Household dietary diversity and Animal Source Food consumption in Ethiopia: evidence from the 2011 Welfare Monitoring Survey

Secondary data were analyzed from the 2011 Ethiopian WMS. The survey covered all rural and urban areas of the country, except three pastoral zones of Afar region and six of the Somali region. Different strata were defined for the urban and rural area. Urban areas were classified into three categories based on their population size: major urban centers, medium and small size towns. In the rural area, agro-ecological zones were used as strata. A two-stage stratified cluster design was used to select sample enumeration areas and households from major urban centers and rural areas. For medium and small towns, a three stage stratified cluster design was used to select sample towns, enumeration areas and households. The 2007 Population and Housing Census data was used as a sampling frame. All major urban centers were directly selected. Sample towns (medium & small size) in each region and sample enumeration areas both in urban and rural areas were selected by probability proportional to size based on the number of households from the 2007 Census. All households from all selected enumeration areas were listed. The listing excluded collective quarters like army barracks, hospitals, police camps and boarding schools. Then, a representative sample of 27,995 households was selected from the enumeration areas for the survey. A structured questionnaire was used to collect socio-demographic and economic data. Dietary data were collected using a dietary diversity questionnaire. The questionnaires, translated into local language, were interviewer-administered and were verified for consistency during pretesting. Interviewers were trained before the actual data collection and supervisors kept track of the field procedures and checked the completed questionnaires everyday to ensure the quality of the data. Detailed participant characteristics and data collection methods have been described elsewhere [26]. The study was approved by the Ethical Review Board of Jimma University, College of Health Sciences.

We used a dietary diversity questionnaire containing commonly consumed food items in Ethiopia to calculate the HDDS and evaluate the household dietary practices. The respondents were asked to report the number of days in the past week they consumed categories of the listed food items. The food groups used to construct the HDDS in this study were constructed from the list of food items the participants reported to consume according to the FAO guideline; cereals, tubers, legumes, meat, egg, vegetables, fruits, oil, sweets, milk and fish, [27]. Consumers of a food item were defined as follows: consumption of the food item at least once in the past week [28]. Irrespective of the frequency, a household which consumed a food group at least once in a week period was scored as 1.The HDDS was constructed as the sum of numbers of food groups consumed over the past week. Higher score indicates higher diversity, as more food groups were eaten.

In the current sample, the minimum and maximum HDDS were one and eleven respectively. Despite the lack of established cut-offs in terms of number of food groups to indicate adequate or inadequate dietary diversity for HDDS, higher HDDS is desirable and larger number of food groups are required to meet requirements of various nutrients [29]. Therefore, the score is converted into tertiles and the higher tertile of the score was taken as high HDDS and the middle and low tertiles were merged together as low HDDS [30, 31]. In the high HDDS category households consumed seven and more food groups while in the low HDDS less than seven food groups in the previous 1 week.

ASF intake was assessed by counting the consumption of animal source products (meat, milk, egg and fish) by the households over the 1 week period. The consumption of each ASF is then described by its distribution among the regions. Composition of the food groups including ASFs; consumed in the higher and lower categories of HDDS was also examined to assess what additional food groups came in to the plate when a household is in the higher HDDS category which also gives insight about the quality of food consumed in the households.

The individual characteristics examined as determinants of dietary diversity were age, sex, literacy status of the household head, and illness of any household member over the past 12 months. It was hypothesized that male sex of the household head and being literate were positively associated with dietary quality; while illness in any member of the household was believed to negatively affect the dietary diversity [30‐33]. The examined household level factors included residence (urban vs. rural), household size (categorized as 4 and less and above 4 based on the average household size), experience of shock (defined as having experienced at least one of the following circumstances in the past 12 months: drought, crop damage, flood, loss of job, robbery, fire accident or land slide), experience of food shortage in the past 12 months, number of crops cultivated (categorized as 3 and less and above 3 crops based on the average number of crops cultivated) and socio-economic score. As there is no income and expenditure data to measure household’s economic status, SES index was constructed. The index was constructed using principal component analysis based on information on housing conditions, ownership of assets and access to basic service. The items used were, ownership of durable goods (TV, Radio, sofa set, refrigerator), crop land, livestock, house, cash crop production, transportation (car, motor bike, cart, horse, donkey, camel), house construction materials, farming materials, source of drinking water, type of the wall, the roof, ceiling of the house, having toilet, kitchen and electricity. The analysis was performed separately for urban and rural households as the characteristics that define wealth in the two scenarios are expected to differ [32, 33]. Three factors were identified both in urban and rural areas and they explained 63 and 59% of the variations in the variables included in urban and rural areas respectively. Factor loadings having a value > 0.5 were the basis for selecting individual variable for the final factors. The variables included in the final factors were source of drinking water, toilet facility, electricity, type of floor, type of wall, ownership of durable goods (4 items recorded as yes and no ), crop land, livestock, ownership of horses, donkey or cart. For analytical purposes, the score is then categorized into tertiles as low, medium and high SES [32]. The variation in the pattern of consumption of all the food groups across socio-economic score was also tested using the chi square test for trend. HDDS was also described in terms of its distribution in the regions to identify regions with high and low HDDS.

The data were coded and entered in to SPSS for windows [SPSS Inc. version 16.1, Chicago, Illinois]. Descriptive statistical analyses were conducted to determine the characteristics of the households and the types of food groups consumed in the reference period. Mean and standard deviation were reported for continuous variables that are normally distributed and frequency distributions for categorical variables. Bivariate analysis was carried out to test differences in HDDS. Multiple logistic regression analysis was used to identify independent predictors of HDDS after controlling for confounding factors. All covariates showing association with HDDS (p < 0.25) were included in the final model. Interaction terms between different predictor variables were evaluated and collinearity was assessed using variance inflation factor. The association between consumption of individual food groups and SES was examined by Chi Square test for trend. Statistical significance was determined by using a P < 0.05 as a cut-off point. The results are presented as proportions and the parameters from the model as β coefficients and OR with 95% confidence interval. All the above analysis took into account the survey design by adjusting for sampling weight, strata and clustered nature of the data by using svy command on STATA version 13.

A total of 27,995 households were involved in the analysis. A little over half of the study households, 52.2% had more than four household members during the time of the survey. Majority of the respondents were found in the age range of above 50 years (29.9%) followed by 31–40 years (25.9%). All of the respondents were heads of household and 75.0% of them were males. About 40.0% of heads of the households were literate and 21.6% of the study households were from urban residence. Only 7.8 and 26.2% reported to have experienced illness and any kind of shock in the past 12 months. Out of the farming households, as many as 36.9% of the households reported to cultivate more than three types of crops. From the total households, 11.2% of them were categorized as having high socio-economic score (Table 1).

The details of the proportion of households who consumed various types of food groups over the previous week are presented in Fig. 1. The mean HHDS was five food groups. Cereals were the most commonly consumed food group by 96.0% of the households, followed by vegetables (81.6%), legumes (76.1%) and oils (75.4%). Fish, egg and fruits on the other hand were the least consumed food groups, in the study households. The mean HDDS is found to be 5 (±1.9).

In all the regions except in the South Nations and Nationalities and People’s Representatives (SNNPR), there was a significant difference between the proportions of households with high and low HDDS. The proportion of households with higher HDDS was highest in Dire Dawa (49.3%), followed by Harari (45.1%) and the capital Addis Ababa (39.0%). Majority of the households in all the regions except Addis Ababa, Diredawa and Afar were also engaged in farming and livestock rearing (Table 2).

The consumption of each food group was further tested against the households socio-economic score (Table 3). The analysis revealed a significant difference in the consumption of all the food groups in the studied households by their socio-economic score.

In addition to reporting the food groups consumed, it is important to know which food groups were predominantly consumed at different levels of HDDS. This provides information on which food groups were eaten by those with the lowest score and which ones were added for those with the higher score. In the current study, there was a notable difference in the consumption of the food groups among the households with respect to their dietary diversity score (Table 4). Cereals (95.1%), vegetables (78.5%), legumes (71.9%) and oil (69.3%) were the food groups consumed in a greater proportion among households with low HDDS. The most frequently reported food groups by the households with higher HDDS in addition to the previous ones were sweets (84.3%), meat (70.8%), tubers (65.2%), milk (62.3%), fruits (54.4%) and egg (50.5%).

Examining the pattern of consuming ASF showed the variability in consumption across the different regions (Fig. 2). The Somali region had the highest intake of milk where 80.7% reported milk consumption in the previous week. The lowest milk intake was found in Benishangul, with only 14.8% reporting milk consumption in the reference period. Meat was consumed in 54.3% of households in Tigray and 47.7% of households in Addis Ababa. Egg was consumed by 38.8% of the households in the capital Addis Ababa, which was a higher proportion compared to other regions. Gambella, Dire Dawa, Addis Ababa and Harari were the regions where relatively all ASF, except fish, were consumed in a better proportion compared to other regions. Fish is consumed in significant proportion (44.8%) only in households in Gambella in the reference period.

This study demonstrated that ASFs, specifically meat, egg and milk in addition to other food groups were consumed in a larger proportion (>50%) among households with higher HDDS and the variation was found to be statistically significant (Table 4).

Predictors of higher HDDS were identified in this research using logistic regression analysis. Based on that, socio-economic score of the higher (AOR = 5.37, P < 0.01) and medium category (AOR = 1.93, P < 0.01), urban residence (AOR = 1.83, P < 0.01), literacy (AOR = 1.79, P <0.01), having more than 4 household members (AOR = 1.18, P < 0.01) and owning livestock (AOR = 1.58, P < 0.01) were found to be positively associated with higher HDDS. On the other hand, the odds of having higher HDDS among female headed households was lower than that of male headed households by 11% (AOR = 0.89, P < 0.05). The likelihood of having higher HDDS among those households who experienced food shortage in the previous 12 months was 62% lower when compared to those households with no shock experience (AOR = 0.38, P < 0.01) (Table 5).

Similar to many LMICs [34], cereals were consumed among many of the households whereas; fruits and animal source products were less consumed. The study revealed that most of the households had low HDDS, though there were notable variations across regions. It was in the two city administrations (Dire Dawa and Addis Ababa) and Harari region that a sizable proportion of households had high HDDS and also consumed ASFs in better proportion compared to other regions. Households in these regions were urban residents with higher access to markets that can supply more food varieties which may have contributed to the higher score. The growing demand by the middle class of the population and other external economic changes are causing domestic prices of food items to rise [35]. As a result, farmers are selling majority of their products to the urban consumers and many of the products are not accessible at an affordable price to the poor majority in rural areas. Moreover, urban households are believed to be exposed to information about the health benefits of a diversified diet through different media outlets [30]. Afar, Amhara and Somali were the regions where more than 80% of the households were having a low HDDS. Majority of the households in these regions are pastoralists and traditional farmers who are dependent on marginal resources whose availability are dependent on seasonality. They face challenges like recurrent droughts, flood, erratic rain fall patterns and high temperature [36, 37] that threaten their livelihood practices. Their frequent mobility as a result of the aforementioned disasters and a very traditional way of farming may have resulted in lesser productivity and then consumption of a less diversified diet.

The current study demonstrated that the proportion of ASF consumption is higher among households with higher HDDS. The multiple logistic regressions also revealed a positive and significant association between owning livestock and higher HDDS. Therefore, the increased ASF consumption could be from a direct consumption of what they own or through increased purchasing ability of the households as they generate income from owning livestock. For majority of households, livestock is an important asset of regular income that the households rely on and be disposed of in hard times to provide a safety net [38]. As reported previously [30] and demonstrated in the current study also, HDDS is positively associated with a higher socio-economic status of the households where by the households are able to purchase variety of foods including ASF to diversify their diet.

Though there is a significant difference in the consumption of all food groups in both categories, our study indicated that ASFs were introduced to the plates of households in a greater proportion in the higher HDDS category compared to the lower HDDS category. A diversified diet containing ASF is good source of essential micronutrients [15, 39]. Although ASF consumption contributes majorly to dietary quality, decreased access and relatively higher price hampers its consumption in most low income settings [40, 41].

Fruits are among the main food groups which contribute a significant share for under-consumed nutrients in many settings. They reduce the risk of many chronic diseases, help individuals achieve and maintain a healthy weight [42, 43]. Their intake is also commonly taken as an indicator of a healthy overall diet [44]. This study identified that, fruits were also among the food groups which were consumed in larger proportion in the higher HDDS category and the difference in the consumption was also found to be statistically significant. Evidence suggests that consumption of fruits is highly associated with household socio-economic status and urban residence in LMICs [45, 46]. Availability in the markets, availability of storage facilities, change in life style and cultural patterns could also result in a higher consumption of fruits among the households [47].

Consumption of sweet foods and beverages has long been in scrutiny for its risk related to obesity and chronic diseases worldwide [48]. Despite this fact, its consumption in the current study was found to be higher among households with high HDDS. The need for convenience leading to the purchase of more processed foods, availability in the urban markets and recent shifts in the pattern of dietary consumption due to increased income and urbanization [49, 50] could have resulted in its increased consumption.

The consumption of tubers and oils were also much higher in the higher HDDS households in the current sample. If one has to mention largely consumed food groups in Ethiopia, edible oil or cooking oil would be one on top. Though edible oil is one of the main consumer goods with increased price and exist in short supply, the urban consumer, where majority of households with higher HDDS live, are dependent on it for most domestic food preparations. Recent report on food production and consumption patterns in sub Saharan Africa [51] indicated that, there has been a positive growth rate in the production and consumption of cereal, oil seeds and tubers. The report also informed that there is a shift in the production of cash crops to food crops due to increase in the incident of drought conditions. This could apparently result in the increased consumption of these food groups.

The current study identified that households from urban areas and high socio-economic groups have a more varied diet. This finding is comparable with a previous study in Ethiopia and central Mozambique [30, 52], which indicated that urban residence and high household income tertiles were positively associated with having high HDDS. This suggests that in the face of increasing food prices, higher income of the household enhances the capacity to purchase various food items thereby they will have access to consume a diversified and quality diet [53‐55].

Experiencing food shortage in the past 12 months was negatively associated with higher HDDS. Food constraints, which can be expressed through having poor dietary practices, can occur when households experience different types of shocks. The poor dietary practices emanate from behavioral adaptations in response to food shortages in the household [56, 57] which ultimately result in a trend of consuming less diversified diet in the households. Our findings have implications for the current drought situation in Ethiopia where the livelihood of more than 10 million people is already in crisis [58]. In the face of scarce food supplies, households opt to coping strategies expressed by reduced number or size of meal which will negatively affect their nutrient intake [56, 59‐61].

It is also noted in this study that male-headed households had a higher HDDS. As is common in many LMICs, men are mostly household’s decision makers and the sole source of income [62, 63] which could directly affect the variety of diet consumed in the household. . But it is also important to note that there are studies [64] that identified that female-headed households spent more on higher-quality, more expensive and protein-rich foods. This implies the need to empower [65] women through involving them in household decision making, having their own income or managing the family budget could have also a positive impact in consuming quality diet in addition to its diversity. Literacy of household heads was also positively associated with an increased HDDS. Comparable findings from other studies noted that literate household heads spent a significant amount of their food budget on nutritionally important food groups, because of a better understanding on their health benefits [66‐69].

Households with more than four family members were found to be positively and significantly associated with higher HDDS. In these households, 45.3% of the household members were between 15 and 50 years of age. A finding from the 1996 Tanzanian health and demographic survey corroborated that, in a labor intensive socio-economy, households with a larger labor force could be less poor[70]. Families with a larger labor force can also invest more in agricultural practices resulting in an increase production, or increase their income by paid labor.

This analysis was based on very large data set which represents all regions of the country, and provides a good source of information for benchmarking strategies focusing on nutrition surveillance in the country. As there is within household variability of dietary intake within the days of the week, the study used a reference period of 1 week to define consumer of a food group. The dataset used does not provide information on the quantity of food consumed and the adequacy of nutrients (compared to requirements). Thus, we were unable to investigate if HDDS is associated with adequate nutrient intake. Given the fact that most people share food from a common bowl, it is difficult to quantify individual consumption and attribute the nutrient intakes [6, 71]. Nevertheless, dietary diversity score has been validated as a useful proxy indicator to assess the likelihood of meeting micronutrient requirements [71‐73].

In Ethiopia many programs such as Social Safety Net, Agriculture Development Program (AGP) and Livestock Market Development [74] aim to increase household income, agricultural productivity, improve quality and diversity of household diet and as such nutritional status. These programs are believed to help realize the Sustainable Development Goals in nutrition [75]. For these programs to be effective, identifying target populations and vulnerable areas is a key. Our analysis shows that HDDS can be used to either target or evaluate the effect of the strategies employed by the above mentioned programs.