The detailed methodology of the BDHS 2004 has been explained elsewhere. Briefly, this survey is a nationwide, well-designed survey that collected a lot of information using four different questionnaires, including a household questionnaire. The draft household questionnaire, which was developed after a series of meetings among experts, was finally reviewed and approved by the BDHS technical Review Committee. This survey was implemented by Mitra and Associates, a well-known Bangladeshi research firm located in Dhaka, under the authority of the National Institute for Population Research and Training (NIPORT) of the Ministry of Health and Family Welfare. ORC Macro of Calverton, Maryland, USA, provided technical assistance to the project as part of its international Demographic and Health Survey programme, and financial assistance was provided by the U.S. Agency for International Development (USAID/Bangladesh).
Sample design The 2004 BDHS used a stratified and multistage cluster sample, which included 361 primary sampling units (PSUs) (122 from the urban area and 239 from the rural area) from the whole country (which consists of six divisions and 64 districts). The Bangladesh population census of 2001 created enumeration areas, based on a convenient number of dwelling units, for collecting data. Because these sketch maps of enumeration areas were accessible, the 2004 BDHS considered enumeration areas as the PSUs. In each division, the list of enumeration areas constituted the sampling frame for the 2004 BDHS. For collecting data, Mitra and Associates first conducted a household listing operation in all the selected PSUs from 3 October 2003 to 15 December 2003. Then a systematic sample of 10 811 households (an average of 30 households per PSU) was selected from the selected PSUs. Among the 10 811 selected households, 10 523 households were occupied during the survey time, of which 10 500 households (99.8%) were interviewed successfully. The household questionnaire was used to list all the usual members and visitors in the selected households and the interviewer assigned a unique number (called a line number) to each listed member for identification purposes. From each household, only one respondent completed the household questionnaire. Using the line number of the respondent, we recorded some of their personal characteristics: age, sex, education, marital status and working status, including background characteristics, namely place of residence and division. Information about the dwelling itself was also collected. The variables collected included: the materials used to construct the roof, wall and floor of the house; types of toilets used in the household; sources of dishwashing and drinking water; duration of using the drinking water source; level of arsenic in the water tested by Hach’s EZ Arsenic kit (hereafter Hach kit), marking of tubewell (by green or red) to indicate the safeness from arsenic contamination; number of sleeping rooms in the household; ownership of various consumer goods and amenities such as electricity, radio, television, bicycle, telephone; sufficiency of food in the household for consumption in the whole year. As our dependent variable was related to arsenic concentration in drinking water, from 10 500 household respondents we excluded 35 for which information about arsenic concentration in drinking water was missing. Again, as it is mainly the tubewell water in Bangladesh that is contaminated by arsenic, we further excluded 1349 household respondents who were not drinking tubewell water. Thus we had a total of 9116 household respondents for final analysis.
Testing kit for determining arsenic level in the water Trained interviewers tested the household drinking water using the Hach kit, which is widely used in Bangladesh (Jalil and Ahmed 2003; NIPORT et al. 2005). This kit has a detection limit of 0.0–0.50 mg/l (colour scale for 0.0, 0.01, 0.03, 0.05, 0.07, 0.30 and 0.50 mg/l), which is equivalent to 0.0–500.0 ppb in 50 millilitres (ml) or 0.0–500.0 mg/l of water. Respondents to the household questionnaire were asked to provide a glass of water that the household uses for drinking. If tubewell was mentioned by the respondent as a source of drinking water, this was probed then by matching the answers of two related questions. Interviewers poured 50 millilitres of given water into a special testing vessel, added two reagents in the prescribed order, and quickly closed the vessel with a cap to which a testing strip was attached. Twenty minutes later, the testing strip was removed and matched with a colour chart to determine the level of arsenic in the water.
We used a dichotomous dependent variable (DACW): whether the tested water contained arsenic at the level >50 mg/l (i.e. was arsenic-contaminated) or not. At first, we performed simple (frequency) tests for each covariate variable to show the distribution of respondents, and then cross-tabulations to show the percentage of DACW including P values, based on 2 tests, to indicate the association of each variable with DACW. Later we analysed the covariate variables by binary logistic regression to examine their independent effects on DACW by using odds ratios (OR) and corresponding 95% confidence intervals (CI). SPSS version 10.0 was used for analysing the data.