A novel household water insecurity scale: Procedures and psychometric analysis among postpartum women in western Kenya

Our ability to measure household-level food insecurity has revealed its critical role in a range of physical, psychosocial, and health outcomes. Currently, there is no analogous, standardized instrument for quantifying household-level water insecurity, which prevents us from understanding both its prevalence and consequences. Therefore, our objectives were to develop and validate a household water insecurity scale appropriate for use in our cohort in western Kenya. We used a range of qualitative techniques to develop a preliminary set of 29 household water insecurity questions and administered those questions at 15 and 18 months postpartum, concurrent with a suite of other survey modules. These data were complemented by data on quantity of water used and stored, and microbiological quality. Inter-item and item-total correlations were performed to reduce scale items to 20. Exploratory factor and parallel analyses were used to determine the latent factor structure; a unidimensional scale was hypothesized and tested using confirmatory factor and bifactor analyses, along with multiple statistical fit indices. Reliability was assessed using Cronbach’s alpha and the coefficient of stability, which produced a coefficient alpha of 0.97 at 15 and 18 months postpartum and a coefficient of stability of 0.62. Predictive, convergent and discriminant validity of the final household water insecurity scale were supported based on relationships with food insecurity, perceived stress, per capita household water use, and time and money spent acquiring water. The resultant scale is a valid and reliable instrument. It can be used in this setting to test a range of hypotheses about the role of household water insecurity in numerous physical and psychosocial health outcomes, to identify the households most vulnerable to water insecurity, and to evaluate the effects of water-related interventions. To extend its applicability, we encourage efforts to develop a cross-culturally valid scale using robust qualitative and quantitative techniques.

characteristics as those used in the third phase for survey administration, Activities G and H 1 4 1 (n=241 and n=186, respectively). Participants were briefly interviewed and lent digital cameras to take photographs of water related experiences. A second individual interview explored photographs and was followed by FGDs on most common emergent themes.
To explore experiences of household water use, acquisition, and insecurity.
Non-cohort Kenyan women, n=20 07/2015-10/2015 C. The Delphi Method (S1 Fig., S1 Table) b International experts on water and food insecurity were purposively selected to achieve a range of disciplines and geographic areas and asked to participate in online iterative surveys about HHWI Although the results of our Phase 1 are presented elsewhere to avoid an excessively long based on the size of the storage containers and the amount of water in the container. For instance, 2 4 0 a half-full 20-litre jerry can was measured as 10 litres of stored water. We also measured the 2 4 1 amount of water used daily by the household in litres based on estimates of the amount of water were calculated between cumulative daily recall and responses from the 31 st day. Quantitative data analyses were conducted in six phases including descriptive analyses, 2 5 9 item reduction, extraction of factors, tests of dimensionality, scale reliability, and validity (  To determine the correlation between individual scale items with the sum score of all scale items.
Estimated adjusted item-total correlation coefficients, help to determine which items to drop. [61,63,64]

Item Communalities
To determine the measurement error in each item or the true score variance.
Estimated using principal axis factoring. [62] Kaiser-Meyer-Olkin (KMO) Test for sampling adequacy To measure the proportion of common variance among items and determine whether the data is suitable for factor analysis.
Estimated the sampling adequacy for each item in the model and for the complete model. KMO values between 0.8 and 1 indicate the sample is adequate. [65]

Bartlett Test of Sphericity
To compare the observed correlation matrix to the identity matrix.
Tested the null hypothesis that the correlation matrix has an identity matrix. [66-68]

C. Extraction of Factors
Exploratory Factor Analysis (latent Structure) To measure the structure of a set of observed variables and identify the subset of variables that corresponds to each of the underlying dimensions.

Parallel Analysis
To identify the possible number of factors that can be developed from the data.
Estimated number of identifiable factors from scale items. This was a form of sensitivity analysis to the exploratory factor analysis. [73,61,64,70]

Model Fit Assessment
To determine the fitness of both factor and parallel analyses to the data.
Examined model fit indices against acceptable thresholds (S3 Table).
[74-81] With bifactor analysis, the factor loadings of the general factor were compared to the group factors to help determine the dimensionality of the scale.

Model Fit Assessment
To determine the fitness of both confirmatory factor analysis and bifactor modeling solutions.

Intra-respondent reliability
To assess the stability and consistency of responses by respondents on scale items.
Correlated the sum score of daily retrospective responses on HHWI items for 30 days with scores on a 30-day recall. [87] Coefficient alpha To assess the internal consistency of the scale. i.e., the degree to which the set of items in the scale co-vary, relative to their sum score.
Calculated Cronbach's alpha for scale items at 15 months postpartum and 18 months postpartum. [88]

Coefficient of stability
To assess the degree to which the participant's performance is repeatable; i.e. how consistent their scores are across time.
Estimated the coefficient of stability via Testretest reliability. This was indexed by the correlation coefficient of two assessments of HHWI at two different time points. [63,64,89]

Predictive validity
To determine the degree to which test scores predict criterion measurements to be made in the future.
Estimated the association between HHWI and maternal stress to food insecurity scores. [63,64]

Convergent validity
To examine the evidence that the same concept measured in different ways yields similar results.
Estimated the correlation between HHWI and water quality (E.coli concentrations), time to water collection, amount spent on water for household, and season of data collection. [89-91]

Discriminant validity
To examine the evidence that the concept measured is different from other closely related concepts.
Estimated the correlation between HHWI and per capita household water use. Indicated by predictably low correlations between HHWI and other measures.
[89-91] Differentiation by 'known groups' To examine the degree to which the concept measured behaves as expected in relation to 'known groups'.
Estimated a differential test of means for maternal HIV status, season, water quality, and source of drinking water.
[ [89][90][91] A. Descriptive analyses 2 6 8 First, we estimated proportions, means, and standard deviations of the HHWI module and 2 6 9 participant characteristics. Although there were 5-response categories for the scale items 2 7 0 originally, the sample distribution was skewed to the right (<5%) for "always" for each item. Therefore, "often" and "always" were collapsed for subsequent analyses.  to determine the optimal number of factors that fit the data at 15 months postpartum Horn's PA, which employed the maximum likelihood (ML) estimator. For sensitivity analysis, we employed principal axis factors. both traditional factor and parallel analyses (S3 Table). The fit indices included Chi-square test Square Residual (SRMR≤0.08) [74][75][76][77][78][79][80][81]. Consistent with the factor structure of previous 2 9 6 household water insecurity scales elsewhere [2][3][4]36,40], we assumed our model will produce 2 9 7 similar factor structure for our scale. In order to test the factor structure obtained from the EFA, a test of scale dimensionality was To determine whether to retain a construct as unidimensional or multidimensional, the 3 0 9 factor loadings from the general factor are compared to those from the group factors (sub scales) Square Residual [74-81] (S3 Table). We assessed intra-respondent reliability of scale items retrospectively by comparing daily 3 1 6 recall across 30 days with the sum score of a retrospective recall on the 31 st day. This was to 3 1 7 assess the stability and consistency of responses on scale items.

1 8
The reliability of the scale itself was estimated using coefficient alpha and the coefficient We used predictive (criterion), construct (convergent and discriminant) validity and 3 2 6 differentiation between 'known groups' to assess scale validity. Predictive (criterion) validity 3 2 7 was assessed by examining the associations between HHWI and perceived maternal stress as 3 2 8 well as food insecurity [56,57].

2 9
Convergent validity was measured against time to and from water source and amount of 3 3 0 money spent on purchasing water in the past month. We calculated Pearson product-moment from household water use.
As a final measure of validity, we assessed the scale score by differentiating the position 3 3 8 of 'known groups'. In other words, we expected to have significantly higher HHWI scores for 3 3 9 participants whose water was contaminated with Escherichia coli (E.coli), were HIV positive, 3 4 0 those who used unimproved sources of water, and during the dry season. We used Student's t- Formative work in Phase 1 resulted in the creation of 29 HHWI questions (Activity E, 3 5 0 Table 1). The cognitive interviews (Activity F) indicated people were able to understand the interviewers on how to ask questions without ambiguity and the avoidance of leading prompts. The response options that were considered appropriate for a 4-week recall period were "never" 3 5 5 P 2 0 (0), "rarely" (meaning 1-2 times), "sometimes" (3-10 times), "often" (10-20 times) and "always" 3 5 6 (>20 times). Of the 241 participants who were interviewed at 15 months postpartum, the mean 3 5 9 household size was 3.5 with a range of 1-12 members ( Table 3). The majority of women (90.5%) 3 6 0 interviewed were primiparous, 51.5% were HIV positive, and the mean age was 25 (range 18-39)  (Table 3).

Costs:
Amount spent per month on water (USD 4 ) by women with no access to water in household (n=130)  Mean total time per week spent in water acquisition among women with no access to water in household (hours) (0-21) 5.6 (4.8)

Use:
Per capita total daily water use in liters 5 (20. Notes. 1 HIV-infected women were oversampled to achieve 1:1 serostatus ratio; 2 Rainy months in this dataset were May and October; 3 Unimproved water sources include unprotected dug well, unprotected spring, surface water; Improved water source include piped water, stand pipe, bore hole, protected dug well, protected spring, rain water; 4 USD=United States Dollar converted in May 2016; 5 These data were collected in a subset of 27 households (Activity J); 6 the presence of E.coli was tested using compartment bag test assay.

Water access and use
3 7 2 Of the 241 participants interviewed, nearly half (41.0%) used drinking water from 3 7 3 unimproved sources, and 53.9% did not have access to water in their households or compounds.

7 4
Of the women who had access to water in their households, 8.8% were unimproved sources. for their households, for a mean of 5.6±4.8 hours per week. In 27 households in which we had 3 8 0 data to assess water use and microbial analysis, a mean of 65.5 liters of water was used daily by 3 8 1 households, 6.5 liters were stored for drinking, and a mean of 70.1 were stored for other uses. E.coli in stored drinking water (Table 3).  Notes: "Never" =0, "Rarely"=1-2 times in prior 4 weeks, 'Sometimes"=3-10 times in prior 4 weeks, "Often/always" in prior 4 weeks=11+ times, Ever ≥ 1 in prior 4 weeks; polychoric correlation coefficients=inter item correlation; polyserial correlation coefficients=item-total correlation 3 9 4 The most severe manifestations of water insecurity, such as sleeping thirsty and having no water 3 9 5 in the household whatsoever, were least endorsed (23. 8% and 19.2%) in this population (Fig. 1  about having enough water and drinking water that was considered to be unsafe, were 3 9 8 considerably more common (46.3% and 44.9% respectively), with 3.7% experiencing these 3 9 9 events often or always (Fig.1, Table 4). In total, nine scale items were dropped from the 29-question survey (Fig. 1) We then investigated inter-item (polychoric) and item-total (polyserial) correlations 4 1 2 (Table 4). Inter-item correlations were strong, ranging from 0.67 to 0.97 for the remaining 20  To understand the latent factor structure of our items, we used EFA and the Guttman- Kaiser rule to extract two factors from the data with initial eigenvalues of 15.86 for factor one 4 2 3 and 1.02 for factor two ( scree plots in both analyses showed a single dominant factor (S2 Fig). Specifically, the line for suggested a one-factor solution was most appropriate.  An evaluation of the factor loadings associated with the eigenvalues produced two 4 3 6 solutions, a one-factor model and two-factor model ( Table 6). An examination of the factor 4 3 7 loadings for the two-factor model showed three statistically significant cross loading items  However, the scores on the dominant factor were comparatively higher than the second factor,  All four model fit indices used in this study showed very strong support for a single  Table).

7
Therefore, we selected a unidimensional scale with 20 items. All factor loadings for the recommended threshold of 0.40 (Table 6). Based on the results, we hypothesized that the 4 5 0 remaining 20 items would represent a single construct i.e. a unidimensional scale (Table 7).  For each item, the questions followed the same format "In the last 4 weeks, how frequently…" 1 Did you worry you would not have enough water for all of your household needs? 2 Did you feel angry or frustrated that you would not have enough water for all of your household needs? 3 Did you worry about the safety of the person getting water for your household? 4 Has the time spent fetching water prevented anyone in your household from earning money? 5 Has the time spent fetching water prevented you or anyone in your household from caring for your children? 6 Has anyone in your household asked to borrow water from other people? 7 Has there not been enough water in the household to wash clothes? 8 Have you missed meetings in your community (church, funerals, community meetings, etc.) because there wasn't enough water? 9 Have you missed meetings in your community (church, funerals, community meetings, etc.) because you lacked water to take a bath and you felt too dirty to go? 10 Have you or anyone in your household had to change what was being cooked because there wasn't enough water? 11 Did you or anyone in your household had to go without washing hands after defecating, changing diapers, or other dirty activities because you didn't have enough water? 12 Did you not have enough water to wash your children's face and hands? 13 Did you or anyone in your household have to go without washing their body because there wasn't enough water? 14 Did you or anyone in your household want to treat your water, but couldn't? By treat I mean boiling, using chemicals to treat, or other ways you make your water safe to use or drink. 15 Did you or anyone in your household actually had to drink water that you thought was unsafe? 16 Did you have problems with water that caused arguments/trouble with neighbors or others in the community? 17 Has there not been as much water to drink, as you would like for you or members of your household? 18 Have you or anyone in your household not had enough water to take medications? 19 Have you or anyone in your household gone to sleep thirsty? 20 Have you had no water whatsoever in your household?
Notes: For each question, participants were asked to respond to one of the following options Never (0), Rarely (1-2 times in prior 4 weeks), Sometimes (3-10 times in prior 4 weeks), Often (11-20 times in prior 4 weeks), Always (above 20 times in prior 4 weeks). Questions were asked from the least to the most severe manifestations of water insecurity. We then tested this hypothesis using confirmatory factor model and a bifactor model on  from the confirmatory factor analysis were all significant at p<0.001 (Fig 2A). The bifactor 4 6 0 model focused on accounting for unrecognized distortions created by the three items with cross 4 6 1 loadings ( Fig 2B). Reise et al. [85] suggest that where the factor loadings of the 4 6 2 general/dominating factor are greater than the subfactor, a unidimensional factor is implied. In postpartum, the mean of HHWI was 9.5 ± 12.2 (Mean ± SD), with a range of 0-59 (Fig 3). At 18 4 7 6 months postpartum, the mean of HHWI was 10.1 ± 12.4 (Mean±SD), with a range of 0-57. Our test of intra-respondent reliability of the scale questions produced a strong  To assess predictive criterion validity, we regressed maternal stress and food insecurity 4 9 7 on HHWI score, and found HHWI to be significantly positively correlated with increased To assess discriminant validity, we tested if there would be a low or no relationship 5 0 7 between HHWI and per capita household water use. This relationship was not statistically 5 0 8 significant (r=0.12. 95% CI, -0.30-0.50, p=0.59).

0 9
We also examined the differences between 'known groups' on HHWI scores. Our results

1 0
showed that although the magnitude of the means for the groups measured was in the expected is valid and reliable for the assessment of HHWI among postpartum women in Nyanza region 5 2 2 (Table 7). Our final scale is composed of items measuring different aspects of water insecurity, yet 5 2 4 its latent structure reflects the central assumption of unidimensionality (Tables 5 & 6). This was not assessed in any of these studies with the statistical rigor used here; we encourage future 5 2 9 studies to draw from the methods outlined here for comparable assessment of dimensionality.

3 0
The HHWI scale performed well in terms of recall bias, with a correlation coefficient of makes it impossible to compare our test-retest results to other existing HHWI scales [1].

3 9
Validity was supported in a number of ways. HHWI was positively associated with food 5 4 0 insecurity and maternal stress, indicating predictive validity. This finding also affirms the fact 5 4 1 that water insecurity is inextricably linked with food insecurity and has significant implications water insecurity and maternal stress also points to the psychosocial effects that water insecurity influences of food and water insecurity on health and well-being. In sum, our 20-item HHWI scale (Table 7) is a reliable and well-validated measure of 6 0 9 HHWI among women in Nyanza, Kenya. The implementation of this scale will make it possible 6 1 0 to understand and quantify both the multi-factorial causes and consequences of HHWI (physical, 6 1 1 mental, economic, social and nutritional). Also, the use of the scale will enable the monitoring of in need of support to increase household water security.