3.1. Data Sources To examine both qualitative and quantitative changes in rice consumption patterns at the household level, this study relied on the HIES data collected in 2000, 2005, and 2010 (BBS, 2000, 2005, 2010a), which were made available by BBS, the government of Bangladesh. BBS used a two-stage stratified random sampling to ensure greater precision. In the first stage, more than 500 primary sampling units (PSUs) were selected across the country; in the second stage, 10 to 20 households were selected randomly per PSU to represent rural, urban, and statistical metropolitan areas. In the 2000 HIES survey, 7,440 households were randomly selected from 6 divisions, 64 districts, and 295 subdistricts, out of which 6,316 households were located in rural areas and the rest (1,124) were from urban areas (BBS, 2000). In the 2005 HIES, a total of 10,080 households were randomly selected from 6 divisions, 64 districts, and 351 subdistricts, out of which 6,400 were located in rural areas and the rest (3,680) were located in urban areas (BBS, 2005). Finally, in the 2010 HIES (BBS, 2010a), a total of 12,240 households were randomly selected from 7 divisions, 612 PSUs, 64 districts, and 381 subdistricts, out of which 7,840 were from rural areas and the rest (4,400) were from urban areas. Thus, the present study is based on information collected from 29,760 households, of which 20,556 were from rural areas and the rest (9,204) were from urban areas.
3.2. Sampled Household Characteristics and Rice Consumption Patterns presents the demographic characteristics of the sampled households at different expenditure levels. To capture the heterogeneity in rice consumption among the sampled households, we divided all sampled households into four expenditure quartiles. The first expenditure quartile consists of the poorest households with the lowest per capita real expenditure on food per day; by contrast, the fourth expenditure quartile consists of the richest households with the highest per capita real expenditure on food per day. On average over the years sampled, per capita, daily real total food expenditure was nearly Bangladesh taka (BDT) 4 14, but it was only BDT 7.58 in the case of the poorest sampled households belonging to the first expenditure quartile, 3.2. Sampled Household Characteristics and Rice Consumption Patterns Table 1 presents the demographic characteristics of the sampled households at different expenditure levels. To capture the heterogeneity in rice consumption among the sampled households, we divided all sampled households into four expenditure quartiles. The first expenditure quartile consists of the poorest households with the lowest per capita real expenditure on food per day; by contrast, the fourth expenditure quartile consists of the richest households with the highest per capita real expenditure on food per day. On average over the years sampled, per capita daily real total food expenditure was nearly Bangladesh taka (BDT) 4 14, but it was only BDT 7.58 in the case of the poorest sampled households belonging to the first expenditure quartile, Interestingly, Table 1 also shows that more than 50% of the sampled households belong to the third and fourth expenditure quartiles, located in Dhaka and Chittagong divisions. Bangladesh has seven administrative divisions, among which the capital city Dhaka is located in the Dhaka division and the largest seaport is located in the Chittagong division. These two divisions are relatively more industrialized than other divisions. The incidence of income poverty (headcount rate) also tends to be lower in Chittagong and Dhaka divisions than in other divisions. The findings in Table 1 thus confirm that the richest sampled households that belong to the fourth expenditure quartile are more likely to be highly educated and urban households, a majority of which are located in Dhaka and Chittagong divisions, two of the most industrialized divisions in Bangladesh.
4. Conceptual Framework To identify the drivers of change in rice consumption and grain-type preference at the household level in Bangladesh, we apply a fixed-effect regression approach. We separately estimate the rice consumption functions for households belonging to separate expenditure quartiles to encounter the problem of heterogeneity in rice consumption among different income groups. Particularly, we specify the following model-
Yitj = Xitjβ + Vj + εitj ,
where Yitj is a vector of dependent variables that includes per capita daily consumption of total rice, fine-grain rice, and ordinary-grain rice in grams by household i in year t (= 2000, 2005, 2010) located in a geographic unit j, which is a mauza in this case5 belonging to one of the four expenditure quartiles; and is a K × 1 vector of explanatory variables that includes the following: daily real per capita total expenditure on all food items, daily real per capita total expenditure on rice, the price of fine-grain rice (in BDT/kg), the price of ordinary-grain rice (in BDT/kg), age of the household head, years of schooling of the household head and spouse, number of family members, a dummy for sex that assumes a value of 1 if a household head is male and 0 otherwise, a dummy for the household’s location that assumes a value of 1 if a household is located in a rural area or 0 otherwise, a dummy for the household’s location in Dhaka and Chittagong divisions that assumes a value of 1 if a household is located in Dhaka and/or Chittagong and 0 otherwise, year dummies for 2005 and 2010 (for which the base year is 2000), and multiplicative dummies for years 2005 and 2010 with urban households.
It is important to mention that because of the cross-sectional nature of the data, we could not apply the household-level fixed-effect estimation method to estimate the equation. Instead, we applied a mauza-level fixed-effect estimation method to estimate the equation. In our equation, the mauza-level fixed effect is represented by Vj, which is time-invariant in nature; represents the disturbance term with the white-noise property; and represents the unknown parameters to be estimated.