The study sample frame was generated from community clinic health assistant records, which have the advantage of being the centralized government document of record containing the population frame for all households with children under five years of age. The health assistant dataset included data for all three upazilas of interest. Based on an examination of the extant health assistant dataset described above, the study excluded 11 unions (out of a total of 41 unions) located in these three upazilas. Six of the unions were removed because data were not available. A further five unions were removed because they only had one community clinic (the study design requires each union to have at least one community clinic for each of the two treatment conditions). The final sampling frame included 78 community clinics located in 30 unions. The sample frame was generated within each community clinic, and the units in the frame are households with children aged between 3 months and 18 months of age, which were situated in the selected community clinics' catchment areas during the period of the baseline data collection. The rationale for restricting the frame to households with children aged three months or older was that the main developmental assessment tool chosen for the evaluation-the BSID-III-has not been previously validated on children under the age of three months in Bangladesh. Early child development specialists consider the BSID-III test to be the gold standard assessment of development for children under 42 months of age, and it has been adapted by the team for use in Bangladesh. Because the BSID-III test is only valid for children under 42 months of age, we had to restrict the upper age limit of participating children to 18 months or younger at the time of baseline data collection in order to collect valid endline data 24 months later. To be eligible, the household had to reside in the catchment area during the baseline data collection period (November 2013-January 2014). Initial Sampling: Using the health assistant records, the team created a list of households with at least one child aged between 3 and 18 months during the baseline data collection period. The team used a reference date of October 21, 2013, to calculate the age (in months) of the target children, and the team will collect endline data by October 2015, when the children will still be under 42 months of age. Finally, within each community clinic catchment area, we randomly selected 33 households with children aged between 3 months and 18 months (as of October 21, 2013). The same set of households surveyed during the baseline data collection period will be surveyed during the endline data collection period. Replacement Sampling: Anticipating that some households would be ineligible or would refuse to participate in the study, the team developed rules for replacing ineligible or "out-of scope households" and refusal households, following the guidance of two survey methodologists from AIR. Twenty additional replacement households were randomly selected from within each community clinic and included in a separate list, with each household randomly sorted from 1 to 20. When any of the originally selected 33 households were found to be ineligible or refused to participate, the field interviewer replaced it with the first household from the 20-household replacement list. Field interviewers continued replacing households in order. A careful differentiation was made between ineligible and refusal households. Ineligible or "out-of scope" households: This category includes households that were randomly selected to be part of the sample but did not fit the target sample description of "Households with children from 3-18 months of age that live in the selected community clinics' catchment areas during the period of the baseline data collection." Out-of-scope households included the following cases: a) Households that had permanently left the catchment area. These 300 households had resided in the catchment area during birth record data collection, but by the time of the baseline data collection they had relocated to a different residence outside the catchment area. In these cases, more than one source (such as neighbors or health assistants) confirmed that the household had moved. b) Households with incorrect location information in birth records. In 291 cases, the selected households were not able to be located. This class of out-of-scope households includes two groups. The first group consists of the households who did not permanently reside in the catchment area of the selected community clinic, but had been registered in the health assistant record because they received services while they were visiting relatives or otherwise transiting through the community clinic's catchment area. The second group consists of households whose birth records were fabricated. This was confirmed to be the case in two community clinics, where a large number of households could not be located. (In response to this finding, the field data team met with the relevant HA, as well as representatives from Save the Children). c) Households with children ineligible due to inaccurate date of birth. In 173 households, the birth records had an inaccurate date of birth for the child, and the child was not in the age range of 3-18 months old. d) Households with temporarily absent families. In 159 cases, the households were located but the respondents were not available for interview because they were not in the village and were temporarily staying elsewhere (often visiting relatives). Refusals: This category includes both households that refused to participate in the study and households that began but did not complete data collection. Thirty-nine eligible households (1.5% of the sample) did not agree to fully participate in the study. In 12 cases, the household refused to participate in any capacity. In 27 cases, the households began the household survey but later decided not to complete data collection (i.e., they did not participate in the BSID-III test or the anthropometric measures). For all 39 cases of refusal, the data collectors completed a non-complier questionnaire that captured some basic characteristics of this group to compare with the compliers. Field Sampling: In cases where the field team was unable to complete data collection with a full set of 33 households in a community, even after exhausting the 53 randomly selected households (33 households from the original sample and 20 replacement households), the study employed an additional field replacement process. A total of 454 households from among the 2,574 were sampled using this method. The field replacement process was necessary because a new random selection from the birth record was impractical; either the birth record data were inaccurate or households had relocated. In order to locate replacements, the field team visited a household neighboring the missing household. If there was an eligible child in that household and that child also appeared in the master list that was collected from the health assistant, we selected that household. If this was not the case, we asked to be referred to the nearest households (within the area of the missing house) with infant children, and we repeated the process. These households were then cross-checked with the list of 53 households to avoid duplicative data collection, and the field team visited the nearest household with an infant child that most closely matched (in terms of the age range and the gender of the missing child) the random selection and neighbors' information. If the original neighbor's household contained an eligible child, the interview was performed there. If the field team was unsuccessful in locating the nearest eligible household, the process was repeated by asking neighbors of the next missing household in the sample. As noted, this process began only after the original list of 53 households in a community clinic was exhausted.