The REACH Khulna Household Survey was conducted between December 2017 and February 2018 to collect quantitative data on various indicators of multidimensional poverty, the state of drinking water services and vulnerability to water-related hazards. The survey was carried out in Polder 29, one of the 139 polders (embanked islands) in southwestern coastal Bangladesh. Polder 29 has a land area of 80 km2 and covers five unions (Tier-4 administrative boundary) across Dumuria and Batiaghata upazilas (sub-districts) of Khulna district. The polder is further sub-divided into 77 mouzas (Tier-5 administrative boundary), with a total population of 58,000 and 17,000 households (BBS, 2011). The survey was administered to 2103 households in two stages: phase 1 involved 978 households spatially distributed across the polder, while phase 2 involved 1125 households within 16 chosen mouzas with relatively higher risk profiles. A stratified random sampling method was used to allocate equal proportions of households from each mouza. During the survey, enumerators used a ‘random route/walk sampling’ technique, where households were selected at certain intervals, determined by the total number of households in the mouza, the sample size per mouza, and the settlement patterns. The survey was administered in Bangla through an electronic form developed in ONA (https://ona.io/), and was conducted by 15 trained local enumerators.REACH (www.reachwater.org.uk) is a nine-year global research programme (2015 - 24) to improve water security for 10 million poor people in Asia and Africa. REACH is led by the University of Oxford with an international consortium of partners and funded by the Foreign Commonwealth and Development Office (FCDO). REACH research focuses on eight observatories across three countries (Bangladesh, Ethiopia and Kenya).

Data from the 2011 National census was used to categorise the 77 mouzas in Polder 29 into low, moderate and high risk profiles, based on selected water security and poverty indicators including drinking water stress, sanitation facilities, waterlogging, erosion, soil salinity, cropping intensity, and asset index. Of the 77 mouzas, some were excluded due to lack of settlements. The survey was administered to 2103 households in two phases. Phase 1 involved 10% of the total number of households in the high risk mouzas and 5% in the low/ moderate risk mouzas, resulting in a total of 978 households across 66 mouzas. Phase 2 incorporated an additional 20% for the 16 high risk mouzas. The household heads or their spouses were the target respondents; however, since the survey contained detailed questions on agricultural activities, the presence of male respondents proved to be necessary at times. Ethical approval for the survey was granted by the University of Oxford's Central University Research Ethics Committee (CUREC). Researchers and enumerators ensured that all respondents participated voluntarily through informed consent, and data with personal identifiers was stored in secured platforms. The survey was administered in an electronic form developed in ONA (https://ona.io/), a Mobile Survey Platform that provided a hosted server for uploading, editing, viewing and submitting forms. The questionnaire was first programmed in English and then translated to Bangla. A total of 15 male and female enumerators were recruited and trained over a five-day period. The training involved two pilot exercises carried out in two separate villages with differential risk profiles. Each enumerator was asked to survey one household in each village, so that they understood the various issues they could face during the survey. The enumerators were supervised by two field officers who reported to the researchers on a regular basis. As we did not have a sampling frame (that is, the list of all households within the mouzas), it was difficult to follow strict probabilistic sampling procedures in selecting the households. Instead a ‘random route/walk sampling’ technique was used, in which the enumerator was assigned a starting point in each mouza and asked to select households at certain intervals, which was determined by the total number of households in the mouza, the sample size per mouza, and the settlement patterns. Enumerators were instructed to select households of different wealth categories, often identified by the housing structure. They were also asked not to concentrate only on easily accessible households, such as those beside the roads, but to cover all areas within the mouza. To ensure quality control, the data was accessed remotely from the server and checked daily by the researchers, who communicated with the field supervisors to settle any queries. In addition, the enumerators were asked to note down their comments separately at the end of each survey. At the end of each day, the enumerators shared the issues (if any) they faced with the field supervisors, who in turn kept a log of the discussion.