Environmental Health and Household Demographics ImpactingBiosand Filter Maintenance and Diarrhea in Guatemala: AnApplication of Structural Equation ModelingDaniel William Divelbiss,*,† Dominic Louis Boccelli,† Paul Allan Succop,‡ and Daniel Barton Oerther#

†School of Energy, Environmental, Biological, and Medical Engineering, College of Engineering and Applied Science, University ofCincinnati, Cinccinnati, Ohio 45221, United States‡Department of Environmental Health, Division of Biostatistics and Epidemiology, College of Medicine, University of Cincinnati,Cincinnati, Ohio 43219, United States#Environmental Research Center, Missouri University of Science and Technology, Rolla, Missouri 65409, United States

*S Supporting Information

ABSTRACT: In rural health development practice, engineers and scientistsmust recognize the complex interactions that influence individuals’ contact withdisease-causing pathogens and understand how household habits may impactthe adoption and long-term sustainability of new technology. The goal of thisstudy was to measure the effect of various environmental health factors andhousehold demographics on the operation and maintenance of the Biosandfilter (Centre for Affordable Water and Sanitation Technology, Calgary,Alberta, Canada) and diarrhea health burden in the region. In July and August2010, randomized household surveys (n = 286) were completed in ruralGuatemala detailing water access, sanitation availability, hygiene practice, socio-economic status, education level, filter operation and maintenance, and diarrheahealth burden of the home. A hypothesized structural equation model wasdeveloped based on a review of published research and tested using thesurveyed data. Model-derived parameter estimates indicated that: (a) proper personal hygiene practices significantly promoteproper filter operation and maintenance; and (b) higher household education level, proper filter operation and maintenance, andimproved water supply significantly reduce diarrhea health burden. Additionally, a high level of unexplained variance in diarrheaindicated the filter, though protective of health, is not the only factor influencing diarrhea.

1. INTRODUCTION AND OBJECTIVES

Diarrheal disease has been identified as the second highestcontributor to childhood mortality globally.1 The commonlyaccepted paradigm is that safe water, adequate sanitation, andproper personal hygiene practices prevent diarrhea.2 Further-more, deficiencies of these three variables have been attributedto causing 88% of global diarrheal deaths.3

In addition to safe water, adequate sanitation, and properhygiene, household education level has been shown byresearchers to have an influence on diarrhea. The patternpersists even when household income and economic status arecontrolled suggesting that higher levels of education isprotective of household health.4,5 Higher levels of householdincome have been correlated with better housing conditionsincluding adequate sanitation and improved water supply.6

Additionally, high levels of income allow households to spendmore on nutritious food, medicine, and healthcare that candirectly impact household health. Overall, health in thedeveloping world is dependent on a complex network ofenvironmental health factors.The Biosand filter − promoted by the Centre for Affordable

Water and Sanitation Technology − has been employed in

water quality interventions in the developing world forapproximately 20 years. Recent maintenance, access by childrenor animals, depth of fine sand bed less than 46 cm, andextremely poor influent water quality increases the chance of afilter having unacceptable filtrate.7 These identified failuremodes could be affected by external factors similar to those thatimpact diarrhea disease in the home despite proveneffectiveness of the filter in laboratory and field trials.8−14

Therefore, external factors (e.g., economic status, educationlevel) could also have adverse effects on the water qualityprovided to the end user, ultimately impacting health.Because many factors influence diarrhea disease, a holistic

view of environmental health must be evaluated. Existing case-control studies of single interventions give promisingresults,10,11 but if all attributes within the environment arenot quantified then the observed successes may not be scaledup to regional or national policies. A powerful multivariate

Received: September 6, 2012Revised: November 12, 2012Accepted: November 15, 2012Published: November 15, 2012

Article

pubs.acs.org/est

© 2012 American Chemical Society 1638 dx.doi.org/10.1021/es303624a | Environ. Sci. Technol. 2013, 47, 1638−1645

analysis technique capable of testing multiple hypotheses withina complex environment is structural equation modeling.Structural equation modeling (SEM) is a quantitative

method used to estimate to what extend a hypothesizedmodel is supported by sample data.15 SEMs share thecharacteristics of requiring a priori specifications of thetheorized relationships and allow the application of bothobserved and latent variables. Model specification is commonlybased on theories, previously published research, andexperience of the researchers.16 The model is then testedwith primary data and, if the data gathered is inconsistent withthe model, the researcher must reject or modify the model.In this study, a conceptual model based on the existing body

of knowledge of acute diarrhea pathology was compiled tocreate a hypothesized SEM. Collected primary data was appliedto the hypothesized model and modifications were made to themodel to improve the fit. The modification reduced the numberof interactions defined to create the most parsimonious modelpossible. The resulting influence pathways in the model identifymultiple possible points for future interventions. Onemethodological advantage of SEM over other analysis is theability to specific variables in the model to be simultaneouslyindependent and dependent (i.e., filter operation andmaintenance is influenced by various factors and independentlyinfluences diarrhea health burden).15 This feature allows themodel to be representative of the system under study.The objectives of this study were (1) to develop a

hypothesized SEM describing the factors affecting operationand maintenance of the Biosand filter and household diarrheaburden, (2) to generate and deploy a survey instrument togather primary data at the site of a household based point-of-use water quality treatment project employing the Biosand filterin rural Guatemala, and (3) to test the hypothesized modelwith the collected data to identify the strength of interactionsamong demographics and environmental health factors that

influence Biosand filter operation and maintenance andhousehold diarrhea burden.

2. HYPOTHESESThe following hypotheses, based on findings in the literature,can be simultaneously tested through the use of SEM analysis.

• Increased household educational level has a negativeeffect on the severity of diarrhea burden.

• Increased socio-economic status has a negative effect onthe severity of diarrhea burden.

• Poor hygiene practices have a positive effect on theseverity of diarrhea burden.

• Additional water treatment beyond the filter has anegative effect on the severity of diarrhea burden.

• Access to an improved water source has negative effecton the severity of diarrhea burden.

• Access to adequate sanitation has a negative effect on theseverity of diarrhea burden.

• Proper water storage has a negative effect on the severityof diarrhea burden.

Limited research has been performed to identify filteroperation and maintenance characteristics that have a specificresponse to various household characteristics. One such studyattempted to identify the determinants for proper filteroperation and maintenance; however, the households used inthe study had exceedingly high attrition rates and the long-termuse portion of the study was not completed.17

The following hypotheses were suggested based on expect-ations of how certain household characteristics would affectfilter operation and maintenance and findings in the literatureassociated with the impacts on diarrheal burden:

• Increased socio-economic status has a positive effect onfilter operation and maintenance.

• Better personal hygiene practices have a positive effect onfilter operation and maintenance.

Table 1. Indictor Variables, Relative Latent Variables, Types, and Definitions

indicator latent variable type definition

use storage in home household water storage dictomous household stores domestic water on-site in container(s)tap installed in storage containers household water storage dictomous water container(s) have tapssmall mouth storage container household water storage dictomous water container(s) have small mouthscovered storage containers household water storage dictomous water container(s) have coverspaternal education level household education level continuous ability to read and write plus the sum of years of formal educationmaternal education level household education level continuous ability to read and write plus the sum of years of formal educationall speak Spanish in home household education level dictomous household members speak Spanishfloor material socio-economic status dictomous unimproved (dirt) or improved (concrete/tile)wall material socio-economic status dictomous unimproved (wood/cane) or improved (concrete block)household population density socio-economic status continuous number in household divided by number of roomschild under age 5 diarrheal health burden dictomous a child under 5 resides in the householdmore than one child under 5 diarrheal health burden dictomous more than one child under 5 resides in the householddiarrhea most persistent diarrheal health burden dictomous diarrhea is identified by the household as being one of the most persistent

diseases affecting the householdat least one day with diarrhea in pasttwo weeks

diarrheal health burden dictomous the household has suffered at least one day with diarrhea in the past two weeks

three or more days with diarrhea inpast two weeks

diarrheal health burden dictomous the household has suffered three or more days with diarrhea in the past twoweeks

water source independent variable dictomous unimproved/improved by WHOa standardssanitation independent variable dictomous unimproved/improved by WHO standardsadditional water treatment independent variable dictomous household performs water treatment in addition to the biosand filterpresence of soap in the home independent variable dictomous household has designated hand washing area with soapfilter operation and maintenance score independent variable ordinal filter operation and maintenance scored based on known failure modesaWHO, World Health Organization.

Environmental Science & Technology Article

dx.doi.org/10.1021/es303624a | Environ. Sci. Technol. 2013, 47, 1638−16451639

• Additional water treatment has a positive effect on filteroperation and maintenance.

• Increased household educational level has a positiveeffect on filter operation and maintenance.

• Proper filter operation and maintenance is expected tohave a negative effect on diarrhea burden due to theresults from case-control studies.10,11

3. METHODS3.1. Survey Instrument and Data Gathering. The survey

was a highly structured instrument used to identify specificattributes of individual households. The information gatheredconsisted of family demographics, age distribution, educationlevels, water sources and usage, sanitation practices, watertreatment practices, on-site water storage, hygiene practices,and diarrhea burden. An additional 15 questions were includedthat detailed filter installation, use, and status. All survey

respondents were informed of the purpose and consented tocontributing their information prior to beginning the survey.The sample was selected from a pool households of ten

communities by house number that previously participated in aBiosand filter program at least six months prior. Further detailsare available in the Supporting Information. All filters in thedata set were constructed by the same team of filter techniciansand a microbial water quality analysis was completed for asubset of the filters in participating households. The resultsshowed the filters constructed by the team were performingwithin the expected levels of baseline removal efficacy.18

Further details are also available in the Supporting Information.3.2. Observed and Latent Variables. An advantage of

SEM analysis is that models can be composed of both observedand latent variables. A latent variable represents a variablewhich cannot be directly observed. Therefore, severalobservable variables that are believed to be influenced by thelatent variable are used as surrogates. This characteristic of the

Figure 1. Measurement model − Rectangles represent indicators, ovals represent latent variables, e represents measurement error of the observedvariables. The overall figure is the initial measurement model. The objects filled in gray and associated pathways were eliminated from the finalmeasurement model. The observed correlation between Household Education Level and At Least One Child Under 5 has been omitted for figureclarity.

Environmental Science & Technology Article

dx.doi.org/10.1021/es303624a | Environ. Sci. Technol. 2013, 47, 1638−16451640

method is a distinct advantage when the system under studyhas difficult to measure factors.Table 1 presents a total of 20 observed indicators that were

extracted from the survey for each household, which includesixteen dichotomous responses, three continuous responses,and one ordinal measure. These 20 observed indicators wereestimated to impact the diarrhea health burden − the primarylatent variable of interest. The initial SEM model of diarrhealhealth burden is hypothesized to be a function of 5 other latentvariables and one observed variable. The 6 latent variables aswell as the observed variables that are influenced by the latentvariable are further described in the Supporting Information.3.3. Two-Step Analysis Approach. The analysis was

completed as a two-step process.15 First, a measurement modelwas created to determine if the selected indicators (Table 1)were appropriately represented by the relative latent variablesand to be sure that the latent variables were not highlycorrelated. The measurement model consisted of probitregressions that accommodate the categorical indicators andlinear regression for the continuous indicators all beingregressed by the continuous latent variables. Second, astructural model was defined to determine relationshipsamong latent and independent variables. The structuralmodel is a linear model to accommodate continuous observedand latent variables that are regressed on continuous latentvariables and continuous and binary observed variables. Amodel-generating approach was implemented to modify theoriginal hypothesized model when prudent. Details regardingevaluating the model fit are included in the SupportingInformation.

The MPlus 4.2 software package was used to build andevaluate the model.

3.4. Analysis. The confirmatory approach to evaluate latentvariables suggested by the hypotheses was completed with thedata available from household surveys. However, some of thedata was expected to not be appropriate for the analysis due tolack of variance in observed variables. Therefore, thedichotomous variables were further screened and variableswith less than 5% of the data assigned to one of the categorieswere eliminated from the measurement model of the associatedlatent variable constructs. Figure 1 shows the initial measure-ment model after these data have been removed. The indicatorvariables sanitation, water supply, soap used in the home andadditional water treatment lacked sufficient variance to describelatent variables in the measurement model but were still eligibleto be used as independent variables in the structural equationmodel.The model fit of the four remaining latent variables was

inadequate to be utilized in further analysis in its initialconfiguration; χ2 (39, n = 286) = 138.83, χ2/df = 3.56, p <0.001, CFI = 0.838, TLI = 0.826, and RMSEA = 0.095.Modification index Lagrange Multipliers were used to suggestimprovements to the model. A correlation between the latentvariable Household Education Level and the indicator At leastone child under age 5 in the home was included. Thismodification was supported in theory by local conditions andsocial structure. When households were surveyed, nodistinction was made in the age of the heads of householdfrom household to household. This is significant because onlyin the past 20 years has education beyond elementary levels

Figure 2. Initial hybrid model (structural and measurement models combined).

Environmental Science & Technology Article

dx.doi.org/10.1021/es303624a | Environ. Sci. Technol. 2013, 47, 1638−16451641

been available to residents. Therefore, younger heads ofhousehold with young children are more likely to have receivedhigher education, whereas older heads of household with olderchildren did not have the same opportunities.Additionally, a high correlation between two of the observed

variables, specifically At Least One Child Under Age 5 andMore Than One Child Under Age 5, effectively rendered thetwo variables as a single facet of the model. Therefore theobserved variable More Than One Child Under Age 5 waseliminated from the model. Additionally, the analysis yieldednonpositive-definite estimates for the factor loadings of thelatent variable of Household Water Storage. Nonpositivedefinite estimates are due to linear dependencies amongvariables in the data. Because of this characteristic in the data,the latent variable Household Water Storage and the associatedindicators were eliminated from the analysis.The resultant model, Figure 1 excluding gray objects, had a

nearly adequate fit to the data satisfying three of the four fitindices produced by the software package; χ2 (19, n = 286) =55.183, χ2/df = 2.90, p < 0.001, CFI = 0.937, TLI = 0.927, andRMSEA = 0.082. Additionally, the parameter estimates were allstatistically significant at the 0.05 level. The measurementmodel now describing the three latent variables and tenobserved variables was sufficient to move forward with thestructural equation analysis.Figure 2 presents the measurement model that was specified

to include the structural regression paths among latent variablesand additional observed independent variables of interest. Thestructural equation model regressed diarrhea health burden onthe latent variables Household Education Level, and Socio-

economic Status, and the observed independent variables thatwere not included in the final measurement model: FilterOperation and Maintenance, Adequate Sanitation, ImprovedWater Supply, Soap Present in the Home, and AdditionalWater Treatment. Additionally, Filter Operation and Main-tenance was a mediator in the model, as this variable wasregressed on the latent variables Household Education Leveland Socioeconomic Status as well as the independent observedvariables Soap Present in the Home, and Additional WaterTreatment. The model fit did comply with the expected modelevaluation criteria to provide a nearly adequate fit, χ2 (43, n =286) = 86.875, χ2/df = 2.02, p < 0.001, CFI = 0.922, TLI =0.911, and RMSEA = 0.060. However, the parameter estimatesfor Socio-economic Status, Additional Water Treatment, SoapPresent in the Home, and Adequate Sanitation on DiarrheaHealth Burden were nonsignificant at a 0.2 level. These pathswere deleted from the model incrementally to create the mostparsimonious and statistically defensible model possible thatalso conformed to the theoretical assumptions of the system.Initially, the path from Socio-economic Status to DiarrheaHealth Burden was removed because the p-value was the largestof the four nonsignificant parameters. Additionally, morepublished research exists on how water treatment, hygienepractices, and sanitation affect diarrhea than socio-economicstatus.1−3 Removing the pathway from Socio-economic Statusto Diarrhea Health Burden had little effect on the model fit andthe significance of the other nonsignificant parameter estimates.The remaining nonsignificant parameter values were removedfrom the model. The resultant model, Figure 3, had animproved fit to the data, χ2 (36, n = 286) = 70.791, χ2/df =

Figure 3. Final hybrid model (structural and measurement). The standardized and unstandardized (listed in parentheses) parameter estimates arelisted next to the associated pathway. * p < 0.05, ** p < 0.1, # p < 0.15, measurement error terms (e) were removed to reduce congestion. Weightadded to arrows for emphasis; color indicates direction of influence, red is negative influence, green is positive influence.

Environmental Science & Technology Article

dx.doi.org/10.1021/es303624a | Environ. Sci. Technol. 2013, 47, 1638−16451642

1.97, p < 0.001, CFI = 0.933, TLI = 0.920, and RMSEA =0.058. A review of the residuals showed that the modeladequately explained 35% of the total variance in FilterOperation and Maintenance and 26% of the total variance inDiarrhea Health Burden. Filter Operation and Maintenanceonly accounted for 7% of the variance in diarrhea burden.

4. RESULTS4.1. Filter Significance. In the structural model, Filter

Operation and Maintenance was regressed on four variables:two latent variables and two dichotomous observed variables.The unstandardized parameter estimates yield the change in theoutput variable as a function of a one-point change in the inputvariable. For example, a one-point increase in HouseholdEducation Level results in a 0.155 decrease in Filter Operationand Maintenance.Household Education Level had a negative effect on proper

Filter Operation and Maintenance. The negative relationship ofeducational level and filter maintenance suggests that personswith lower levels of education are more likely to maintain theirfilter properly. This result is opposed to the original hypothesis,and counters the accepted convention that higher levels ofeducation leads to improved health.5 Streatfield (1990) foundthat education matters very little when specific knowledge of anissue is made available19 suggesting formal education has littleto do with comprehension and compliance with filtermaintenance tasks. This finding could be more indicative ofindividuals with lower education level have few opportunitiesavailable outside of the home which affords additional time tocare for the filter than the knowledge to carry out maintenance.Socio-economic Status had a negative effect on Filter

Operation and Maintenance. This estimate implied thathouseholds of lower socio-economic status, as defined byquality of home construction materials, typically cared for theirfilter better than household with better homes and livingconditions. This result was contrary to the originally proposedhypothesis. However, it should be noted that there was noliterature on household characteristics impacting proper filteruse and maintenance. This negative effect could suggest twopossible characteristics of the study population. One is thathouseholds with higher socio-economic status had enoughdisposable income to buy bottled water and therefore do notstrictly manage their filter. A second is that households withlower socio-economic status gained a sense of pride in owningtheir filter, which was valued for providing clean water and wasone of the few durable products owned by the home. Anotherpossibility is that the latent variable construct created for thisanalysis does not fully represent socio-economic status as itonly considers characteristics of the family dwelling and notspecific family owned items or income. Frost (2005) whenstating a high socio-economic status is protective of health,included aspects of sanitation and water supply as criteria forhigher status.6 An inventory of family owned items would havebeen a more robust measure of socio-economic status but localprogram staff suggested against this query due to personalsecurity concerns. Also, Household Education Level and Socio-economic Status were not significantly correlated in the analysiswhich is surprising considering Frost (2005) observed thateducation is a vehicle for socioeconomic development6 and,conversely, Bicego (1993) stated that higher levels ofsocioeconomic status create more opportunities for educa-tion.20 This pattern was not followed in the study region as allcommunities had elementary schooling available free of cost

which could eliminate any link between socioeconomic statusand education at lower levels. When students wish to enrollbeyond elementary school, they often must travel to anothercommunity for secondary education, which requires trans-portation and daily food costs. However, probably due to thelow number of heads of household which progressed beyondelementary school (approximately 13% of sample), theexpected correlation of socio-economic status and highereducation was not observed.Additional Water Treatment had a negative effect on proper

Filter Operation and Maintenance. This indicated thathouseholds which complete additional point-of-use watertreatment were less likely to maintain their filter properly.This result was contrary to the originally proposed hypothesis,which stated that the household taking the time for additionalwater treatment would be carefully managing their filter as well.Upon returning to the data set, it was found that, of the 70households that no longer utilized the filter (thereby scoring azero on the filter operation and maintenance scale), 48households (69%) also completed additional water treatment.This would suggest households that do not use the filter oftencomplete additional water treatment to improve water quality,hence the negative effect that additional water treatment has onfilter operation and maintenance. These households preferother, possibly simpler, means of treatment due to eitherperceived or actual filter failure.Soap Present in Home has a positive effect on proper Filter

Operation and Maintenance. The direction and magnitudesuggest that households who practice keeping soap available foruse also tend to maintain their filter properly. This agreed withthe original hypothesis and suggests that best householdpractices regarding hygiene practices reinforce proper filteroperation and maintenance. Therefore, households that arealready accustomed to completing personal hygiene practicesmay also be willing to accept new solutions to improve theirwater quality.

4.2. Diarrhea Significance. In the structural model,Diarrhea Health Burden was regressed on three variables; onelatent variable, one continuous observed variable, and onebinary observed variable. Household Education Level had anegative effect on the severity of Diarrhea Health Burden. Asthe education level of the household increases, there was adecrease in diarrhea health burden. This effect was expected, asprevious research had shown a negative relationship involvinghousehold education level and health burden on the home.4,5

Direct negative correlation to years of maternal education andchildhood mortality have been determined and credited therelationship to mothers having better access to public healtheducation due to their experience in the academic environ-ment.5 This model is consistent with previous research findings.Filter Operation and Maintenance had a negative effect on

the severity of Diarrhea Health Burden. This effect wasexpected, as previous research had shown that the Biosand filteris effective in reducing diarrheal disease in the field.10,11 Themetric used in evaluating filter operation and maintenance is ameasure of how well the filter is maintained, which is directlyrelated to filter performance.7 Therefore, the model suggests abetter maintained filter results in lower diarrhea health burden.Improved Water Supply had a negative effect on the severity

of Diarrhea Health Burden. This suggests that a householdhaving an improved water supply will have a lower diarrheahealth burden than households with unimproved water

Environmental Science & Technology Article

dx.doi.org/10.1021/es303624a | Environ. Sci. Technol. 2013, 47, 1638−16451643

supplies. This conclusion is corroborated by numerous studiesand resources.2,21,22

All model-derived impacts associated with the observedparameter estimates on Diarrhea Health Burden complied withexisting published literature. These results would suggest thatthe model is producing reasonable parameter estimates with theSEM methodology.4.3. Indirect and Total Effects. Even though Socio-

economic Status, Additional Water Treatment, and SoapPresent in Home have no statistically significant direct effecton Diarrhea Health Burden, an indirect effect for each can becalculated through the mediator Filter Operation andMaintenance. An indirect effect is the influence a variable hason another when the only connection is facilitated though amediator variable in a causal chain.16 The product of theparameter estimate of the variable on the mediator and theparameter estimate of the mediator on the final outcomevariable are used to calculate the indirect effect. Table 2 shows

the sum of the direct and indirect effects of the causal variableson one observed and one latent variable known as the totaleffect. In addition to the variables, which have no direct effecton Diarrhea Health Burden, the total effect was calculated forHousehold Education Level as this effect is mediated throughFilter Operation and Maintenance. The total effect ofHousehold Education Level differs very little from the directeffect previously mentioned.The total effect of Socio-economic Status on Diarrhea Health

Burden is positive suggesting households with higher statuslevels tend to have poorer health. This unexpected outcomecould be due to survey respondent bias; further comments onbias are included in the Supporting Material. The total effect ofAdditional Water Treatment on Diarrhea Health Burden waspositive as well suggesting the practice of Additional WaterTreatment could be insufficient to prevent users from comingin contact with water-borne disease. The total effect of SoapPresent in Home on Diarrhea Health Burden was negative

suggesting there could be a combined effect of proper filter useand proper hygiene practices to reduce diarrheal diseaseburden, which is supported by the literature.23

As previously stated, the residuals showed the modeladequately explained 35% of the variance in filter operationand maintenance and 7% of the variance in diarrhea healthburden. The global measures of fit were previous deemednearly adequate and the significant p-values for the parameterestimates affirm the usefulness of the model. Therefore, FilterOperation and Maintenance and Diarrhea Health Burden dohave influences not explained by the model. This unexplainedvariance could be due to exclusion of important parameters. Forexample, other factors that could account for a portion of theunexplained variance in Filter Operation and Maintenanceinclude the size of family or time spent in the home.Additionally, food preparation techniques or a chronic ailmentcould be responsible for unexplained variance in DiarrheaHealth Burden. Of the two variables, Diarrhea Health Burdenhas a lower level of explained variance. The model doesillustrate the effect which the various documented factorsinfluence diarrhea in the home but only a small amount. Eventhough this effect is small, the magnitude is still significant andattributable to a reduction in diarrhea.

4.4. Discussion. This study is the first to use SEM toinvestigate the relationships among household demographics,environmental health factors, Biosand filter operation andmaintenance, and household diarrhea health burden. Theresults illustrate how demographics, infrastructure, andpractices within the home have a significant effect on properoperation and maintenance of the Biosand filter. Additionally,these effects on filter operation and maintenance subsequentlyaffect diarrhea health burden in conjunction with otherindependent effects. The community is a complex system ofinteractions which directly and indirectly influence the health ofits residents. Policy makers and development practitioners mustrecognize that single target interventions (e.g., improving waterquality) have a limited influence on the entire system.The multiple hypotheses originally defined in the study to

describe the various effects of environmental health factors andhousehold demographics on diarrhea burden failed to berejected and confirmed the findings of previous studies. Thehypotheses defined to describe the relationships betweenenvironmental health factors and household demographics onproper filter operation and maintenance were not all confirmed.To date, no other study has been published to demonstrate arelationship between socio-economic status, household educa-tion level, or soap present in the home, and additional watertreatment on proper operation and maintenance of the Biosandfilter. These hypotheses were proposed based on anticipatedresults rather than previous findings.It should be noted that the model did not fully explain the

variance present in filter operation and maintenance anddiarrhea health burden. Factors exterior to the model arecausing this variance and require further investigation toidentify. Despite these unmeasured external factors, the effect ofa properly maintained Biosand filter was shown to be significantand protective against diarrhea. Therefore, the filter deserves tobe recognized as an effective tool in diarrhea reduction in anactual, developing-world scenario. This reaffirms the idea thatthere is no single factor or intervention which determineshealth, but rather many factors must work in concert toimprove health.

Table 2. Direct, Indirect, and Total Effects of CausalVariables on Filter Operation and Maintenance andDiarrhea Health Burden with Standardized ParameterEstimates

causal variable

filter operationand

maintenance

diarrheahealthburden

direct effect household education level −0.155 −0.170socio-economic status −0.122additional water treatment −0.527soap present in home 0.139improved water supply −0.169filter operation andmaintenance

−0.119

indirect effect household education level 0.018socio-economic status 0.015additional water treatment 0.063soap present in home −0.017

total effect household education level −0.155 −0.152socio-economic status −0.122 0.015additional water treatment −0.527 0.063soap present in home 0.139 −0.017improved water supply −0.169filter operation andmaintenance

−0.119

Environmental Science & Technology Article

dx.doi.org/10.1021/es303624a | Environ. Sci. Technol. 2013, 47, 1638−16451644

While development practitioners working in the field canexpect to see modest reductions in diarrhea due to a filterintervention, the larger reductions observed in highly structuredand monitored interventions may not be typical.10,11 Previousstudies have created an ideal intervention environment viaconsistent home visits, filter monitoring, and prompting onoperation and maintenance, which may have createdHawthorne effects. Further, Aiken (2011) in an evaluation ofthe Biosand filter in the Dominican Republic found evidence tosuggest households participating in longitudinal research trialshave higher rates of continued filter use than household notparticipating in trials.12

Comparable reductions in diarrhea may not be possible if adevelopment authority is not equipped to provide similaroversight. However, this study has exhibited that the Biosandfilter is capable of reducing diarrhea proportionally to thedegree at which the filter is properly used and maintained.Therefore, development practitioners ought not to bediscouraged when a filter intervention does not provide thesame magnitude of results as a highly monitored, scientificstudy. Rather, practitioners should recognize the significant andmodest gains from the Biosand filter and look for additionalinterventions to further improve health.

■ ASSOCIATED CONTENT

*S Supporting InformationDetails regarding the study site, data gathering methodology,the survey instrument, and definitions of variables. Thismaterial is available free of charge via the Internet at http://pubs.acs.org.

■ AUTHOR INFORMATION

Corresponding Author*E-mail: divelbdw@mail.uc.edu. Fax: 513 556 4171.

NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTS

This material is based on work supported by the NationalScience Foundation Graduate Research Fellowship Grant1102690 and the John A. and Susan Mathes Endowmentfrom the Missouri University of Science and Technology.Additionally, the authors acknowledge the extensive in-countylogistical assistance by Gerber Esau Perez, water quality testingassistance by Tomas Leon, and data analysis conversations withDiana Castellanos-Bucey.

■ REFERENCES(1) Diarrhhoea: Why Are Children Still Dying and What Can Be Done:The United Nations Children’s Fund (UNICEF)/World HealthOrganization (WHO). Office of Communication: New York, 2009.http://www.unicef.org/health/files/Final_Diarrhoea_Report_October_2009_final.pdf(2) Acute Diarrhea in Adults and Children: A Global Perspective. WorldGastroenterology Organisation (WGO). 2008. http://www.w o r l d g a s t r o e n t e r o l o g y . o r g / a s s e t s / e x p o r t / u s e r fi l e s /Acute%20Diarrhea_long_FINAL_120604.pdf(3) Black, R. E.; Morris, S. S.; Bryce, J. Where and why are 10 millionchildren dying every year? The Lancet. 2003, 361, 2226−2234.(4) Shukr, R. I.; Ali, S.; Khanum, T.; Mehmood, T. Is there a linkbetween maternal illiteracy and childhood diarrhea? Rawal MedicalJournal. 2009, 34 (2), 199−202.

(5) Cleland, J. G.; van Ginneken, J. K. Maternal education and childsurvival in developing countries: The search for pathways of influence.Social Science & Medicine 1988, 27 (12), 1357−1368.(6) Frost, M. Maternal education and child nutritional status inBolivia: Finding the links. Social Science & Medicine 2005, 60 (2),395−407.(7) Fewster, E.; Mol, A.; Wiesent-Brandsma, C. The Long-TermSustainability of Household Biosand Filtration. 30th WEDC Interna-tional Conference, 2004, Vientiane, Lao PDR.(8) Palmateer, G.; Manz, D.; Jurkovic, A.; McInnis, R.; Unger, S.;Kwan, K.; Dutka, B. Toxicant and parasite challenge of Manzintermittent slow sand filter. Environ. Tox. 1999, 14 (2), 217−225.(9) Stauber, C.; Elliott, M.; Koksal, F.; Ortiz, G.; DiGiano, F. A.;Sobsey, M. Characterisation of the biosand filter for e. coli reductionsfrom household drinking water under controlled laboratory and fielduse conditions. Water Sci. Technol. 2006, 54 (3), 1−7.(10) Stauber, C.; Ortiz, G.; Loomis, D.; Sobsey, M. A randomizedcontrolled trial of the concrete Biosand filter and its impact ondiarrheal disease in Bonao, Dominican Republic. The American Journalof Tropical Medicine and Hygiene 2009, 80 (2), 286−293.(11) Tiwari, S.; Schmidt, W.; Darby, J.; Kariuki, Z.; Jenkins, M.Intermittent slow sand filtration for preventing diarrhoea amongchildren in Kenyan households using unimproved water sources:Randomized controlled trial. Tropical Medicine & International Health2009, 14 (11), 1374−1382.(12) Aiken, B. A.; Stauber, C. E.; Ortiz, G. M.; Sobsey, M. D. Anassessment of continued use and health impact of the concrete biosandfilter in Bonao, Dominican Republic. American Journal of TropicalMedicine and Hygiene 2011, 85 (2), 309−317.(13) Fabiszewski de Aceituno, A. M.; Stauber, C. E.; Walters, A. R.;Meza Sanchez, R. E.; Sobsey, M. D. A randomized controlled trial ofthe plastic-housing biosand filter and its impact on diarrheal disease inCopan, Honduras. American Journal of Tropical Medicine and Hygiene2012, 86 (6), 913−921.(14) Stauber, C. E.; Printy, E. R.; McCarty, F. A.; Liang, K. R.;Sobsey, M. D. Cluster Randomized Controlled Trial of the PlasticBioSand Water Filter in Cambodia. Environ. Sci. Technol. 2012, 46,722−728.(15) Schumacher, R. E.; Lomax, R. G. A Beginner’s Guide to StructuralEquation Modeling; Lawrence Erlbaum Associates: Mahwah, N.J., 2004.(16) Kline, R. B. Principles and Practice of Structural EquationModeling; Guilford Press: New York, NY, 1998.(17) Vanderzwaag, J.; Atwater1, J.; Bartlett, K.; Baker, D. Fieldevaluation of long-term performance and use of biosand filters inPosoltega, Nicaragua. Water Quality Research Journal of Canada 2009,44 (2), 111−121.(18) Sobsey, M. D.; Stauber, C. E.; Casanova, L. M.; Brown, J. M.;Elliott, M.A. Point of use household drinking water filtration: Apractical, effective solution for providing sustained access to safedrinking water in the developing world. Environ. Sci. Technol. 2008, 42,4261−4267.(19) Streatfield, K.; Singarimbun, M.; Diamond, I. Maternaleducation and child immunization. Demography 1990, 27 (3), 447−455.(20) Bicego, G. T.; Boerma, J. T. Maternal education and childsurvival: A comparative study of survey data from 17 countries. SocialScience & Medicine 1993, 36 (9), 1207−1227.(21) Fewtrell, L.; Kaufmann, R.; Kay, D.; Enanoria, W.; Haller, L.;Colford, J. Water, sanitation, and hygiene interventions to reducediarrhoea in less developed countries: A systematic review and meta-analysis. The Lancet - Infectious Diseases 2005, 5 (1), 42−52.(22) Ozkan, S.; Tuzun, H.; Gorer, N.; Ceyhan, M.; Aycan, S.;Albayrak, S.; Bumin, M. Water usage habits and the incidence ofdiarrhea in rural Ankara, Turkey. Transactions of the Royal Society ofTropical Medicine and Hygiene 2007, 101 (11), 1131−1135.(23) Ejemot R. I.; Ehiri J. E.; Meremikwu.M. M.; Critchley, J. A.Hand washing for preventing diarrhoea. Cochrane Database ofSystematic Reviews. 2008, Issue 1. Art. No.: CD004265. DOI:10.1002/14651858.CD004265.pub2.

Environmental Science & Technology Article

dx.doi.org/10.1021/es303624a | Environ. Sci. Technol. 2013, 47, 1638−16451645