Statistical analyses.

All calculations were performed using Stata^® SE version 14.1 (StataCorp LP, College Station, TX). To mirror the UDHS complex sampling design, calculations were performed using Stata's SVY module. We stratified the analyses a priori by rural/urban residence and child age group (0 to < 2 years and 2 to < 5 years). Bivariate associations were evaluated using unweighted Student's t tests, Pearson's correlations, contingency tables, and analysis of variance. To obtain P values corrected for the complex survey design, linear and logistic regressions were performed using Stata SVY coding. Differences in medians were tested using weighted Somer's D adjusted for household clustering. Two-tailed P values less than 0.05 were deemed statistically significant.

Weighted multiple linear regression models were constructed using SVY:REG Stata coding to account for household clustering and stratification. Potential covariates for multivariable regressions were based on their demonstrated associations with child HAZ. First, household ownership for each type of animal was included in the model; households with horses/donkeys/mules were excluded from the analyses because none was recorded in urban areas and less than 0.1% were recorded in rural areas. To arrive at the final models, six categories of variables, which included WASH, child, maternal, household, geographic, and dietary characteristics, were tested as confounders of the relationship between child HAZ and domestic animal ownership using the 10% criterion of confounding.