**Background:** Nutritional deficiency remains a severe problem in developing countries, and household food adequacy does not guarantee individual nutritional adequacy. Intrahousehold food allocation inequality can exacerbate nutritional deficiencies in certain subgroups. While prior literature has examined sex-based discrimination in food intake, few studies have analyzed dietary diversity inequality at the individual level across all household members simultaneously. This study applies parental investment (PI) theory to hypothesize that dietary diversity inequality exists by family roles (fathers, mothers, sons, daughters, grandparents) and age groups (children, adults, elderly).
**Methods:** A cross-sectional survey was conducted in three districts of Bangladesh (Dhaka, Jashore, Satkhira) from February to June 2019. A total of 811 households (219 urban, 592 rural) comprising 3248 subjects participated. In urban areas, occupation-based randomization was used; in rural areas, household lists from local NGOs were used with random number generator selection. Dietary data were collected using a 24-hour recall method. Dietary diversity scores (DDS) were calculated based on 9 food groups per FAO guidelines: starchy staples; dark green leafy vegetables; other vitamin A rich fruits and vegetables; other fruits and vegetables; organ meat; meat and fish; eggs; legumes, nuts and seeds; and milk and milk products. A 15g minimum quantity threshold was applied. Poverty was estimated using the cost of basic need (CBN) method, with poverty lines of 2929 BDT (36.08 USD) in urban and 2019 BDT (24.87 USD) in rural areas. Statistical analyses included chi-squared tests, Mann-Whitney tests, Wilcoxon matched-pairs signed-rank tests, ordinary Poisson regression, and two-level random intercept Poisson regression to account for clustering of subjects within households.
**Key Results:** The mean DDS for the overall sample was 4.88 (out of 9). Urban subjects had significantly higher DDS (5.61) than rural subjects (4.63). Poor households comprised 17% of the overall sample (5% urban, 22% rural). The Wilcoxon matched-pairs signed-rank test showed significant distributional differences in DDS between fathers and mothers (Z=3.36, p=0.01), fathers and sons (Z=2.07, p<0.04), fathers and daughters (Z=1.64, p<0.10), and fathers and grandparents (Z=4.96, p<0.01). The Mann-Whitney test found a significant difference between children and adults (Z=−4.61, p=0.01) but not between elderly and adults (Z=1.59, p=0.11). In the Poisson regression models (Model 1-3 and 2-3), the expected DDS of poor households was 15% lower than nonpoor; rural subjects had 12% lower DDS than urban subjects. Father education increased expected DDS by 0.7% per year of schooling; mother education increased it by 1% per year. Grandfathers' expected DDS was 19% lower than fathers (ordinary Poisson) and 16% lower (random intercept model); grandmothers' was 14% lower and 13% lower, respectively. Children's expected DDS was 8% lower than adults (ordinary) and 6% lower (random intercept). The cluster-specific random effect was significant, with a standard deviation of 0.16 in Model 2-3, indicating household-level characteristics have a greater effect on DDS than household poverty (γ3=−0.14).
**Clinical Implications:** This study identifies grandparents and children as the most vulnerable subgroups for inadequate dietary diversity within Bangladeshi households, irrespective of poverty or urban/rural residence. The findings suggest that policies should target these groups specifically. Father and mother education improve overall household dietary diversity but do not resolve intrahousehold inequality, indicating that education alone is insufficient. The authors recommend awareness and education programs targeting fathers and mothers to address intrahousehold food allocation inequality. The study contributes to understanding how to achieve SDG targets related to nutrition and health at the household level. Limitations include the use of single 24-hour dietary recall (potential recall bias), lack of data on nutritional knowledge, food preferences, and health-related variables, and the cross-sectional design preventing causal inference.