Sampling Data were collected from 188,835 households that participated in the Nutritional Surveillance Project (NSP) of Bangladesh in 2003–2005. The NSP was established in 1989 by Helen Keller International and the Institute of Public Health Nutrition of the Government of Bangladesh. A stratified multistage cluster design was used and indicators were chosen based on the United Nations Children’s Fund’s conceptual framework of the causes of malnutrition. Rural households were selected from 4 sub districts in each of the 6 divisions of Bangladesh: Barisal, Chittagong, Dhaka, Khulna, Rajshahi, Sylhet, and the Chittagong Hill Tracts. Sampling was designed to statistically represent each of the divisions as well as the entire country of Bangladesh. Data collection Two-person field teams, trained by Helen Keller International, collected data every 2 mo coinciding with the different seasons of Bangladesh, resulting in a total of 18 rounds of data collection over the 3 y covered in this paper. A new cross-sectional sample was drawn each round. Quality control was ensured by refresher trainings before each round of data collection and by random supervisor visits with repeated collection of a subsample of the data within 24 h of the initial data collection. A structured coded questionnaire was used to collect data. The household head or another adult household member provided information on the household’s composition, weekly expenditures, land ownership, parental education, and other socioeconomic and health indicators. Each household was asked to report expenditures from the previous 7 d on rice, wheat, eggs, pulses, fish, vegetables, fruits, meat, poultry, milk, sweet biscuits, snacks, spices, sugar, cooking oil, and other foods. Data were collected on the following nonfood expenditures for the previous month: medical care, education, housing, agricultural inputs and livestock purchases, electricity, fuel, taxes, loan payments, and other household items. Expenditure variables were collected in Bangladeshi taka. A 7-d recall was used to collect information on the number of days each household consumed typical foods commonly available in rural Bangladesh, including yellow/orange fruits and vegetables, green leafy vegetables (shak), meat, chicken, fish, eggs, or pulses (dal). Data analysis A simple household dietary diversity score was created from the non grain food groups described above. The score was calculated as the unweighted sum of the number of days in the previous week households consumed at least one item from each of the 7 food groups described above. The range of possible scores was 0–49. In keeping with our interest in the validity of simple measures, the proxies used in this analysis for food security and household wealth are straightforward. In the absence of data on energy availability at the household level, we use household expenditures as our comparison measures for food security. Household level food consumption and food expenditure data were collected in the NSP using a 7-d recall. Though we consider dietary diversity and food security at the household level, we created per capita expenditure variables for use in the analysis to control for the effects of household size on spending. Monthly per capita non grain food expenditure represents the sum of reported 7-d expenditure on the food categories given in the expanded list above, except rice and wheat, multiplied by the mean number of weeks per month (4.34 wk/mo). Because few of the poorest households purchase large amounts of rice, a monetary value for in-kind rice was included in total food expenditure calculations (21). Rice produced on a household’s land, received as in-kind payment for labor, and received as a gift was reported in kilograms and assigned a monetary value determined by the daily market price of rice recorded at the time of the survey. This monetary value was summed with cash expenditure on grain and non grain foods and divided by household size to calculate monthly per capita total food expenditure. Extreme outliers of total food expenditure were identified and removed during data cleaning. Nonfood expenditures, collected using a 1-mo recall period, were summed and divided by household size. The result was added to per capita total food expenditure to arrive at an estimate of monthly per capita total expenditure, a proxy for household income. The amount of cultivable land owned by the household, another important socioeconomic indicator in Bangladesh, was reported in decimals. For analysis, we first calculated the hectare (ha) equivalent of the decimals owned and then created a categorical land ownership variable using cutoff values consistent with previous research in Bangladesh (2,22). Households owning 0 ha were categorized as landless. Land ownership fell into the following categories: marginal, which included any land up to 0.2 ha; small, from 0.201 to 1.0 ha; medium, from 1.01 to 2.0 ha; and large, which included any land .2 ha. The area of the main dwelling, in square meters, was calculated by multiplying the reported length and width. For both maternal and paternal education, values .12 were recoded as 12 y of education because of the small sample size of individuals with more than a high school level education. Statistical analyses were performed using SAS Survey (SAS Institute). To represent the rural population of Bangladesh, analyses were weighted according to the population of each of the 7 divisions of Bangladesh and adjusted for the multi-stage cluster design of the NSP. The unit of analysis was the household and those with multiple children were counted only once. Spearman correlation tests were used to examine associations between dietary diversity score and expenditure variables, parental education variables, amount of cultivable land owned by household, the area of the main dwelling, sex of household head, and number of members in household. Differences in mean dietary diversity score between quintiles of total expenditure and food expenditure were examined using 1-way ANOVA with multiple comparison tests undertaken using the Tukey-Kramer adjustment (P , 0.05). ANOVA models were also used to compare mean weekly household food expenditures, non grain food expenditures, and expenditures on each individual food group by quintile of dietary diversity score. Multiple linear regression models were used to test associations between each expenditure variable and the score, controlling for land ownership, parent education, number of household members, and area of main dwelling. Values in the text are means 6 SD.