Study setting, design, and sampling procedure

The Federal Democratic Republic of Ethiopia is located in the Horn of Africa, sharing borders with Eritrea and Djibouti to the north, Somaliland to the northeast, Somalia to the east, Kenya to the south, South Sudan to the west, and Sudan to the northwest. The country is divided into nine regional states and two city administrations. Data were collected in five regions and two city administrations: Amhara, Oromia, Somalia, Southern People and Peoples Nationalities, Sidama, Dri Dawa City Administration, and Addis Ababa City Administration. These areas account for 90.4% of the national population [26].

A population-based cross-sectional survey was conducted from August to September 2022. A multistage stratified cluster sampling method was used to choose households for the study. In the first stage, districts were selected from each study region based on a sampling frame developed by the Central Statistical Agency (CSA) for the 2021 Household Income Consumption Expenditure Survey (HICES). Ninety-nine enumeration areas (EA) were chosen from selected districts using the lottery method. A household list was obtained for each of the selected EAs, which were used as a sampling frame for the final stage of household selection. A household had to contain at least one member of the study target group. From the revised listing, twenty eligible households pre-EA were selected randomly.

Sample and sample size determination

The study population was women of reproductive age (15–49 years). Pregnant and lactating women were excluded since their dietary behavior and requirements likely differed from the general population. The subjects were drawn from eligible households in the study areas. The sample size was calculated based on the known prevalence of low dietary diversity [6]. A single population proportion formula was used to estimate the sample size needed regionally based on the prevalence of indicators using a 0.05 desired leave standard error, a 95% confidence level, and a design effect of 1.5. The sample size was adjusted for region-specific average household size, region-specific percentage of the target population, a household response rate of 94.5%, and an individual response rate of 80%.

where, n = sample size, Z _𝞪/2 = standard errors from the mean corresponding to the 95% confidence level = 1.96, P = prevalence, d = allowable error = 0.05, DEFE = design effect = 1.5, Ave. HH size = average household size from each region, %of Target PP = proportion of the target population from each region, HHRR = household response rate (%) = 94.5%, IRR = individual response rate (%) = 80%. After using the formula, we calculated the total sample size to be 1,980.

Data collection and measurement

A structured questionnaire with four modules was designed and administered to gather comprehensive information on respondents’ characteristics and various topics related to food systems and dietary outcomes. The questionnaire was created in English but was translated into Amharic and Affan Oromo as spoken in the survey area. The questionnaire was programmed in the CSPro software package using pre-coded responses, and tablet computers were used to collect the data. Several procedures were followed during the survey’s design and implementation to ensure data quality. First, data collection training modules and field guidelines were prepared. Qualified and experienced data collectors were recruited and underwent two weeks of intensive data collection training. After the training, a mock test was conducted among the participants. Finally, a pre-test was conducted in the Oromia region (in Bishoftu, near Addis Ababa), and after the necessary adjustment, the survey tool was finalized.

Socio-demographic and time use

The socio-demographic and the women’s time-use questionnaire were obtained from Ethiopia Central Statistics Agency (CSA) health and demographic and time-use surveys. These tools underwent testing and validation [2, 27]. The household characteristics module collected data on household members’ age, sex, education, residency, assets, wealth, employment, and income. It also examined kitchen characteristics such as the source of drinking water, hand washing facilities, cooking fuel, cookstove type, available cooking utensils, and propensity to prepare new food. The women’s time use module gathered information on how women allocate their time.

Nutrition-related KAP assessment

The nutrition-related KAP module followed the guidelines of the Food and Agriculture Organization (FAO); this tool has been tested and validated in Malawi, Cambodia, and Mexico [28]. In this study, we assessed participants’ nutrition-related knowledge by calculating the correct answers to 27 questions across three key areas: infant and young child feeding, micronutrient intake, and malnutrition. Each question was answered with a yes or no response; correct responses were awarded one point, while incorrect responses received a zero score. Good knowledge was where the participant correctly answered more than 50% of the knowledge questions. Attitudes toward nutrition were measured using a three-point Likert scale comprising 1 (agree), 2 (neutral), and 3 (disagree) with a total of 23 questions. These questions were in four areas: benefits of nutrition, barriers to healthy eating, susceptibility to nutritional issues, and severity of dietary complications. The total nutrition-related attitude score was compiled from these subcategories, and the average score was then calculated based on the number of statements agreed upon. A positive attitude was where the participants agreed with more than 50% of the attitude questions. For nutrition-related practices, 18 questions were asked, which included actions such as handwashing, using clean fuel, treating water, attending cooking demonstrations, and food preservation. The results were scored based on recommended health and nutritional practices, earning one point for appropriate practices or zero otherwise. The average score was calculated. Good practices were where the participant reported applying more than 50% of recommended practices.

Dietary quality assessment

The food frequency and dietary diversity module was adapted from the Food and Agriculture Organization (FAO) dietary assessment tools [5, 29]. The food frequency questionnaire was tested and validated in Ethiopia, Tanzania, and Morocco [30,31,32,33]. Tailored to fit Ethiopia’s list of available food options, it comprises 86 specific lists of food and beverages and offers response categories to indicate the frequency of consumption over seven days. Also, the individual dietary diversity questionnaire (24-hour dietary recall) was tested and validated for several age/sex groups as proxy measures for macro and/or micronutrient adequacy of the diet [3, 5, 34]. Dietary quality was assessed by determining the dietary diversity score (DDS), minimum dietary diversity for women (MDD-W), and mean adequacy ratio (MAR). DDS is the average number of food groups women consume, calculated by summing the ten food groups (Grains, white roots and tubers, and plantains; pulses (beans, peas, and lentils); nuts and seeds; milk and milk products; meat, poultry, and fish; Eggs; Dark green leafy vegetables; other vitamin A-rich fruits and vegetables; and other fruits) consumed during the last 24 h [5]. MDD-W was calculated for women who have consumed at least five of the ten possible food groups during the last 24 h [4]. The MAR was calculated as an overall measure of nutrient adequacy.

Nutrient intake was estimated using a seven-day food frequency and calculated using an Ethiopian food composition Tables [30, 35]. Missing values were added from the Kenyan food composition Table [36]. The NAR was calculated as the women’s nutrient intake ratio for both macronutrients and micronutrients relative to the recommended allowance in the Ethiopian Food-based dietary guidelines and Dietary Intake Reference [29, 37, 38].

Econometrics approach

An econometric approach was employed to analyze three distinct dietary outcomes for women: DDS, MDD-W, and MAR, which served as the dependent variables. The analysis aimed to establish the relationships between these dietary outcomes and various factors, including nutrition-related KAP, and kitchen characteristics, which were considered the independent variables that facilitate or hinder outcome variables. To analyse the first dependent variable, the DDS, we had to consider that it can only take nonnegative integral values. Therefore, a count data modelling approach was necessary, and the Poisson model was a suitable choice as it accommodates the discrete nature of the dependent variable. In contrast, the MDD-W is a binary value, so the logistic model was appropriate. The MNA is continuous, and for this result, the ordinary least square method (OLS) method was suitable for modelling. We modelled women’s dietary outcomes (DD, MDD, and MNAR) as a function of nutrition-related KAP, kitchen characteristics, and selected sociodemographic characteristics using the following regression model:

Where; (:{DDs}_{i}) is dietary diversity score, (:{MDDs}_{i}) is minimum dietary diversity score, (:{MAR}_{i}) is the mean adequacy ratio of nutrients, (:{KAP}_{i}:)is a vector of variables capturing nutrition-related knowledge, attitude, and practice, (:{KC}_{i}) is a vector of variables capturing kitchen characteristics, (:{SD}_{i}) is a vector of variables capturing socio-demographics, and (:{epsilon:}_{i}) is a random error term.

Data analysis

We used a combination of quantitative methods to gain insights into the factors affecting women’s dietary quality. SPSS version 16 statistical software was used to analyze descriptive statistics such as the mean, median, standard deviation, and percentages of the collected data. To assess the normality, we used the Shapiro-Wilk test. To further examine and understand the influencing factors, we deployed multinomial regression, which allowed us to determine the extent to which various factors contributed to the quality of women’s diets. Following regression analysis, we conducted post-estimation tests to verify the accuracy and validate the assumptions. To check for homoscedasticity, we used Cameron and Trivedi’s decomposition test and the Breusch-Pagan/ Cook-Weisberg test for heteroscedasticity, while for multicollinearity, we employed the Variance Inflation Factor (VIF) test. Missing data have been imputed. Variables with a p-value less than 0.05 were considered to indicate statistical significance.