## INTRODUCTION

Safe drinking water is essential for the survival and maintenance of good health of people. Providing water security plays a major role in poverty reduction and livelihood improvement, and water scarcity and/or contamination has adverse effects on fertility and migration patterns (Falkenmark 1990) and human health (Dungumaro 2007). When provision and availability of drinking water become inadequate, people are forced to use contaminated and unsafe water sources, resulting in water-related diseases that cause loss of productive working hours and an increase in health expenditure. For instance, (Satterthwaite 2003) finds that inadequacies in provision for piped water, sanitation and drainage result in problems with insect-borne diseases such as malaria, and other diseases related to lack of water and use of poor-quality water. Poverty alleviation and access to safe and sufficient water are positively related, because provision of adequate water resources helps in achieving food security and improving human development indicators, including human health. Poor households frequently do not have access to quality water both in terms of sources of water and methods of water treatment, which makes them vulnerable to a variety of water-borne diseases, thereby adversely reducing their bare-minimum earnings and increasing their expenditure on medicines (Wang et al. 2005).

In poor countries, lack of availability of convenient and easily accessible water resources forces households to travel long distances and often means that children are employed to collect water, adversely affecting productive activities of adults and education of children (Mehta 2014). In addition, poorer households may use water from open sources, often contaminated, which makes these people more vulnerable to water-borne diseases. According to WHO (2010), by the end of 2010 about 89% of the world's population (6.1 billion people) used improved drinking water sources, a figure which is even higher by 1% than the Millennium Development Goals (MDG) target of 88%. However, this means that approximately 11% of the world's population (783 million people) still does not have access to safe drinking water. Hence the expansion of access to safe and reliable drinking water sources, especially in the continents of Africa and Asia, is one of the priorities of the Millennium Development Goals program.

In this context, issues of household socioeconomic conditions as they relate to water availability are not very well covered in water literature, particularly in the context of Bhutan. In Bhutan, about 69% of the country's population lives in rural areas, where most of the water-borne and water-related diseases such as diarrhoea, typhoid, skin infections, conjunctivitis, dengue and malaria are prevalent. These diseases are among the leading causes, of infant and child deaths in the country. In most cases, research tends to focus on factors leading to an inadequate supply of water, and not on exploring factors which hinder or enable households' access to water. It is argued that in certain instances, failure to obtain water from a safe source is caused by the inability to pay for it. Understanding household socioeconomic conditions offers a way of linking availability of water and ability of households to obtain water from a safe source. In this context, analysis of household-level determinants of access to drinking water sources is an important tool for gaining a comprehensive understanding of the behaviour that shapes a household's use of safe drinking water.

In order to contribute to this body of knowledge, the present study makes an attempt to examine the patterns of access to water, and to identify and analyze the factors that influence households' access safe drinking water sources, analyze factors determining the extent of households' travel to fetch drinking water, and assess the effects of access to safe drinking water to human health in Bhutan, using the data from the Bhutan Living Standard Survey 2012. The paper is organized as follows. The second section presents the propensity score matching (PSM) approach as a conceptual framework for assessing the effects of safe access to water on human health. Data, methods and variable description are presented in the third section. The fourth section presents the empirical results and discussions. Finally, the conclusion and policy implication are presented in the fifth section.

## THE PSM APPROACH: A CONCEPTUAL FRAMEWORK FOR LINKING HOUSEHOLD ACCESS TO DRINKING WATER AND HUMAN HEALTH

In this section, the PSM approach is described as a simple conceptual framework for linking the effects of drinking water with human health. It is assumed that there are two categories of rural households in Bhutan, i.e. households having access to safe drinking water and the households having no access to safe drinking water, and the households having access to safe drinking water have higher utility levels as compared to households having no access to safe drinking water, as presented in Equation (1).
1
The rural households' access to safe drinking water is influenced by number of socioeconomic (), demographic (), institutional (ξ) household and farm level () factors as represented in Equation (2).
2

On the contrary, if it is assumed that the characteristics of these two sets of households, with and without access to safe drinking water, are similar, then it would be incorrect, as in the real world the socioeconomic and other characteristics of beneficiary and non-beneficiary households tend to be quite different from each other in many aspects. If the households are quite different from each other than the comparison and estimates may produce misleading results. In order to correct this sample selection bias, PSM approach has been employed. Matching involves pairing treatment and comparison units that are similar in terms of their observable characteristics. When the relevant differences between any two units are captured in the observable covariates, which occurs when outcomes are independent of assignment to treatment conditional on pretreatment covariates, matching methods can yield unbiased estimates of the treatment impact (Dehejia & Wahba 2002).

PSM follows that the expected treatment effect for the treated population is of primary significance. This effect may be given as:
3
where is the average treatment effect for the treated (ATT), denotes the value of the outcome for adopters of the new technology, in our case households with access to safe drinking water and is the value of same variable for non-adopters, in our case households without safe drinking water. As noted above, a major problem is that we do not observe . Although the difference can be estimated, it is potentially a biased estimator.
In the absence of experimental data, the propensity score-matching model (PSM) can be employed to account for this sample selection bias (Dehejia & Wahba 2002). The PSM is defined as the conditional probability that a household has access to safe drinking water, given pre-access characteristics (Rosenbaum & Rubin 1983). To create the condition of a randomized experiment, the PSM employs the unconfoundedness assumption, also known as conditional independence assumption (CIA), which implies that once Z is controlled for, technology adoption is random and uncorrelated with the outcome variables. The PSM can be expressed as:
4
where I is the indicator for access to safe drinking water and Z is the vector of pre-access characteristics. The conditional distribution of Z, given p(Z) is similar in both groups of households having access to safe drinking water and those that do not have access to safe drinking water.
After estimating the propensity scores, the average treatment effect for the treated (ATT) can then be estimated as:
5

PSM rests on two strong assumptions, i.e. CIA and the common support condition. The CIA states that once the observable factors are controlled for, access to safe drinking water is random and uncorrelated with the outcome variables. The common support condition states that matching can only be performed over the region of the common support. There are a number of matching algorithms which can be employed to estimate the PSM, i.e. nearest neighbour matching (NNM), kernel based matching (KBM), radius matching (RM) and mahalanobis metric matching (MMM). As the main purpose of the PSM is to balance the covariates before and after matching, a number of balancing tests have been employed in the current analysis such as a reduction in the median absolute bias before and after matching, the value of before and after matching and the p-value of joint significance of covariates before and after matching.

## DATA, METHODS AND VARIABLE DESCRIPTIONS

Table 1

Percentage of householdS by sources of water during 2012, 2007 and 2003

Water sourcesRural
Urban
201220072004201220072004
Tap in dwelling 72.8 44.8 43.66 87.2 82.9 78.14
Public & neighbour 23.2 40.9 31.81 11.8 16.5 20.66
Well 1.1 1.7 4.27 0.4 0.2 0.17
Natural sources 2.8 12.6 20.26 0.5 0.5 1.03
Total 4.350 6.856 1.688 4.619 2.942 2.319
Water sourcesRural
Urban
201220072004201220072004
Tap in dwelling 72.8 44.8 43.66 87.2 82.9 78.14
Public & neighbour 23.2 40.9 31.81 11.8 16.5 20.66
Well 1.1 1.7 4.27 0.4 0.2 0.17
Natural sources 2.8 12.6 20.26 0.5 0.5 1.03
Total 4.350 6.856 1.688 4.619 2.942 2.319

Source: Compiled from BLSS 2003, 2007 and 2012.

The objectives of the paper are to identify and analyze the factors that influence a household's ability and/or inability to access safe drinking water sources, and analyze factors determining household travel to fetch drinking water. To identify and analyze the factors that determine a household's ability to access safe drinking water, a logistic regression model has been employed, whereas for analyzing factors affecting household travel to collect drinking water, the censored least absolute deviation (CLAD) model has been applied. Based on the available literature on factors influencing household access to drinking water and our own understanding of the local household behaviour, it is assumed that various household socioeconomic and demographic characteristics may also influence household access to drinking water in Bhutan. These factors include age and gender of the household head, household size, years of schooling, landholding size, wealth of household, access to markets and others. Table 2 presents the descriptions and summery statistics of the variables used in the models mentioned above.

Table 2

Variable descriptions and summery statistics

VariableDescriptionMeanStd dev.
Age of household head Age of the farmer in number of years 49.13 15.33
Gender of household head 1 if the gender of the household head is male and 0 for female 0.66 0.37
Number of children under 15 years Children under the age of 15 in the household 1.48 1.35
Elderly member in the households Total adult in the household over the age of 65 0.34 0.62
Family size (household size) Total number of family members living in the household 4.8 2.2
Education of household head Number of years of schooling of the household head 1.83 3.94
Distance to food market/shop Distance to the nearest food market/shop in h 1.61 5.7
Distance to water source Distance of the safe water in min 9.36 21.64
Water access  1 if the household have access to safe drinking water and 0 otherwise 0.73 0.39
Operational land Total operational land in acres 1.28 2.08
Livestock assets  Livestock assets owned by the household 1.87 1.05
Mobile owned 1 if the household have mobile facility and 0 otherwise 0.91 0.08
Internet owned 1 if the household have access to internet facility and 0 otherwise 0.05 0.09
Toilet facility 1 if the household have improved toilet facility and 0 otherwise 0.81 0.03
Mean per-capita monthly income Mean per-capita monthly household income in ngultrum 2.319 1.562
Mean per-capita expenditure Mean per-capita monthly household food expenditure in ngultrum 1.399 1.142
Number of observations 4,173
VariableDescriptionMeanStd dev.
Age of household head Age of the farmer in number of years 49.13 15.33
Gender of household head 1 if the gender of the household head is male and 0 for female 0.66 0.37
Number of children under 15 years Children under the age of 15 in the household 1.48 1.35
Elderly member in the households Total adult in the household over the age of 65 0.34 0.62
Family size (household size) Total number of family members living in the household 4.8 2.2
Education of household head Number of years of schooling of the household head 1.83 3.94
Distance to food market/shop Distance to the nearest food market/shop in h 1.61 5.7
Distance to water source Distance of the safe water in min 9.36 21.64
Water access  1 if the household have access to safe drinking water and 0 otherwise 0.73 0.39
Operational land Total operational land in acres 1.28 2.08
Livestock assets  Livestock assets owned by the household 1.87 1.05
Mobile owned 1 if the household have mobile facility and 0 otherwise 0.91 0.08
Internet owned 1 if the household have access to internet facility and 0 otherwise 0.05 0.09
Toilet facility 1 if the household have improved toilet facility and 0 otherwise 0.81 0.03
Mean per-capita monthly income Mean per-capita monthly household income in ngultrum 2.319 1.562
Mean per-capita expenditure Mean per-capita monthly household food expenditure in ngultrum 1.399 1.142
Number of observations 4,173

Household, income and expenditure are expressed in Bhutanese ngultrum or nu (exchange rate: 1 US$= 53.44 nu at time of study; purchasing power parity rate: 1 US$ = 18.203 nu).

Table 3

Differences in the key characteristics of the households having access and having no-access to safe drinking water

VariableWater accessNo water accessDifferencet-values
Age of household head 45.30 51.79 −6.49** −2.15
Gender of household head 0.72 0.61 0.11*** 2.94
Number of children under 15 years 1.40 1.56 −0.16 −1.27
Old age member in the households 0.37 0.32 0.05 0.80
Family size (household size) 4.27 5.32 −0.61* −1.82
Education of household head 1.97 1.65 1.32** 2.07
Distance to food market/shop 1.73 1.51 −0.222*** −2.75
Distance to road 0.71 0.84 −0.07* −1.86
Distance to water source 10.82 6.20 −4.62* −1.66
Water access 0.72 0.61 0.11* 1.90
Operational land 1.52 1.00 0.52** 2.24
Livestock 1.87 1.62 0.25* 1.82
Mobile 0.93 0.85 0.08* 1.85
Internet 0.07 0.03 0.4* 1.74
Improved toilet facility 0.88 0.78 0.10** 1.83
Mean per-capita monthly income 2.519 2.013 −506** −2.35
Mean per-capita expenditure 1.575 1.205 −370* 1.80
VariableWater accessNo water accessDifferencet-values
Age of household head 45.30 51.79 −6.49** −2.15
Gender of household head 0.72 0.61 0.11*** 2.94
Number of children under 15 years 1.40 1.56 −0.16 −1.27
Old age member in the households 0.37 0.32 0.05 0.80
Family size (household size) 4.27 5.32 −0.61* −1.82
Education of household head 1.97 1.65 1.32** 2.07
Distance to food market/shop 1.73 1.51 −0.222*** −2.75
Distance to road 0.71 0.84 −0.07* −1.86
Distance to water source 10.82 6.20 −4.62* −1.66
Water access 0.72 0.61 0.11* 1.90
Operational land 1.52 1.00 0.52** 2.24
Livestock 1.87 1.62 0.25* 1.82
Mobile 0.93 0.85 0.08* 1.85
Internet 0.07 0.03 0.4* 1.74
Improved toilet facility 0.88 0.78 0.10** 1.83
Mean per-capita monthly income 2.519 2.013 −506** −2.35
Mean per-capita expenditure 1.575 1.205 −370* 1.80

Note: The results (***, **, *) indicate significance at 1, 5 and 10 percent levels, respectively.

## EMPIRICAL RESULTS AND DISCUSSION

Table 4

VariableCoefficientz-values
Age of household head −0.12*** −2.73
Gender of household heada,b 0.11** 2.08
Family size (household size) 0.13*** 4.10
Education of household head 0.12** 2.25
Distance to market (in min) −0.13** −2.14
Distance to road (in min) −0.24** −2.04
Distance to water source (in min) −0.10*** −2.76
Operational land (in acres) 0.16** 1.98
Livestock assets (TLU) 0.12** 2.04
Mobile owneda,c 0.04 1.13
Internet owneda,d 0.08* 1.92
Improved toilet facilitya,e 0.05 1.47̀
Per capita monthly household income 0.09** 2.03
Monthly per capita food expenditure 0.11** 2.16
Constant 0.27** 2.25
LR  135.27
Prob >  0.000
Pseudo  0.24
Number of observations 4.170
VariableCoefficientz-values
Age of household head −0.12*** −2.73
Gender of household heada,b 0.11** 2.08
Family size (household size) 0.13*** 4.10
Education of household head 0.12** 2.25
Distance to market (in min) −0.13** −2.14
Distance to road (in min) −0.24** −2.04
Distance to water source (in min) −0.10*** −2.76
Operational land (in acres) 0.16** 1.98
Livestock assets (TLU) 0.12** 2.04
Mobile owneda,c 0.04 1.13
Internet owneda,d 0.08* 1.92
Improved toilet facilitya,e 0.05 1.47̀
Per capita monthly household income 0.09** 2.03
Monthly per capita food expenditure 0.11** 2.16
Constant 0.27** 2.25
LR  135.27
Prob >  0.000
Pseudo  0.24
Number of observations 4.170

Note: The results (***, **, *) indicate significance at 1, 5 and 10% levels, respectively.

cDo not have mobile phone.

dDo not have internet.

eDo not have improved toilet.

### Factors affecting household travel to fetch drinking water

Table 5

Determinant factors affecting household travel to distance to access safe drinking water (Tobit estimates)

VariableCoefficientt-values
Age of household head −0.07** −2.32
Gender of the household heada,b −0.06* 1.69
Family size (household size) 0.08** 2.19
Education of household head 0.10*** 2.74
Distance to market −0.12*** −3.16
Operational land 0.11*** 2.71
Livestock asset 0.13** 2.12
Mobilea,c 0.14** 2.08
Interneta,d 0.15* 1.67
Improved toilet facilitya,e 0.03 1.40
Mean per-capita monthly income 0.09** 2.16
Monthly per capita food expenditure 0.05** 1.99
Constant 0.175*** 2.79
Pseudo  0.39
Sample size 4.170
Uncensored sample 3.061
VariableCoefficientt-values
Age of household head −0.07** −2.32
Gender of the household heada,b −0.06* 1.69
Family size (household size) 0.08** 2.19
Education of household head 0.10*** 2.74
Distance to market −0.12*** −3.16
Operational land 0.11*** 2.71
Livestock asset 0.13** 2.12
Mobilea,c 0.14** 2.08
Interneta,d 0.15* 1.67
Improved toilet facilitya,e 0.03 1.40
Mean per-capita monthly income 0.09** 2.16
Monthly per capita food expenditure 0.05** 1.99
Constant 0.175*** 2.79
Pseudo  0.39
Sample size 4.170
Uncensored sample 3.061

Note: The results (***, **, *) indicate significance at 1, 5 and 10% levels, respectively.

cDo not have mobile phone.

dDo not have internet.

eDo not have improved toilet.

The family size coefficient is positive and significant at 5% level indicating that large families normally can travel longer distances to have access to safe drinking water. This may be because people in large family is having less opportunity costs and hence can spend time to travel longer distance to fetch water. The coefficient of the variable education is positive and significant at 1% level, which suggests that educated households can travel longer distances to have access to safe drinking water. In this case, the awareness about the importance of safe drinking water dominates the opportunity costs that the educated household may face. The distance to market coefficient is negative and significant at 1% level of significance, indicating that the greater the distance to market, the less the households have access to safe drinking water.

The operational land is positive and significant at 1% level of significance. Similarly livestock ownership is positive and significant at 5% level indicating that wealthier households have easy access to safe drinking water. The mobile ownership and internet access were included as development indicators and the coefficients are positive and significant, which suggest that overall development is likely to generate demand for safe drinking water. The households' income coefficient is positive and significant at 5% level, indicating that higher income households are likely to travel long distances to fetch safe drinking water. Here, also access to safe drinking water dominates.

### Impact of safe drinking water on human health

The impact of safe drinking water on human health is estimated by employing the PSM. The matching method has been employed to correct for potential sample selection bias that may arise due to systematic differences between the households having access to safe drinking water and having no access to safe drinking water. The most important parameter of interest is ATT, i.e. difference in outcome of the treated and non-treated. In the current analysis, a number of different matching algorithms are employed, i.e. nearest neighbour matching (NNM), KBM, RM and MMM. The nearest neighbour matching matches with the nearest neighbour only, the KBM takes the weighted average of all the non-participants and then matches. In RM, each treated subject is matched with a corresponding control subject that is within a predefined interval of the treatment subject's propensity score. In MMM, the subjects are ordered randomly and then the distance between the treated and control subjects is calculated. The treatment and control are matched based on the smallest Mahalanobis distance. The process is repeated until each treatment subject is matched and then the unmatched control subjects are removed.

Table 6

Matching algorithmOutcomeATTt-valueCritical level of hidden biasNumber of treated (A)Number of control (B)
NNM Stomach disorder −0.10* −1.95 1.40–1.45 2.310 1.493
Dehydration −0.16*** −3.22 1.85–1.90 2.146 1.720
Medicine expenditure −248** −1.98 1.10–1.15 2.035 1.627
Skin diseases −0.02* −1.65 1.30–1.35 2.519 2.766
KBM Stomach disorder −0.09** −2.05 1.20–1.25 2.416 1.572
Dehydration −0.15** −2.34 1.35–1.40 2.239 1.725
Medicine expenditure −385** −2.10 1.25–1.30 2.462 1.763
Skin diseases −0.05** −2.07 1.30–1.35 2.031 1.467
RM Stomach disorder −0.12** −1.97 1.35–1.40 2.217 1.584
Dehydration −0.13*** −2.56 1.95–2.00 2.138 1.369
Medicine expenditure −415*** −3.03 1.50–1.55 2.046 1.433
Skin diseases −0.01 −1.42 – 2.418 1.671
MMM Stomach disorder −0.13** −2.18 1.45–1.50 2.064 1.329
Dehydration −0.10*** −3.27 1.10–1.15 2.316 1.472
Medicine expenditure −375*** −2.93 1.40–1.45 2.258 1.230
Skin diseases −0.03 −1.22 – 2.353 1.670
Matching algorithmOutcomeATTt-valueCritical level of hidden biasNumber of treated (A)Number of control (B)
NNM Stomach disorder −0.10* −1.95 1.40–1.45 2.310 1.493
Dehydration −0.16*** −3.22 1.85–1.90 2.146 1.720
Medicine expenditure −248** −1.98 1.10–1.15 2.035 1.627
Skin diseases −0.02* −1.65 1.30–1.35 2.519 2.766
KBM Stomach disorder −0.09** −2.05 1.20–1.25 2.416 1.572
Dehydration −0.15** −2.34 1.35–1.40 2.239 1.725
Medicine expenditure −385** −2.10 1.25–1.30 2.462 1.763
Skin diseases −0.05** −2.07 1.30–1.35 2.031 1.467
RM Stomach disorder −0.12** −1.97 1.35–1.40 2.217 1.584
Dehydration −0.13*** −2.56 1.95–2.00 2.138 1.369
Medicine expenditure −415*** −3.03 1.50–1.55 2.046 1.433
Skin diseases −0.01 −1.42 – 2.418 1.671
MMM Stomach disorder −0.13** −2.18 1.45–1.50 2.064 1.329
Dehydration −0.10*** −3.27 1.10–1.15 2.316 1.472
Medicine expenditure −375*** −2.93 1.40–1.45 2.258 1.230
Skin diseases −0.03 −1.22 – 2.353 1.670

Note: ATT stands for the average treatment affect for the treated, the results (***, **, * ) are significant at 1%, 5% and 10% levels, respectively, for the nearest-neighbor matching, the calipers are reported, while for the kernel-based matching, the band widths are reported, the numbers in each section of the column (A) and (B) are different because different calipers and bandwidths are used.

The main purpose of the PSM is to balance the covariates before and after matching. The results regarding covariates balancing are presented in Table 7. A number of balancing tests like median absolute bias before and after matching, value of R-square before and after matching and the joint significance of covariates before and after matching are estimated. The median absolute bias is quite high before matching and is quite low after matching. Before matching, the bias is in the range of 15.28–27.14. The percentage bias reduction is in the range of 58–81%. The percentage bias reduction indicates that, after matching, the farmers with and without access to safe drinking water are very much similar to each other. The value R-square is quite high before matching and is quite low after matching, indicating that, after matching, there are no systematic differences between the participants and non-participants. The p-value of joint significance of covariates should always be rejected after matching, and should be accepted before matching, hence implying that, after matching, both the households with and without access to safe drinking water are very much similar to each other. The covariates matching results are in line with previous studies like Ali & Abdulai (2010) and Ali & Sharif (2011).

Table 7

Indicators of covariates balancing before and after matching

OutcomeMedian absolute bias before matchingMedian absolute bias after matchingPercentage bias reductionValue of R2 before matchingValue of R2 after matchingp-value of joint significance of covariates before matchingp-value of joint significance of the covariates after matching
Matching algorithm: NNM
Stomach disorder 22.57 4.21 81.34 0.185 0.003 0.001 0.735
Dehydration 27.14 6.79 74.98 0.226 0.004 0.002 0.622
Medicine expenditure 18.52 5.36 71.05 0.132 0.005 0.003 0.713
Skin diseases 19.23 6.55 65.93 0.189 0.005 0.004 0.646
Matching algorithm: KBM
Stomach disorder 18.55 6.29 66.09 0.113 0.005 0.004 0.724
Dehydration 19.34 4.57 76.37 0.251 0.006 0.001 0.512
Medicine expenditure 20.41 7.18 64.82 0.396 0.007 0.003 0.485
Skin diseases 17.69 5.26 70.26 0.241 0.003 0.004 0.373
Matching algorithm: RM
Stomach disorder 17.39 5.12 70.55 0.209 0.003 0.006 0.576
Dehydration 15.28 6.34 58.50 0.326 0.004 0.003 0.468
Medicine expenditure 22.31 5.43 75.66 0.413 0.005 0.002 0.316
Skin diseases 24.10 6.35 73.65 0.510 0.006 0.004 0.637
Matching algorithm: MMM
Stomach disorder 19.35 5.80 70.02 0.362 0.001 0.004 0.412
Dehydration 18.27 4.76 73.94 0.429 0.002 0.007 0.351
Medicine expenditure 19.44 5.29 72.78 0.310 0.004 0.005 0.520
Skin diseases 20.15 6.82 66.15 0.473 0.005 0.004 0.361
OutcomeMedian absolute bias before matchingMedian absolute bias after matchingPercentage bias reductionValue of R2 before matchingValue of R2 after matchingp-value of joint significance of covariates before matchingp-value of joint significance of the covariates after matching
Matching algorithm: NNM
Stomach disorder 22.57 4.21 81.34 0.185 0.003 0.001 0.735
Dehydration 27.14 6.79 74.98 0.226 0.004 0.002 0.622
Medicine expenditure 18.52 5.36 71.05 0.132 0.005 0.003 0.713
Skin diseases 19.23 6.55 65.93 0.189 0.005 0.004 0.646
Matching algorithm: KBM
Stomach disorder 18.55 6.29 66.09 0.113 0.005 0.004 0.724
Dehydration 19.34 4.57 76.37 0.251 0.006 0.001 0.512
Medicine expenditure 20.41 7.18 64.82 0.396 0.007 0.003 0.485
Skin diseases 17.69 5.26 70.26 0.241 0.003 0.004 0.373
Matching algorithm: RM
Stomach disorder 17.39 5.12 70.55 0.209 0.003 0.006 0.576
Dehydration 15.28 6.34 58.50 0.326 0.004 0.003 0.468
Medicine expenditure 22.31 5.43 75.66 0.413 0.005 0.002 0.316
Skin diseases 24.10 6.35 73.65 0.510 0.006 0.004 0.637
Matching algorithm: MMM
Stomach disorder 19.35 5.80 70.02 0.362 0.001 0.004 0.412
Dehydration 18.27 4.76 73.94 0.429 0.002 0.007 0.351
Medicine expenditure 19.44 5.29 72.78 0.310 0.004 0.005 0.520
Skin diseases 20.15 6.82 66.15 0.473 0.005 0.004 0.361

## CONCLUSION AND POLICY IMPLICATIONS

This study examines the pattern of household use of drinking water sources, household access to safe drinking water, and the effects of safe drinking water on human health. It also makes an attempt to identify and analyze the factors affecting household access to safe drinking water sources in Bhutan. The descriptive statistics show that the percentage of both urban and rural households that use safe sources of water has increased during the years 2003–2012. The results also show that a much larger proportion of urban households have access to piped water in the dwelling or compound compared with rural households. Two important findings are clearly evident from the descriptive statistics: firstly, urban dwellers have better access to piped water in the dwelling or compound than rural households, possibly because piped water is considered to be the most reliable, safest and easiest source of drinking; secondly, access to safe drinking water increases with an increase in economic status.

The empirical analysis of the determinant factors affecting household access to safe drinking water suggests that level of education, gender, age, economic status, access to market, and location of the household are all factors that influence household access to safe drinking water in Bhutan. The results show that with an increase in the level of education, the likelihood of using safe drinking water increases, because educated households are expected to be more aware with regard to water safety, and also face a high opportunity cost of drinking unsafe water. Educated households often have paid employment in government or the private sector, and usually live in apartments or houses in urban and semi-urban areas which have piped water in the dwelling. Households with younger, male members, and of large size are more likely to have access to safe drinking water, as they can afford to travel the distance to ensure safe drinking water for their family. Economic status of the household such as landholding and livestock holding size are other key determinants of household access to safe drinking water.

The CLAD model estimates the determinants of household travel to fetch drinking water show that level of education, age of the household head, economic status of the household and location of the household are the key factors that influence household decisions to travel the distance to collect safe drinking water. It is found that households with higher levels of education and those with a younger household head are more likely to travel distance to fetch drinking water. Households with higher economic status (e.g. landholding and livestock holding size) are also likely to travel to ensure safe drinking water for their families. Since access to clean drinking water has numerous positive effects on the well-being of people, policies should be aimed at providing piped water in the compound/dwelling, as this is the safest, most reliable and easiest way of accessing water. In the context of provision of safe drinking water to rural households in Bhutan, it should be pointed out that although it may not be possible to connect each rural household with piped water, given the constraints they face in terms of economic disadvantages and location, an awareness must be created among households of the importance of adopting appropriate methods to treat their drinking water.

## REFERENCES

REFERENCES
Ali
A.
Abdulai
A.
2010
.
Journal of Agricultural Economics
61
,
175
192
.
Ali
A.
Sharif
M.
2011
.
Asian Economic Journal
25
,
413
428
.
Blake
P. A.
Ramos
S.
Macdonald
K. L.
Rassi
V.
Gomes
T. A. T.
Ivey
C.
Bean
N. H.
Trabulsi
L. R.
1993
.
Journal of Infectious Diseases
167
,
627
632
.
Dehejia
R. H.
Wahba
S.
2002
.
Review of Economics and Statistics
84
,
151
161
.
Dungumaro
E. W.
2007
.
Physics and Chemistry of the Earth, Parts A/B/C
32
,
1141
1147
.
Falkenmark
M.
1990
.
Population and Development Review
16
,
81
94
.
Isham
J.
Kahkonen
S.
2002
How do participation and social capital affect community-based water projects? Evidence from Central Java, Indonesia
. In:
The Role of Social Capital in Development: An Empirical Assessment
(Grootaert, C. & van Bastelaer, T., eds). Cambridge University Press, Melbourne, Australia
. pp.
155
187
.
Jalan
J.
Ravallion
M.
2003
Journal of Econometrics
112
,
153
173
.
Jessoe
K.
2013
.
Journal of Environmental Economics and Management
66
,
460
475
.
Mehta
L.
2014
.
World Development
59
,
59
69
.
Rahut
D. B.
Behera
B.
Ali
A.
2015
.
Water Resources and Rural Development
5
,
1
16
.
Rice
E.
Johnson
C.
Wild
D.
Reasoner
D.
1992
.
Letters in Applied Microbiology
15
,
38
40
.
Rosenbaum
P. R.
Rubin
D. B.
1983
.
Biometrika
70
,
41
55
.
Satterthwaite
D.
2003
.
The Annals of the American Academy of Political and Social Science
590
,
73
92
.
Wang
J.
Xu
Z.
Huang
J.
Rozelle
S.
2005
.
Environment and Development Economics
10
,
769
799
.
WHO
2010
Progress on Drinking-Water and Sanitation – 2010 Update
.
World Health Organization
,
Geneva
,
Switzerland
.
World Bank
1993
.
The World Bank Research Observer
8
,
47
70
.
Zhang
J.
2012
.
Journal of Health Economics
31
,
122
134
.