The primary objective of this study was to examine the effect of income diversification on food security in the study areas. Hence, both quantitative and qualitative approaches were employed in the study.
2.4. Measuring Food Security
Various approaches have been used to measure different dimensions (availability, access, utilization, and stability) of food security. However, it was not been easy to get a single measure that was comparable over time and places, and also captured all these dimensions at the same time [
19,
20,
21,
22].
Most researchers and policymakers use two commonly used measures of food insecurity which are household calorie consumption (nutrition-based) and a measure related to household perspectives on food security (experience-based) [
17]. These two household-level measures assess one dimensions of food security which is access to food. Hence, in this study, to measure the household food security status (access), the study used both daily adult equivalent per capita calorie consumption (nutrition-based), a quantitative approach, and the experiential measure, a qualitative approach.
To determine the calorie intake of individuals, we used the Ethiopian food composition table, part III, which was developed by the Ethiopian Health and Nutrition Research Institute (EHNRI) [
23]. We identified 103 food items out of 769 food items that were commonly used in the study areas. Hence, the foods eaten by the family members in the last seven days were collected. We used local common instruments such as bowl, sack, cup, can, pot, spoon, kerepe, and others to measure the size of food, and converted them into kilograms. However, the foods consumed outside of the home and the possible food wastes were not considered. Nevertheless, in rural Ethiopia, people do not commonly buy and eat outside the home, except in a few cases when people may be offered food by their neighbors, friends or relatives. Moreover, we observed that the magnitude of food waste after table (leftovers) in rural areas of the country is smaller than the food waste in urban areas.
Concerning the experience-based measurement, the HFIAS measure used by the U.S to estimate the prevalence of food insecurity in the country was adapted [
20]. The measure was designed to capture the predictable reactions and responses of a household indicating that food was of insufficient quality and quantity, and that there was anxiety over the food supply (for the period of the previous 30 days in our case). Finally, to facilitate the application of the statistical model, the food insecurity continuous data were converted into a binary form by classifying households as food secure, where the household did not experience any food insecurity situations and rarely worried about having enough food, or otherwise, as food insecure [
17].
2.5. Data Analysis
Researchers have used different methods to measure the level of household income diversity. The Herfindahl–Hirschman index, Shannon’s diversity index and Simpson’s index of diversity (SID) are among the most commonly used measures [
24,
25]. Recently, the SID has been widely adopted to measure the effect of income diversity on rural livelihood [
8,
11,
25]. Therefore, we adapted the SID method for this study, and it is specified as follows:
where, SID is the Simpson’s index of diversity, N is a total number of sources of income, and
is the proportion of household income coming from i
th sources. The value of SID always falls between 0 and 1. The SID was equal to zero if a given household had only one source of income and approached one as the number of sources of income increased. The closer the value of the index was to one, the more the degree of diversification, and the closer the value of the index was to zero, the more the degree of specialization.
Additionally, the daily calorie intake method (reported undernourishment or nutrition-based) and household perceptions on food security (experience-based) were used to measure food security (access). Hence, in this study, the average daily minimum calorie intake of a moderately active adult equivalent, 2280 kilocalorie (kcal), was used as the food security threshold [
14,
26]. Consequently, the food security index (FSI) was determined as follows [
25].
where, FSI is the food security index (status) of i
th household,
is the actual daily per capita calorie intake of the i
th household, and R is the recommended per capita daily calorie intake (2280kcal). When, FSI ≥ 1, the given household was categorized as food secure, otherwise, it was categorized as food insecure.
A binary logistic regression was also employed to determine the effect of household income diversification on food security. For estimating the logit model, the dependent variable was food security which takes a value of 1 if the household is food secure and 0 if not. Thus, mathematically the logit model was specified as follows [
27]:
where,
is the probability of household being food secure.
where
i = 1, 2, 3…n,
is intercept,
is regression coefficients to be estimated,
is a disturbance term that affects
but not observed by the researcher, and
is vectors of independent variables.
The probability that a household was food insecure is:
Then, the odds ratio is written as:
The left-hand side of Equation (4),
, is simply the odds ratio in favor of food secure. It is the ratio of the probability that a household is secure in food to the probability that the household is not secure in food. Finally, by taking the natural log of Equation (4) the log of odds ratio is written as:
where,
is log of the odds ratio in favor of food secure, which is not only linear in
but also linear in the parameters.
Moreover, multiple linear regression was used to determine the effect of income diversity on the food security index. The model is specified as follows [
27]:
where,
is the food security index;
,
, …
are explanatory variables;
0 is the intercept,
1 is the slope coefficient on
,
is the coefficient on
, and so on; and
is error term.
The study employed an instrumental variable (IV) method to overcome the omitted variable bias and the possibility of reverse causality where the food security condition of a household determines the level of household income diversification (an endogeneity bias). Thus, the household income diversity index was treated as an endogenous variable and access to credit service, proximity to market, and having training or special skill in craftsmanship were considered as instruments. This approach is similar to that used by Babatunde and Qaim [
12] in different contexts. Studies suggested that access to credit and proximity to a market encourage households to engage in various economic activities, and therefore income diversity increases [
28,
29]. Additionally, if the head of the household or any other family member has training or a special skill in craftsmanship, it initiates that family to engage in different non-farm activities available in their area. Zerai and Gebreegziabher [
30], for instance, identified that households with special skills initiated the households to look for non-farm activities in addition to farm work. Similarly, Amogne et al., [
31] suggested that the lack of entrepreneurial training was among the main factors that limited households to not look for additional sources of income.
The chosen instruments were valid (did not have direct correlation with food security) because of the following reasons: First, in Ethiopia, government and private financial institutions give credit to rural households with a strict follow-up to ensure that the farmers utilize the money for the intended purpose (mainly to do a business or to finance their agriculture). Hence, credit services are not aimed at buy food materials. Secondly, in rural Ethiopian situations, the lack of infrastructural facilities (road and transportation) exacerbates the food security condition and prohibits households from accessing the market rather than the distance to the market. In this study, for instance, the average distance from a household to the nearby market (Ambo town) was only 1.53 hours. Third, special skills, training or experience of households encourages them to get involved in different income generating activities which impacts the food security status of the households.
Finally, an independent samples t-test was used to determine whether the mean of the SID and daily calorie intake of the rural and semi-urban households were statistically different from one another or not.