Appropriate household water treatment methods in Ethiopia: household use and associated factors based on 2005, 2011, and 2016 EDHS data

The Ethiopian Demographic and Health Survey (EDHS) conducted in 2005, 2011, and 2016 were used as a data source for analysis. The surveys were conducted based on a nationally representative sample households that provide estimates at the national and regional levels. Women of child-bearing age 15–49 and all men aged 15–59 in randomly selected households across the country were the target groups for the surveys [8, 9, 12]. This study considered all households in the country in each survey as a targeted population as it was intended to generalize the point-of-use water treatment practices at the national level. Data collection was taken place by interviewing respondents from the selected households [8, 9, 12]. The Federal Ministry of Health supported the survey implemented by the Central Statistical Agency with technical assistance from the ICF International through its MEASURE DHS project. MEASURE DHS is a 5-year project to assist institutions in collecting and analyzing data needed to plan, monitor, and evaluate population, health, and nutrition programs [13].

The sample in each EDHS was designed to provide population and health indicators at the national and regional levels. The sample design allowed for specific indicators to be calculated for each of Ethiopia’s 11 administrative regions: 9 regional states (Tigray, Afar, Amhara, Oromiya, Somali, Benishangul-Gumuz, Southern Nations Nationalities and Peoples (SNNP), Gambela, and Harari) and 2 city administrations (Addis Ababa and Dire Dawa). Accordingly, 14,500 households from 540 clusters in 2005 EDHS; 17,817 households from 624 clusters in 2011 EDHS, and 16,650 households from 645 clusters in 2016 EDHS were selected for the surveys. In each survey, each of the regions was stratified into urban and rural areas, and samples of enumeration areas (EAs) were selected independently in each stratum. The samples were selected using a stratified, two-stage cluster design. First, clusters were selected from the list of enumeration areas from the population and housing census sample frame (1994 census for 2005 EDHS and 2007 census for 2011 and 2016 EDHS). Second, households were selected after listing of households in all of the selected EAs for sampling frame. The detail of the sampling technique is found elsewhere [8, 9, 12].

Households reported using appropriate water treatment methods were considered for analysis as an outcome of an interest. According to 2011 and 2016 EDHS reports, adding bleach, boiling, filtration, and solar disinfection (SODIS) were taken as using appropriate water treatment methods. Therefore, households’ reported use of boiling, adding bleach, filtering, or solar disinfecting took a binary form such that the use was considered a yes (1 = if the household had used either of the methods during each survey) and no otherwise (0 = if the household had used neither during each survey). The predictor variables are residency of households (categorized into urban and rural), household wealth status categorized into five quintiles (poorest, poorer, middle, higher, and highest), presence of under 5-year-old child in the household, education status of household head, owning television and radio, type of water sources (improved versus unimproved), sex of household head, age of household head, and intermittent water supply.

The “svy” command in Stata was used to weight the survey data for the adjustment of cluster sampling design. The weighted data were analyzed using descriptive (frequency and proportion), bivariate, and multivariable regression analysis. Bivariate regression was applied to determine unadjusted effects of each of the variables on household use of the treatment methods during each survey. Then, we subsequently included the variables with a p value < 0.25 for multivariable regression to assess the independent effect after controlling other variables [14]. In addition, a two-level multilevel modeling was employed to determine whether the effect of identified predictors vary due to regions. The two-level multilevel logistic regression that considers a collection of groups (regions) and within-group (random households selected within regions) is used for analysis because data from a random sample of level 1 households were collected from level 2 regions. For all analysis, the significant association of predictor variables was considered at p value < 0.05. All statistical analyses were conducted using Stata version 14.0 (Stata Corp, College Station, TX, USA). Multi-collinearity diagnostics were conducted to exclude the variable/s with the variance inflation factor (VIF) of greater than 10 from multivariable regression.

Data about appropriate water treatment and characteristics attributable to it were analyzed using 13,711, 16,672, and 16,650 households respectively for 2005, 2011, and 2016 EDHS. Table 1 shows the household characteristics of the three surveys. Of total household heads, about 68% in 2005, 57% in 2011, and 55% in 2016 had no formal education. A small number of households (4.9%) owned a television in the 2005 survey, and the number increased to 13.8% in 2016 survey. More than 70% of adult women were responsible to fetch water in each of the surveys. There was an increasing trend in the number of households using improved water sources in three surveys. The number of households accessed the water on the premises was 8.4% in 2005 and increased to 20.1% in the 2016 survey. The number of households that accessed drinking water supplies within 30 min in 2011 and 2016 surveys was almost the same.

The number of households treating their water prior to drinking with any treatment options was 8.0% in 2005, 10.2% in 2011, and 9.4% in 2016 (Table 2). In 2005 EDHS data, a total of 3.0% of households had used one of the appropriate treatment methods. Explicitly, 2.4% of households had reportedly used boiling, 0.2% of households had reportedly used bleach, and 0.3% of households had reportedly used a filter as point-of-use water treatment methods. In 2011, the number of households that reportedly used boiling, bleach, and filter as household water treatment methods was 2.7%, 5.8%, and 0.2%, respectively. In this survey, there was no household that reportedly used SODIS. The total number of households that reportedly used the appropriate household water treatment methods in 2011 was 8.2%. In 2016, 6.5% of households had reportedly used either of the listed treatment methods to treat their drinking water prior to drinking. Specifically, 2.2%, 3.4%, 1.1%, and 0.1% was the number of households that reportedly used boiling, bleach, filter, and SODIS, respectively, as a household water treatment method. In general, appropriate water treatment use is least in the 2005 survey and highest in 2011 survey.

Table 3 indicates the bivariate regression of the household use of appropriate point-of-use water treatment methods in the three surveys. In the bivariate regression analysis, the socio-demographic and economic characteristics of households were associated with the use of appropriate point-of-use water treatment methods. The odds of using appropriate point-of-use water treatment methods in urban area are 1.51 (95% CI = 1.03, 2.22) times higher in 2005, 1.80 (95% CI = 1.32, 2.45) times higher in 2011, and 1.58 (95% CI = 1.17, 2.13) times higher in 2016 compared to rural area. Household head with higher education had the odds of 7.42 (95% CI = 4.78, 11.52) times higher in 2005, 6.65 (95% CI = 4.61, 9.58) times higher in 2011, and 3.27 (95% CI = 2.29, 4.67) times higher in the reported use of appropriate water treatment methods compared to house head who had no education. The odds of reported use of appropriate treatment methods among households found in the highest wealth quintile in 2005, 2011, and 2016 were 2.77 (95% CI = 1.39, 5.54), 5.02 (95% CI = 3.19, 7.90), and 1.83 (95% CI = 1.06, 3.18) times higher, respectively, than households found in poorest wealth quintile.

Table 4 shows multivariable regression analysis on the use of appropriate point-of-use water treatment methods and associated factors. There was no collinearity among predictor variables as the multi-collinearity diagnostics did not show variables with VIF of greater than 10. In all surveys, education status of household head and owning radio had a significant association with the household use of the water treatment methods after controlling other variables. The household head with higher education in 2005, 2011, and 2016 had 5.99 (95% CI = 3.48, 10.33), 3.61 (95% CI = 2.56, 5.07) and 3.43 (95% CI = 2.19, 5.37) times higher odds ratio, respectively, to use appropriate treatment methods as compared to household head who had no education. Households owned a radio in 2005, 2011, and 2016 had 1.58 (95% CI = 1.02, 2.47), 1.50 (95% CI = 1.21, 1.86), and 1.34 (95% CI = 1.03, 1.74) times higher odds ratio, respectively, to use the appropriate water treatment methods as compared to the households who did not own radio. There is sufficient evidence to accept the hypothesis that the reported use of appropriate treatment methods is high in the households of educated household heads and owned a radio in all the three surveys. But, we fail to accept the hypothesis that there is a high reported use of the treatment methods in the households located in urban areas, found in high wealth quintiles, owning a television, dependent on unimproved water sources, and having children with age below 5 years in 2005 EDHS surveys. From the pooled data regression, the odds of treating water with an appropriate water treatment were 2.78 (95% CI = 2.16, 3.57) and 2.18 (1.64, 2.89) times higher in 2011 and 2016, respectively, compared to 2005 DHS survey. Households with primary, secondary, and higher education head had 1.71 (1.44, 2.02), 2.17 (1.73, 2.73), and 3.58 (2.79, 4.59) times higher odds to use appropriate water treatment methods, respectively, as compared to household heads with no education. Moreover, the odds of treating water with appropriate household treatment methods is higher in the households of being in higher and highest wealth quintiles, being living in rural, dependent in unimproved water sources, and owning radio and television (Table 5). In a pooled data analysis, there are sufficient evidences to accept the hypothesis that the reported use of appropriate treatment methods is high in the households of living in rural, educated household heads, found in high wealth quintiles, owned a radio, owned a television, and dependent on unimproved water sources.

The data had a two-level hierarchical structure with households at level 1 nested within 11 regions at level 2. We began by the empty (intercept only) two-level model for a dichotomous outcome variable (household use or non-use appropriate treatment methods). The output shows that there was a variation in the use of appropriate water treatment due to the variation in regions. The interclass correlation from empty model indicates that 15.3% of variations in 2005, 9.1% of the variation in 2011, and 29.6% of variations in 2016 are attributable to the difference in the region (Table 6). The variations in the use of treatment methods due to the difference in the region were significant in each survey regardless of whether there was a higher within-region variation compared to between-regions variation. The higher within-region variation than between-regions variations suggests that the use of point-of-use water treatment methods is a nationwide problem.

Similar to the output of multivariable regression analysis, the fixed part of multilevel modeling indicates that household use of appropriate water treatment options generally was associated with the educational status of household head, wealth status, and ownership of radio and television. The 2016 survey shows that supplying water intermittently to household had 1.26 (95% CI = 1.06, 1.49) times higher odds of using treatment methods regardless of the region where the households are located (result not indicated in the table). In 2005 and 2011 EDHS, households dependent on unimproved water sources were 34% and 37% more likely to use water treatment methods, respectively, compared to households using improved water sources. In 2016, households’ use of either boiling, filtration, bleach, and SODIS was associated with the presence of under 5-year-old children in the household (AOR 1.12; 95% CI = 0.99, 1.27) (Table 7).

The EDHS data we analyzed were collected cross-sectionally and, therefore, have the following limitations: (1) The responses were liable to biases (social desirability bias); (2) The analysis fails to show the cause and effect relationship between independent variables and dependent variable [30]. The data did not indicate whether households claimed using the treatment methods were confirmed users and how consistent is the water treatment. In addition, the data did not explicitly indicate different types of filtration methods used by households. Psychosocial factors, one of the factors associated with WASH technology adoption and use, were not included in the data for analysis. The Ethiopia Demographic and Health Survey is conducted in every 5 years via appropriate procedures of selecting households that would represent the whole population in the country. Therefore, the representativeness of the data is one of the strengths when compared with area-specific studies being conducted in the country.