2.7.2. Demographic and Contextual Covariates

The demographic profile of the countries was summarized using mean and standard deviation (SD) for continuous data and count (percentage) for categorical data. The age of the child was computed by subtracting the date of birth of the child from the date of administration of the first application of the measure. Decimal age was used throughout, whereby 1 year is equal to 1, 6 months is equal to 0.5 and 3 months is equal to 0.25, and so on. Weight-for-age *z*-scores (WAZ), height-for-age *z*-scores (HAZ) and weight-for-height *z*-scores (WHZ) were constructed according to the WHO 2006 standards [24,25]. They were computed in R using code provided from WHO Anthrostat [26]. In order to make the FCI score comparable across different ages, a generalized partial credit model (GPCM) was used in the R package MIRT [27]. Maternal education was divided into four main categories: no school, primary only, secondary only, and above secondary. In order to make the responses between the DHS wealth index data comparable across countries, we also constructed a two-parameter logistic IRT model using MIRT [27] to create a socioeconomic status (SES) score. The generation of the FCI and SES scores are detailed elsewhere [28].

#### 2.7.3. Missing Data

Missing IYCD item responses (1.6%) were not imputed as the IRT model uses full maximum likelihood estimation. Missing covariate data were imputed using the R package MICE [29]. The numbers of missing data points for covariates were: HAZ *n* = 29; WAZ *n* = 23; maternal education *n* = 5; sex *n* = 1; FCI *n* = 6; SES *n* = 6; urban/rural *n* = 0.
