the 1, 5, and 10% level, respectively.

aDummy variables.

bExcluded category: tenant.

cExcluded category: farmer not head.

dExcluded category: non-fragmented land.

eExcluded category: slope land.

fExcluded category: no land leveling practice.

gExcluded category: nuclear family.

hExcluded category: lack of access to metal road.

iExcluded category: no tractor ownership.

jExcluded category: no tube well ownership.

kExcluded category: no car ownership.

lExcluded category: no access to credit.

mExcluded category: no access to extension service.

nExcluded category: no membership to any organization.

oExcluded category: Baluchistan.

Land fragmentation was included as a dummy variable and the coefficient is positive and significant indicating that farmers with the fragmented land holding normally face greater water-scarcity issues. The land slope coefficient is positive although non-significant. The land leveling and the land holding size coefficients are negative and significant, indicating that farmers who have leveled the land and have higher land holdings normally face fewer water-scarcity problems.

The family size coefficient is positive and highly significant at the 1% level of significance, indicating that farmers with a larger family size normally face greater water-scarcity problems. The joint family was included as dummy variable and the coefficient is negative and significant, indicating that farmers living in a joint family normally face fewer water-scarcity problems.

The metal road and tractor ownership coefficients are positive and non-significant. Most importantly the tube well ownership coefficient is negative and highly significant at the 1% level of significance, indicating that farmers who own tube wells normally face fewer water-scarcity problems. The car ownership coefficient is negative and significant, while the number of livestock owned is positive and significant.

The access to a credit facility, extension services and organization membership are all negative and significant, indicating that farmers having this support normally when facing fewer water-scarcity problems. The provincial dummies were also included in the model.

The value of R-square is 21% and LR-chi square is highly significant at the 1% level of significance, indicating the robustness of the variables included in the model.

Impact of water scarcity on cereal crop yields, household income and poverty levels

The impact of water scarcity on cereals crop yields, household income and poverty level was estimated by employing the PSM approach. In the current analysis, two different matching algorithms i.e. NNM and KBM are employed. In the case of propensity score matching, the most important parameter of interest is the ATT, i.e. the difference in the outcome of the farmers facing irrigation water scarcity and those farmers not facing the irrigation water scarcity with the similar propensity score. The caliper for NNM and the bandwidth for KBM are reported in Table 3. The ATT for wheat yield is negative, indicating that farmers facing water scarcity have lower wheat yields: in the range of 20–29 kg both in NNM and KBM, respectively. Similarly the rice yields are lower: in the range of 38–43 kg both in NNM and KBM, respectively. The maize yields are lower, in the range of 15–17 kg both in NNM and KBM. These results have important policy implications, namely the yields of the cereal crops are significantly lower on farms which experience water scarcity compared with similar farmers facing no water-scarcity problems.

Table 3

Impact of water scarcity on cereal crop yields, household income and poverty levels

Matching algorithmOutcomeCaliper/bandwidthATTt-valueCritical level of hidden biasNumber of treatedNumber of control
NNM Wheat yield 0.001 −20.31*** −3.17 1.20–1.25 356 451 
Rice yield 0.05 −38.57** −2.04 1.45–1.50 399 473 
Maize yield 0.01 −15.25** −1.98 1.15–1.20 362 485 
Income 0.003 −8,032* −1.75 1.80–1.85 378 494 
Food security 0.005 −0.07* −1.67 1.30–1.35 316 426 
Poverty 0.06 0.05* 1.73 1.25–1.30 322 491 
KBM Wheat yield 0.1 −29.60*** −2.63 1.35–1.40 379 478 
Rice yield 0.04 −43.18*** −2.58 1.15–1.20 352 444 
Maize yield 0.05 −17.24*** −3.04 1.50–1.55 385 460 
Income 0.001 −10,741*** −3.41 1.25–1.30 401 452 
Food security 0.002 −0.12** −2.15 1.40–1.45 383 471 
Poverty 0.003 0.07** 2.32 1.35–1.40 370 453 
Matching algorithmOutcomeCaliper/bandwidthATTt-valueCritical level of hidden biasNumber of treatedNumber of control
NNM Wheat yield 0.001 −20.31*** −3.17 1.20–1.25 356 451 
Rice yield 0.05 −38.57** −2.04 1.45–1.50 399 473 
Maize yield 0.01 −15.25** −1.98 1.15–1.20 362 485 
Income 0.003 −8,032* −1.75 1.80–1.85 378 494 
Food security 0.005 −0.07* −1.67 1.30–1.35 316 426 
Poverty 0.06 0.05* 1.73 1.25–1.30 322 491 
KBM Wheat yield 0.1 −29.60*** −2.63 1.35–1.40 379 478 
Rice yield 0.04 −43.18*** −2.58 1.15–1.20 352 444 
Maize yield 0.05 −17.24*** −3.04 1.50–1.55 385 460 
Income 0.001 −10,741*** −3.41 1.25–1.30 401 452 
Food security 0.002 −0.12** −2.15 1.40–1.45 383 471 
Poverty 0.003 0.07** 2.32 1.35–1.40 370 453 

NNM stands for nearest-neighbor matching.

KBM stands for kernel-based matching.

ATT stands for the average treatment effect for the treated.

***,** and * results are significant at the 1%, 5% and 10% levels, respectively.

Water scarcity lowers household incomes in the range of 8,032–10,741 Pakistani rupees both in NNM and KBM. Due to water scarcity, household food security levels are badly affected and the households have less food security (in the range of 7–12%) compared with households having no water-scarcity problems. These results align with the results of previous studies, e.g. Hussain & Hanjra 2004.

The important finding of the current study is the impact on poverty levels. The household head count index of poverty was estimated and included as an outcome variable in the PSM.

The ATT results indicate that water scarcity increases poverty levels by 5–7%. These results align with the results of previous studies, e.g. Ali & Abdulai (2010).

From the empirical findings, it can be concluded that access to irrigation water is affecting cereal crop yields, household food security, income and poverty levels. Significant investment is needed from both the public and private sector if the farming community is to have access to irrigation water to sustain crop yields, household income and food security levels. Otherwise the efforts to reduce poverty levels will be negated. The critical level of hidden bias is also included and it indicates the level up to which the farmers facing water scarcity and farmers not facing water scarcity differ in their odds of having access to irrigation water. The number of treated and the number of control are reported in Table 3.

The main purpose of the propensity score matching is to balance the covariates before and after matching. For that purpose a number of balancing tests are employed, such as the median absolute bias before and after matching, the value of R-square before and after matching and the P-value of joint significance of covariates before and after matching. The results of the matching quality are presented in Table 4. The median absolute bias before matching is quite high and is in the range of 18–23% both in NNM and KBM. After matching a considerable amount of bias has been reduced with the bias in the range of 3–7%. The percentage bias reduction is in the range of 64–81%.

Table 4

Indicators of covariates balancing before and after matching

Matching algorithmOutcomeMedian absolute bias before matchingMedian absolute bias after matchingPercentage bias reductionValue of R-square before matchingValue of R-square after matchingP-value of LR-chi square before matchingP-value of LR-chi square after matching
NNM Wheat yield 18.24 4.23 76.8 0.264 0.002 0.003 0.278 
Rice yield 19.27 5.07 73.6 0.257 0.003 0.002 0.351 
Maize yield 20.18 3.87 80.8 0.234 0.004 0.001 0.443 
Income 20.15 4.71 76.6 0.231 0.003 0.002 0.517 
Food security 21.42 3.98 81.4 0.280 0.002 0.003 0.239 
Poverty 22.54 5.73 74.5 0.432 0.003 0.002 0.361 
KBM Wheat yield 20.36 7.20 64.6 0.591 0.001 0.001 0.453 
Rice yield 21.73 6.54 69.9 0.783 0.004 0.003 0.481 
Maize yield 20.75 4.51 78.2 0.591 0.005 0.002 0.493 
Income 21.67 4.39 79.7 0.842 0.002 0.001 0.260 
Food security 20.43 5.38 73.6 0.649 0.003 0.002 0.482 
Poverty 23.51 6.04 74.3 0.233 0.001 0.003 0.431 
Matching algorithmOutcomeMedian absolute bias before matchingMedian absolute bias after matchingPercentage bias reductionValue of R-square before matchingValue of R-square after matchingP-value of LR-chi square before matchingP-value of LR-chi square after matching
NNM Wheat yield 18.24 4.23 76.8 0.264 0.002 0.003 0.278 
Rice yield 19.27 5.07 73.6 0.257 0.003 0.002 0.351 
Maize yield 20.18 3.87 80.8 0.234 0.004 0.001 0.443 
Income 20.15 4.71 76.6 0.231 0.003 0.002 0.517 
Food security 21.42 3.98 81.4 0.280 0.002 0.003 0.239 
Poverty 22.54 5.73 74.5 0.432 0.003 0.002 0.361 
KBM Wheat yield 20.36 7.20 64.6 0.591 0.001 0.001 0.453 
Rice yield 21.73 6.54 69.9 0.783 0.004 0.003 0.481 
Maize yield 20.75 4.51 78.2 0.591 0.005 0.002 0.493 
Income 21.67 4.39 79.7 0.842 0.002 0.001 0.260 
Food security 20.43 5.38 73.6 0.649 0.003 0.002 0.482 
Poverty 23.51 6.04 74.3 0.233 0.001 0.003 0.431 

The value of R-square is quite high before matching and is quite low after matching indicating that after matching the covariates have been balanced and there are no systematic differences between the farmers facing irrigation water scarcity and those not facing irrigation water scarcity. Similarly the P-value of the joint significance of covariates is also presented in Table 4. The P-value is significant before matching, indicating that before matching both the groups are quite different from each other, while the P-value is non-significant after matching, indicating that after matching both are quite similar to each other. The results align with the results of previous studies, e.g. Ali & Abdulai (2010). Figure 1 indicates the imposition of the common support condition and the balancing of the covariates.
Figure 1

PSM (imposition of the common support condition).

Figure 1

PSM (imposition of the common support condition).

CONCLUSIONS

Irrigation water scarcity is becoming a serious issue in Pakistan with many farmers facing water-scarcity problems. The empirical results of this study indicate that water scarcity reduces the yield of cereal crops which in turn has decreased the food security levels of farming households by 7–12%. Similarly, the household income levels are reduced by around 8,000–10,000 Pakistani rupees. Due to water scarcity the poverty levels are increased by 4–5%. It is concluded that investment is needed from the public and private sector if the water losses are to be decreased and farmers are to have access to irrigation water at the time of need to sustain their crop yields in order to maintain food security levels and reduce poverty in the rural areas of Pakistan.

REFERENCES

REFERENCES
Ashraf
S.
2013
Pakistan's water shortage drips towards disaster. The National. http://www.thenational.ae/thenationalconversation/comment/pakistans-water-shortage-drips-towardsdisaster (accessed 23 December 2015)
.
Barker
R.
Koppen
B. V.
1999
Water Scarcity and Poverty. International Water Management Institute (IWMI), Colombo, Sri Lanka
.
Cruz
R. V.
Harasawa
H.
Lal
M.
Wu
S.
Anokhin
Y.
Punsalmaa
B.
Honda
Y.
Jafari
M.
Li
C.
Huu
N.
,
2007
Asia. Climate change, 2007, impacts, adaptation and vulnerability
. In:
Parry
M. L.
Canziani
O. F.
Palutikof
J. P.
et al
(eds),
Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change
.
Cambridge University Press
,
Cambridge
, pp
469
506
.
Dehejia
R. H.
Wahba
S.
2002
Propensity score matching methods for non experimental causal studies
.
The Rev. Econo. Stat.
84
(
1
),
151
161
.
Government of Pakistan
2003
Establishment Division, Islamabad, Pakistan
.
Heckman
J. J.
Ichimura
H.
Todd
P. E.
1997
Matching as an econometric evaluation estimator: evidence from evaluating a job training programme
.
Review of Economic Studies
64
,
605
654
.
Imbens
G. W.
Wooldridge
J. M.
2009
Recent developments in the econometrics of program evaluations
.
Journal of Econometric Literature
47
,
5
86
.
Jalan
J.
Ravallion
M.
2003
Does piped water reduce diarrhea for children in rural India?
J. Econometrics
112
(
1
),
153
173
.
Molden
D.
Oweis
T.
Steduto
P.
Kijne
J. W.
Hanjra
M. A.
Bindraban
P. S.
2007
Pathways for increasing agricultural water productivity
. In:
Molden
D.
(ed.),
Water for Food, Water for Life: a Comprehensive Assessment of Water Management in Agriculture
.
Earthscan
,
London, UK
and International Water Management Institute (IWMI), Colombo, Sri Lanka
, pp.
219
310
.
Qureshi
A. S.
Shah
T.
Akhtar
M.
2003
The groundwater economy of Pakistan. IWMI working paper no. 64, International Water Management Institute, Colombo, Sri Lanka, p. 23
.
Shah
T.
1993
Groundwater Markets and Irrigation Development. Political Economy and Practical Policy
.
Oxford University Press
,
Delhi, India
.
Sivakumar
M. V. K.
Stefanski
R.
2011
Climate Change in South Asia. In: R. Lal, M. V. K. Sivakumar, S. M. A. Faiz, A. H. M. M. Rahman & K. R. Islam (eds), Climate Change and Food Security in South Asia. Springer Science + Business Media B.V
, pp. 13–28.
Smith
J.
Todd
P.
2005
Does matching overcome Lalonde's critique of non-experimental estimators?
J. of Econo.
125
(
1–2
),
305
353
.
http://www.bsos.umd.edu/econ/jsmith/papers
.
World Bank
2013
Data: Climate Change. Retrieved September 24, 2013. http://data.worldbank.org/topic/climate-change.
Zhao
Z.
2000
Using Matching to Estimate Treatment Effects: Data Requirements, Sensitivity, and an Application.