Title: Determinants of Minimum Acceptable Diet among 6–23 Months Age Children in Ethiopia: A Multilevel Analysis of The Ethiopian Demographic Health Survey

Background Though infant and young children should be fed according to a minimum acceptable diet to ensure appropriate growth and development, only 7% of Ethiopian 6-23 months age children meet the minimum acceptable dietary standards, which is lower than the national target of 11% set for 2016. Therefore, this study aims to assess the individual and community level determinants of minimum acceptable diet among 6–23 months age children in Ethiopia. Methods This study analyzed retrospectively a cross-sectional data on a weighted sample of 2919 children aged 6-23 months nested within 617 clusters after extracting from Ethiopian Demographic and Health Survey 2016 via the link www.measuredhs.com. By employing bi-variate multilevel logistic regression model, variables which were significant at the p-value < 25 were included in multivariable multilevel logistic regression analysis. Finally, variables with p-value < 0.05 were considered as significant predictors of minimum acceptable diet. Results Only 6.1% of 6-23 months age children feed minimum acceptable diet in Ethiopia. Children 18-23 months age (AOR=3.7, 95%CI 1.9, 7.2), father’s with secondary or higher education (AOR=2.1, 95%CI 1.2, 3.6), Employed mothers (AOR=1.7, 95%CI 1.2, 2.5), mothers have access to drinking water (AOR=1.9, 95%CI 1.2, 2.9), mothers with media exposure (AOR=2.1 95%CI 1.1, 2.7) were positive individual level predictors. Urban mothers (AOR=4.8, 95%CI 1.7, 13.2)) and agrarian dominant region (AOR=5.6, 95%CI 2.2, 14.5) were community level factors that significantly associated with minimum acceptable diet of 6–23 months age children. Conclusion Both individual and community level factors were significantly associated with minimum acceptable diet of 6-23 months age children in Ethiopia, suggesting that nutritional interventions designed to improve child health should not only be implemented at the individual level but tailored to community context as well.

individual's characteristics, but also the community level attributes (Fig 1). The  Statistical analysis 143 The data were analyzed using STATA version 12. A sampling weight was used for computing all 144 descriptive statistics to adjust for the non-proportional allocation of the sample to different 145 regions and their urban and rural areas and for the possible differences in response rates. Hence, 146 the actual representativeness of the survey results at both the national and regional levels is 147 ensured. The 'Svy' command was used to allow for adjustments for the cluster sampling design.

148
Tables were used for data presentation. Frequency and percentage were used to report categorical 149 variables.
To examine the effect of multilevel factors on the individual dietary behavior, multilevel 151 modeling approach was used. The nested nature of EDHS data makes the use of traditional 152 regression methods inappropriate because of the assumption of independence among 153 individuals within the same group, assumption of equal variance across groups which are 154 inherent in traditional regression methods are violated. Therefore, in this study, a two-level 155 mixed effect logistic regression analysis was employed in order to estimate both independent 156 (fixed) effects of the explanatory variables and community-level random effects on feeding 157 MAD. The first level represents the individual (children) and the second level is the cluster 158 (community). Hence, the log of the probability of feeding MAD was modeled using two-level 159 multilevel model as follows: Where, Is the feeding of MAD for the individual I in j cluster/community; X ij and Z ij are 162 individual and community level variables for the i th individual in group j, respectively. The β ' s 163 are fixed coefficients indicating a unit increase in X can cause a unit increase in probability β 164 feeding MAD. While, the is intercept that is the effect on feeding MAD when he effect of all β o 165 explanatory variables absent (38). The u j shows random effect (effect of the community on 166 mother's decision to provide MAD) for the j th community. By assuming each community has 167 different intercept (β 0 ) and fixed coefficient (β), the clustered data nature and the within and 168 between community variations were taken in to account. The measures of community variation 169 (random-effects) was estimated as intra class correlation coefficient (ICC) which is proportion of 170 the total variance in feeding MAD due to variables operating at a community level. Hence, the 171 ICC was estimated as: ρ = (σ 2 u / (σ 2 u + π 2 /3)), Where, ρ is the ICC, σ 2 u is the variance at the that 44.9% of the total explained variance in the odds of feeding MAD to a child could be

262
The individual and community level determinants of feeding MAD are presented in Table 5. At 263 individual level: child age, father's education, mother's occupation, time to get drinking water, 264 media exposure; and at community level: region of residence were independent predictors of 265 feeding the child MAD.

266
The odds of feed the child MAD was nearly 4 times (AOR=3.7; 95%CI 1.9, 7.3) higher among 267 children in age between 18-23 months than children at age 6-8 months.  This study used multilevel logistic regression analysis to address these two level factors.

286
Accordingly the individual level factors such as the child's age, mother's occupation, father's 287 education, access to drinking water and media exposure showed a significant association with 288 feeding minimum acceptable diet.

289
The age of the child was positively associated with feeding minimum acceptable diet for 6-23 290 months age children. Children with age 18-23 months were found to be significantly higher odds infants may develop diarrhea due to poor hygienic condition, but mothers could associate this 299 problem with taking different food items and eventually she might not permit the child to taste 300 unfamiliar foods. It can also be attributed to the feeding interest of the child too.

301
Father's education level was significantly and positively associated with feeding the child 302 minimum acceptable diet. Other studies in South Asia, Bangladesh and Nepal had reported a 303 similar finding indicating that children whose fathers had secondary or higher level of education 304 were more likely to be provided with the recommended acceptable diet as compared to the 305 children whose fathers did not have any education [22,23,24]. This could be due to educated 306 fathers were more likely to have information (media exposure), understand educational messages 307 about child feeding easily, might have received lessons on child feeding in the curricula at school 308 that would increase their knowledge about the importance of child feeding. However, different 309 finding was reported from study done in Serilanka, in which paternal education was not 310 associated with any of complementary feeding indicators [25]. This difference could be due to 311 study area and sample size. could also be linked with less food production and higher levels of poverty.
344 Furthermore, the current study found that; even though most variation on feeding the child The proportion of young children aged between 6-23 months receiving minimum acceptable diet 354 was very low (6.1%). This study also showed that this low proportion of minimum acceptable 355 diet among 6-23 months age children is determined by number of individual and community 356 level factors such as age of the child, mother's occupation, father's education level, and media 357 exposure, access of household drinking water and region of residence. Therefore, interventions 358 to improve MAD practice should not only be implemented at the individual level but also be 359 tailored to the community context. Utilizing media to promote feeding young children with 360 minimum acceptable diet through enhancing mother's knowledge on child feeding practices and 361 advocacy for appropriate complementary feeding, particularly on meeting MAD and nutrition 362 education and social and behavior change interventions should be strengthened targeting the 363 infants and children aged 6-8 months and unemployed mothers.