Determinants of the avoidance behaviour of households to cope with unsafe drinking water: case study of Douala and Yaoundé in Cameroon

Abstract

Despite the effectiveness of home water treatment methods, this practice is not yet widespread in many countries, including Cameroon. This study analyses the determinants of the avoidance behaviour of households to cope with unsafe drinking water in the cities of Douala and Yaoundé in Cameroon. The study is based on primary data collected in 2013 from a sample of 789 households in the two cities. The nested logit model is used for empirical analysis. The main findings of the estimated models are as follows: the decision to improve water quality decreases when the head of the household is a man and when there is no child in the household. Furthermore, it decreases when wealth and the level of education are low. In addition, the probability of using a given avoidance method decreases with its cost of adoption and increases with its perceived efficiency (that is, the favourable opinion on the quality of water after treatment). The implications for public policies are discussed in this paper.

Introduction

Access to safe and reliable water remains a daily battle for hundreds of millions of people in the world. According to the World Health Organization and the United Nations Children’s Fund (2017), 844 million people still lack a basic drinking water service, including 159 million people who rely on surface water to meet their daily drinking-water needs. A way to enable individuals to cope with the poor water quality is the use of various avoidance methods. The most promising and accessible avoidance methods are filtration with ceramic filters, chlorination and solar disinfection by the combined action of UV rays and heat (World Health Organization 2012). Generally, several avoidance methods may be combined; for instance, when water is turbid (not clear or contains visible particles), it must first be clarified before being disinfected (chemical disinfection, solar disinfection, boiling) (Cotruvo and Sobsey 2006).

Many studies have found that point-of-use water treatment (that is, water purification at the point of consumption) has a significant effect on both water quality and the reduction in the occurrence of waterborne diseases. The systematic reviews of Speich et al. (2016) and Wolf et al. (2014) support these results. Zin et al. (2013) concluded in their review that promoting household water treatment is most essential in preventing diarrhoea and in reducing the number of deaths caused by it among patients. Moreover, studies suggest that home treatment as compared to source or storage improvements provides the most effective method of ensuring the consumption of clean drinking water (Brick et al. 2004; Fewtrell et al. 2005).

Despite the effectiveness of home water treatment methods, this practice is not yet widespread in Cameroon. According to the National Institute of Statistics (NIS) and ICF International (2012), about 87% of the population do not have the habit of treating water at home. This rate is 76% in Douala and Yaoundé, the two biggest cities of the country that face serious problems of water quality and supply. The consequence of this low water treatment is that, although 71% of households in Cameroon (91% of urban households and 98.6% of households in Douala and Yaoundé) consume water from an improved source (NIS and ICF International 2012), the country regularly faces outbreaks of waterborne diseases, such as the cholera outbreak of 2004 which started from an unprotected well at the Bépanda neighbourhood in Douala and affected more than 7000 individuals causing 130 deaths (Assako Assako et al. 2005). Also, there was another cholera outbreak in May 2010 which affected all the regions of the country and caused several hundreds of deaths. These cholera outbreaks reflect the poor quality of water that people drink. For instance, a research conducted in the neighbourhood of Mvog Betsi in Yaoundé by Nnanga et al. (2014) shows that most of the inhabitants drink water from wells, boreholes or springs and almost all these sources are polluted with Escherichia coli. Another study by Djuikom et al. (2009) conducted in Douala on a water sample from wells shows that the water contains numerous pathogenic bacteria, faecal coliforms and faecal streptococci.

Looking at the huge investments needed to provide safe drinking water to all, it is clear that a large proportion of the world’s population will still have to go for a long time with no access to improved water supply, hence the need for averting actions to cope with poor water quality. It is therefore imperative to understand the factors that may explain household averting behaviours. The objective of this study is to analyse the determinants of households’ avoidance behaviour to cope with unsafe drinking water in the cities of Douala and Yaoundé in Cameroon.

The literature on the determinants of households’ avoidance behaviour is quite extensive. However, no research has addressed the influence of households’ opinion on the effectiveness of avoidance methods in terms of water quality improvement. Indeed, in most existing studies, only the characteristics of individuals/households are often handled, while the characteristics of the avoidance methods are omitted. The main contribution of this study is to test the effects of perceived efficiency of avoidance methods on water averting behaviours. Our findings suggest that the probability of adopting a given avoidance method significantly increases with its efficiency and decreases with its cost of adoption. The findings also suggest the significant effects of wealth, education, gender and the presence of children on avoidance behaviours.

The paper is structured as follows: Sect. 2 presents the literature review, Sect. 3 covers the methodology, Sect. 4 presents the results of the empirical analysis discussed in Sect. 5, and Sect. 6 is the conclusion.

Literature review

The empirical literature on the demand for improved water quality in developed countries is quite rich. Among these studies, a number of them analysing household strategies to cope with unreliable water quality were conducted in the 1980s and 1990s (Smith and Desvousges 1986; Abdalla et al. 1992; Laughland et al. 1993; Larson and Gnedenko 1999). The literature on household avoidance behaviours in developing countries is more recent. Most of the studies carried out in these countries have been in Asia. Few studies like those of McConnell and Rosado (2000) have been conducted in South America. Likewise, few studies such as Dubois et al. (2010), Anderson et al. (2010), Totouom et al. (2012) and Miner et al. (2015) have so far been conducted in Africa.

Two main approaches are often used in the literature to understand factors behind household averting behaviours in response to poor water quality: the analysis of the determinants of households’ choice in adopting avoidance methods on the one hand, and the assessment of the avoidance expenditures followed by the identification of their determinants on the other hand. The econometric models that are selected depend on the question that is addressed: Determinants of avoidance expenditure are commonly estimated using ordinary least squares models, while adoption studies rely either on binary probit/logit models or multinomial logit models. Binary probit/logit models have been used to study household decisions to purify water (Nauges and Van Den Berg 2009; Anderson et al. 2010; Katuwal et al. 2015) or household decisions to adopt a given treatment method (Larson and Gnedenko 1999; Anderson et al. 2010; Katuwal et al. 2015). The multinomial logit model has been used when the entire avoidance strategies available are considered (Haq et al. 2007; Jalan et al. 2009; Katuwal et al. 2015). However, the multinomial logit model is restrictive since it is based on the key assumption of the Independence of Irrelevant Alternatives (IIA), which is hardly conclusive. It is structured in a way that individuals choose between two avoidance options, ‘a’ and ‘b’ independently of the other choices available.¹ Despite the fact that more flexible alternative models have been developed to overcome the questionable IIA assumption, such alternative models are not very much used in the literature.²

Most existing studies only considered socioeconomic characteristics as conditioning factors. Results suggest that avoidance behaviour may be affected by a number of these characteristics. The positive effect of education is highlighted by McConnell and Rosado (2000), Dasgupta (2004), Roy et al. (2004) and Katuwal et al. (2015). The underlying assumption is that education increases awareness about waterborne diseases and knowledge about their prevention. Jalan et al. (2009) in a study carried out in urban India estimated the effects of schooling, exposure to mass media and occupational variables on home water purification. They found that these awareness indicators had statistically significant effects on home water purification and therefore, on the willingness to pay for better drinking water quality.

The positive effect of the presence of children on household averting decisions is also highlighted in existing studies (McConnell and Rosado 2000; Nauges and Van Den Berg 2009; Johnstone and Serret 2012). The idea is that having more children leads to greater awareness of health effects of consuming water of poor quality. Concerning wealth, it is suggested that poor households are less likely to adjust the quality of their water due to related costs. The two most common proxies for wealth are income (Haq et al. 2007; Johnstone and Serret 2012; Katuwal et al. 2015) and wealth indexes constructed on the basis of household ownership of various durables (Jalan et al. 2009). Many studies also point out the positive effect of some socioeconomic characteristics on adoption of averting behaviour such as better occupational status (Jalan et al. 2009; Nauges and Van Den Berg 2009) which may be considered as an indicator of wealth, age (Bontemps and Nauges 2016; Lanz and Provins 2017) due to the greater experience of the aged about the detrimental effects of poor water quality (Lanz and Provins, op. cit.) and being a woman (Abdalla et al. 1992; McConnell and Rosado 2000).

The impact of concern about water quality is also explored in the literature (Abrahams et al. 2000; Jakus et al. 2009; Jain et al. 2014; Beaumais and Veyronnet 2017; Lanz and Provins 2017). Regarding the current water quality level itself, existing studies suggest the positive and significant impact of perceived water quality before treatment on the adoption of water averting behaviour (Nauges and Van Den Berg 2009; Johnstone and Serret 2012; Vásquez et al. 2015). The explanation is that the more people are concerned about the quality of the water they consume or have the perception that it presents a real health risk for them, the more they will purify their water to reduce the risks involved. However, existing studies did not test the impact of perceived water quality after treatment on the choice of avoidance methods. It is however logical to think that the adoption of a given avoidance method over the other may be due to the fact that the preferred method provides better water quality.

Till date, the research of Totouom et al. (2012) remains the only one conducted in Cameroon. The estimated bivariate probit model used in their study highlights the positive and significant impact of education, wealth and the number of children on household avoidance behaviour to cope with unsafe drinking water. However, the study fails to test the impact of avoidance methods’ characteristics on household behaviour due to the lack of relevant data. It only focuses on the decision to treat water and does not investigate the choice of the treatment method.

This study uses a more relevant dataset that allows for the remedying of these shortcomings. Using a nested logit model, this study considers the perceived water quality after treatment as a proxy of the efficiency of avoidance methods in the analysis. It takes into account the potential endogeneity of the subjective quality of water as pointed out by Whitehead (2006). We also assess the impact of the cost of avoidance methods on household behaviour. McConnell and Rosado (2000) are the only few authors to have investigated the effect of such cost in their analysis. Two types of costs are considered in this study: variable costs, equal to the purchase price of inputs used to improve water quality (cotton, fuel for boiling, etc.) and opportunity costs of improvement, equal to the value of time spent improving the quality of water. Appendix 1 provides a table summarising the main characteristics of selected papers from the literature.

Methodology

The econometric model

Following McConnell and Rosado (2000), this study uses the nested logit model to analyse households’ avoidance behaviour to cope with unsafe drinking water in the cities of Douala and Yaoundé in Cameroon. The use of a nested logit (instead of the multinomial logit model largely used in the literature) allows to take correlations between choices available in a particular subset or nest into account (while it maintains the restriction of the IIA between the nests); see McFadden (1978). The use of a nested logit model implies that the choice to adopt a particular coping strategy is dependent on the decision to improve water quality: households first decide whether or not to improve their water quality and later choose their improvement method j (j = 1, 2, ..., m) from a set of available avoidance methods.

Based on the survey carried out, the coping strategies considered in this study are boiling, filtering with cotton, filtering with ceramic filter, use of chemicals (bleach or chlorine) and consumption of bottled water. The indirect utility of a household can be expressed as follows:

$$ {u}_{ij}={v}_{ij}+{\varepsilon}_{ij}={z}_{ij}\alpha +{X}_i\beta +{\varepsilon}_{ij} $$

(1)

where i is the household’s index and j represents the avoidance method.

$$ {v}_{ij}={z}_{ij}\alpha +{X}_i\beta $$

is the deterministic component of utility and ε_ij is the random component known by the household but not observed by the researcher. This is a random utility model developed by McFadden (1974). The error term is assumed to be a random variable independently distributed according to an extreme-value law. X_i is the vector of household socioeconomic characteristics. These variables vary between households but remain constant between alternatives. z_ij is the vector of the attributes related to the treatment options. These attributes vary from one alternative to another and from one household to another.

McFadden (1978) showed that conditional and marginal choice probabilities are given by the multinomial logit formulas P_j/i and P_i. The probability that a household chooses avoidance method j given that it has chosen to improve the quality of its water is

$$ {P}_{j/i}=\frac{e^{\left({\mathrm{z}}_{ij}\alpha \right)}}{\sum \limits_{j=1}^m{e}^{\left({\mathrm{z}}_{ij}\alpha \right)}} $$

(2)

The variables that vary between households but remain constant between alternatives are excluded. The marginal probability that a given household chooses to improve the quality of its drinking water is given by

$$ {P}_i=\frac{e^{\left({X}_i\beta \right)+\upsigma\ {\mathrm{I}}_i}}{\sum \limits_{i=1}^n{e}^{\left({X}_i\beta \right)+\upsigma\ {\mathrm{I}}_i}} $$

(3)

In this formula, I_i is called the inclusive value and is given by

$$ {I}_i=\mathit{\ln}\left(\sum \limits_{j=1}^m{e}^{\left({\mathrm{z}}_{ij}\alpha \right)}\right) $$

(4)

To jointly estimate the models related to the decision of improving water quality and the choice of the avoidance method, the nested logit combines probabilities (2) and (3). The probability of a household’s decision to use avoidance method j to improve the quality of drinking water is

$$ {P}_{ij}={P}_{j/i}{P}_i $$

(5)

The nested logit model is consistent with utility maximisation if and only if the coefficients of the inclusive values parameters are in the unit interval. When they are equal to one, the probabilities of choices are given by the standard multinomial logit. When they are equal to zero, the error terms become perfectly correlated and households choose the alternative with the highest utility.

The parameters of the nested logit model may be estimated by the sequential method or by the maximum likelihood techniques. The maximum likelihood technique is used in this study because it yields more efficient estimates.³To test the IIA assumption, i.e. the importance of using a nested logit model instead of the multinomial logit, the parameters of the inclusive value will be used. In addition, the likelihood ratio test will also be useful to test the null hypothesis of the dissimilarity parameters being equal to 1.

Data

Data used in the study come from a field survey conducted in 2013 among a sample of urban households of Douala and Yaoundé. These cities face serious problems of water supply. The objective of the survey was to provide an overview of the water situation in Cameroon’s households living in urban areas. Information related to all available water sources, collection of water strategies, consumption quantities and avoidance methods to cope with unreliable water quality were collected. Details on socioeconomic and demographic characteristics of the surveyed households were also gathered. Data were collected through personal interviews. The questionnaire was first pre-tested to assess the content of survey tools. Feedback from the pre-test was used to revise the final questionnaire. The estimation of the theoretical sample size of 845 households (522 in Douala and 323 in Yaoundé) to survey was based on the following formula developed by Sudman and Bradburn (1982):

$$ n=\frac{(1.96)^2p\left(1-p\right)}{\lambda^2} $$

This model is recommended for a large population (over 100,000 individuals). n is the sample size to calculate, 1.96 corresponds to the choice of a 95% confidence interval, p represents the proportion of the population showing interest and λ is the tolerable error, that is the margin of error for the survey. The proportion of the population that purify water in the cities of Douala and Yaoundé is 32 and 16%, respectively, in the Third Multiple Indicators Cluster Survey (MICS3) conducted in 2006 by the National Institute of Statistics. Therefore, p = 0.32 for Douala and p = 0.16 for Yaoundé. If the tolerable margin of the error value of 4% is taken, then the above formula will yield the sample size for the cities of Douala and Yaoundé of n₁ = 522 and n₂ = 323, respectively. At the end of the survey, only 789 questionnaires were correctly filled and exploited (491 in Douala and 298 in Yaoundé), corresponding to a response rate of 93% (94% in Douala and 92% in Yaoundé) compared to the set target of 845.

The cartographic data from the Bureau of Census and Population Studies (BUCREP) developed in 2003 under the Third General Census of Population and Housing (RGPH) were used as the sampling frame for the survey. For sample collection, a two-stage random sampling method was adopted: a random selection of a number of enumeration areas within each sub-division of Douala and Yaoundé followed by a random walk selection of a number of households to be interviewed in each enumeration area sampled. A total of 26 enumeration areas involving 26 neighbourhoods were taken (11 in Yaoundé and 15 in Douala). As mentioned above, all the sub-divisions in Douala and Yaoundé were involved in the survey to ensure a sufficient geographical coverage and a spatial representation of the population. There were attempts to achieve an equitable distribution of the number of surveyed households among the sub-divisions in each city (approximately 133 households per sub-division in Yaoundé and 52 households per sub-division in Douala). The random selection of the surveyed households ensured that the sample was representative. The observation units during the survey were households. Information was gathered through face-to-face interviews with the heads of households. If the heads of the households were absent, we spoke to their close relatives, who gave us information about them like their age, marital status and educational levels.⁴

Data collected showed that piped network was the main source of water in the study area. Of the 789 households surveyed in the study, 558 (70.7%) collected water from the piped network. However, among these households, only 360 (45.6%) had pipe borne water at home. The 198 other households were supplied by public taps. In general, respondents had a positive opinion on the quality of the water supplied (about 85% of the whole sample). Regarding avoidance methods used to get better water quality, survey data showed that out of the 789 surveyed households, 402 (about 51% of the whole sample) usually did something to get better water quality. Households were asked to only report the main averting action usually undertaken. Description of the avoidance methods used, and the households’ socioeconomics and demographics characteristics are reported in Sect. 4 of the paper.

The explanatory variables

The explanatory variables used in this study can be classified into two main categories:

Socioeconomic characteristics

Wealth

Multiple component analysis has been used to build the households’ wealth index based on several variables such as housing characteristics (access to electricity, access to piped water, presence of modern toilets, quality of the building materials, etc.) and ownership of some durable items (cars, mobile phones, etc.). This wealth index is further normalised in order to obtain values that vary between 0 and 1. This normalisation is based on the following formula:

$$ {\mathrm{wealth}}_N=\frac{wealth_G-\min \left({wealth}_G\right)}{\max \left({wealth}_G\right)-\min \left({wealth}_G\right)} $$

where wealth_G is the gross indicator and wealth_N is the normalised indicator. Values of the normalised wealth index are grouped to obtain four ordered classes represented by four dummies. It is expected that worse-off households are less likely to improve their water quality because of the costs involved.

Education

This variable measures the level of education of the heads of households. It is a categorical variable which has been classified into three dummies: Primary education (1 if the head of the household has gone at most through primary education, 0 otherwise), Secondary education (1 if the head of the household has gone at most through secondary education, 0 otherwise), and Higher education (1 if the head of the household has gone through higher education, 0 otherwise).

Male

This variable captures the gender of the head of the household. It is equal to 1 for male-headed households and 0 for female-headed households.

No child

This variable is equal to 1 for households that do not have at least one child under five and 0 otherwise.

Characteristics of the avoidance methods

Quality

It is the respondent’s opinion on the quality of water after treatment with each of the avoidance methods available. For each of the avoidance methods, respondents were asked to give their opinion on the quality of water after treatment. Of the four answers proposed in the questionnaire, the answers ‘bad’ and ‘too bad’ (or ‘very high risk’ or ‘high risk’) were coded 0, while answers ‘good’ and ‘excellent’ (or ‘moderate risk’ and ‘no risk’) were coded 1. The variable Quality is used as a proxy variable to measure the effectiveness of the avoidance methods.

The introduction of perceived water quality in averting-decision models can potentially cause endogeneity bias (Whitehead 2006; Bontemps and Nauges 2016; Lanz and Provins 2017). However, it is not obvious to have appropriate and valid instruments and those chosen by Vásquez et al. (2015), Whitehead (2006) and Danielson et al. (1995) are questionable. Bontemps and Nauges (2016) used regional variables as instruments for the household’s perceived health impacts of tap water in 2011. Beaumais and Veyronnet (2017) and Lanz and Provins (2017) modelled perceived water quality as a function of objective water quality. However, such objective measure of water quality is not available in our dataset. To avoid the problem of endogeneity bias arising from the introduction of subjective quality of water in averting behaviour models, the opinion of each respondent is replaced in the study by the average opinion of the respondents of the municipality having the same avoidance behaviour (no treatment of water, boiling, filtering with cotton, filtering with ceramic filter, use of chemicals or consumption of bottled water). This average opinion is equal to the share of households of the municipality that have a favourable opinion on the quality of water after treatment with each of the above strategies. Two types of opinion are considered in this study: opinion on water safety and opinion on overall water quality.⁵

Cost

This variable is the financial cost associated with the acquisition and the use of each avoidance method. Two types of costs are considered in the present study: variable cost and total cost. For households who boil their drinking water, variable cost is given by the average purchase cost of fuel used each month. For households who purify their water with bleach, chlorine or cotton, variable cost is determined by the average purchase cost of these inputs per month. For households who use a ceramic filter, variable cost is determined by the ratio between the purchase cost of the filter and the number of years from the purchasing date of the filter to the date of survey. In order to get a monthly value, this ratio is then divided by 12. For households who consume bottled water, variable cost is given by the average cost of purchasing water each month. Variable cost is associated with a specific avoidance method and is provided only for households that use the method. For the other households who improve drinking water quality, they are associated the average variable cost calculated over the entire sub-sample of households using the avoidance method. Total cost is equal to the sum of variable cost and opportunity cost of time spent treating water. This opportunity cost is given by the average time (in hours) spent every month to improve water quality multiplied by the head of the household’s hourly income.⁶ The head of the household’s hourly income is obtained by dividing his monthly income by his/her monthly work time (in hours). Assuming that people work averagely for 35 h a week,⁷ the monthly work time is equal to 35 × 4 = 140 h.

For households who use bleach, chlorine or cotton to treat their water, the time spent treating water is calculated by the average time spent going to and from the usual point of purchase multiplied by the average number of trips made per month. For households who consume mineral water, this time is determined by the average time spent going to and from the usual point of purchase multiplied by the average number of trips made per month. Since people were explicitly and exclusively questioned about the time spent fetching bleach/chlorine, cotton or mineral water, it is assumed that trips are specially made to purchase bleach/chlorine, cotton or mineral water.⁸ That is, there are no multi-purpose trips. For households who boil water, the average time spent treating water is calculated by the average time spent boiling water each time multiplied by the number of times the water is boiled in a month. For households who use an ordinary filter, the opportunity cost is zero.

The opportunity cost of time associated with a specific coping strategy is defined only for households who use the strategy. For the other households who improve their drinking water quality, they are associated the average opportunity cost calculated over the entire sub-sample of households using the coping strategy. The table in Appendix 2 summarises the variables used in the study.

Empirical results

Descriptive statistics

The data collected show that different avoidance methods are used by the surveyed households. The following question was asked to the respondent: ‘In your household, do you usually undertake averting actions (filtering, boiling, purchasing bottled water, etc.) to have a better drinking water quality in order to reduce associated health risks?’ For the respondents who answered yes to the question, an additional question was asked: ‘What is the main averting action usually undertaken to have better water quality?’ with the following proposed answers ‘boiling, filtering with cotton, filtering with ceramic filter, use of chemicals (bleach or chlorine), purchasing and consumption of bottled water, solar disinfection, others (mentioned)’. The full questionnaire is available in Appendix 3.

Table 1 below provides the distribution of surveyed households by avoidance method.

Table 1 Distribution of households by avoidance method

It is observed from Table 1 that the use of ceramic filters remains the main avoidance method used by the surveyed households. They are used by 36% of households (33 and 41% respectively in Douala and Yaoundé). Filtering of water with cotton is also a coping strategy widely used. The proportion of surveyed households filtering water with cotton is 34, 23 and 30% respectively in Douala, Yaoundé, and the total sample. ‘Solar Disinfection’ and ‘others’ are the least used methods with a percentage of 0.8 and 0.50%, respectively. These two methods are not considered in the econometric analysis. The consumption of bottled water is marginal among the surveyed households (8%). In developing countries in particular, bottled water is reserved for the few fortunate households. Table 2 provides descriptive statistics of the different explanatory variables used in the econometric analysis.

Table 2 Descriptive statistics of explanatory variables

Table 2 also shows that about 10% of heads of households have gone through at most the primary education, 47% the secondary education and 43% the higher education. Regarding the heads of households’ gender, statistics show that about 80% of households are headed by men. As far as children are concerned, the survey shows that about 60% of households do not have at least one child below five. Concerning wealth index, it has been categorised into four dummies. The average wealth score of households is 0.25 with the maximum value of 1 and the minimum value of 0. A higher value of the index indicates a wealthier household.

Concerning the respondents’ perception of the water quality after treatment, Table 2 shows that about 86% of the respondents are satisfied with the overall quality of water. This percentage is 87% when water quality in terms of health safety is considered. Regarding avoidance expenditures, statistics reveal that the average variable cost of water treatment is 3656 FCFA per month. These costs vary from a minimum value of 0⁹ to a maximum value of 56,000 FCFA. When the opportunity cost of time spent improving water quality is added to costs, the average total cost becomes 4651 FCFA per month. This total cost represents about 3% of the average monthly income of the households which is 158,000 FCFA.

Estimation results

After removing households with missing data from the sample, a total of 769 households were finally considered. Given the nature of the nested logit model used, the number of observations on each variable is six per household.¹⁰ Estimates therefore involved 769 × 6 = 4614 observations. The results of the first estimation carried out are reported in Table 3.

Table 3 Estimation results of the nested logit model

Several tests were further conducted in order to check the robustness of the results. The results of two alternative specifications are reported in Table 4. The first alternative model is model 1 which is close to the one reported in Table 3. The difference here is that a specific aspect of water quality, namely the degree of safety (or inversely, the low level of health risk), is incorporated into the model as an explanatory variable instead of the overall water quality. The results are not significantly different from the previous ones. They suggest that a favourable opinion about the effectiveness of an avoidance method increases its probability of adoption.

Table 4 Results of the robustness checks

Column 3 of Table 4 presents the results of the estimation of the second alternative specification. In model 2, variable cost is used instead of the total cost. This specification is used to test the robustness of the negative and significant impact of the cost of an avoidance method on its probability of adoption. The second alternative model thus incorporated the variable cost and the degree of safety. Once more, results are not different from those presented in Table 3. In conclusion, the results of the different alternative estimations are very similar and close to those presented in Table 3. Results show that the sign and the significance of all the variables remain the same, indicating robustness of the findings.

The estimated parameter of the inclusive value in the branch ‘improve’ is equal to 0.68. This means that the five avoidance methods are more substitutable among themselves, than with the alternative ‘do not improve’. The parameter of the inclusive value for the branch "do not improve" option is set to be 1 because it is a degenerated branch (or single option). In addition, the likelihood ratio test indicates that the null hypothesis of the dissimilarity parameters equal to 1 is rejected at the 1% level, justifying the appropriateness of the nested logit model used.

Discussion

The study explores the effect of wealth on the decision to improve drinking water quality. The coefficients of the different wealth dummies are negative and statistically significant (the top wealth quartile is used as reference). This result suggests that the more a household is poor, the less likely it will improve its drinking water quality. The interpretation seems to be that avoidance methods for some households involve significant costs, such that the income constraint becomes a factor that limits their choices. Such a result is not new in the literature. Previous studies such as that of Jalan et al. (2009) discussed the influence of wealth on the likelihood of treating water.

As expected, Primary education and Secondary education variables have a negative coefficient (however, the coefficient of Secondary education is not significant). These findings suggest that as compared to households with heads that have gone through higher education, households with less educated heads are less likely to improve the quality of their drinking water. The reason is probably that income varies with the level of education. The education effect could therefore be correlated with the wealth effects. Furthermore, less educated households’ heads are possibly less aware of the detrimental effects of consuming contaminated water. As suggested in the literature, a low level of education limits understanding of the issues related to the availability of drinking water at home such as health benefits for instance (Briand and Loyal 2013). Actions of education and sensitization of the population on the health effects of unsafe water consumption should thus be undertaken by health authorities in order to prompt the population to pay special attention to the quality of water they drink. Such actions which may be conducted through media are likely to produce positive effects on the population in support of the Health Belief Model (Rosenstock 1974), which suggests that warnings or sensitization campaigns about health issues serve as a signal that triggers the perceived threat and the likelihood of the action. As stated by Figueroa and Kincaid (2010), several studies confirmed the positive effect of the different communication channels used by water treatment interventions for disseminating water treatment and hygiene messages, especially mass media (radio and television), community participation (health clubs, schools, water committees), entertainment education (local theatre, videos) and interpersonal communication. Also, in a study conducted by Jalan and Somathan (2008) about the extent to which information affects the demand for environmental quality, authors found that households initially not purifying their water and told that their drinking water was possibly contaminated were 11% more likely to begin some form of home purification in the next 8 weeks than households that received no information. Moreover, an additional year of schooling of the most educated males in the households is associated with a 3% rise in the probability of initial purification.

This study also tested the effect of the gender of the households’ heads on household avoidance behaviour. The result is consistent with those of many other existing studies like McConnell and Rosado (2000) who found that female-headed households are more likely to adopt averting behaviour as compared to male-headed households. The negative coefficient of the variable Male might be explained by the fact that women in general are in charge of taking care of sick persons in a household. Therefore, they are more concerned about the health detrimental effects of consuming unsafe water than men to the extent that in case of sickness, the time that they would have invested in productive and remunerative jobs, education of children or the preparation of meals will be reallocated to sick persons. Furthermore, studies on intra-household bargaining power such as the one of Schmidt (2012) provided supportive evidence of the positive impact of women power on health outcomes and thus, on the demand for water quality.

The findings of the study also suggest that a household without at least one child less than five is less likely to adjust the quality of water than a household with young children. Indeed, since children are more vulnerable to health risks from unreliable water than adults, households’ members seem to be less sensitive to water quality issues in the absence of children. The impact of the presence/absence of children on the choice to purify water or to consume bottled water has been shown by many authors (McConnell and Rosado 2000; Nauges and Van Den Berg 2009; Johnstone and Serret 2012; etc.).

The findings also show that the estimated cost of improving water quality has a negative and statistically significant coefficient, indicating that the more an avoidance method is expensive, the less likely it will be chosen. This result is consistent with the demand theory which states that as the price of a good increases, its demand decreases, ceteris paribus. The negative impact of the cost of water treatment on household choice is also highlighted in the study by McConnell and Rosado (2000).

As to the impact of the respondents’ opinion on the quality of drinking water, findings show that the coefficient of water quality is positive and statistically significant at 1%. This result suggests that the more respondents consider an avoidance method to be efficient (that is, it produces better water quality), the more likely they will adopt it. This result is consistent with expectations.

Conclusion

Based on data collected from a sample of 789 households in Douala and Yaoundé, this study uses a nested logit model to identify factors that influence household avoidance behaviours to cope with unsafe drinking water in the cities of Douala and Yaoundé in Cameroon. The regressions carried out suggest robustness of the findings. In particular, results reveal that the probability of adopting avoidance methods decreases with their costs and increases with their perceived efficiency (favourable opinion on the quality of water after treatment). The implementation of policies that aim at making averting methods more affordable, as well as, the sensitisation of the public about the efficiency of water treatment as a means to improve the quality of water and reduce waterborne diseases are recommended. Priority should be given to the promotion of the most efficient methods. At the same time, the public should be sensitised (through mass media for instance) about the detrimental health effects of consuming unsafe and unreliable water. This sensitisation should be permanent and not only conducted during periods of cholera outbreaks, as it used to be the case. Moreover, authorities should put in place appropriate measures to make averting technologies more affordable for the public. They should specifically intervene on the supply side by facilitating the setting up of companies that produce averting technologies (ceramic filter for instance). Such policies could take the form of tax exonerations on the production or the importation of the inputs. It is really surprising to observe that till date, no company producing ceramic filters is present in the country and there is no official policy that supports such local production. The local production of coping technologies should result in the increase of the supply and the reduction of the prices. The drop in the prices of bottled water in the country that arises from the current development of companies producing bottled water is appreciable and will probably increase the rate of consumption of bottled water that remains very low in the country.

Appendices

Appendix 1

Table 5 Selected papers from the literature

Appendix 2

Table 6 Description of the variables

Appendix 3. Questionnaire

Part I—Socioeconomic and demographic profile of the household

Part II – Relation with drinking water