The paper draws on the Gallup World Pool (GWP) data, that included the FIES in 2014. The FIES module surveys an experience-based food insecurity individual scale, together with other social, economic and demographic characteristics of the respondents (Gallup 2017). The GWP conducts nationally representative surveys annually in 147 countries. The study adopted a three-stage sampling procedure to select the sample (ibidem). Country sample sizes of 1000 individuals, representative of the male and female resident population aged 15 and over (in very large countries such as India and China, sample sizes up to 5000), obtaining a sample of 150,000 individuals. The survey enables the collection of cross-culturally comparable information from individual respondents, and allows also an estimate of food insecurity in rich and developed countries.
The Food Insecurity Experience Scale
The importance of a measure of food insecurity centred on individuals rather than only on countries or regions was recognised starting with the work of Sen in the eighties (Sen 1981, 1983). At the World Food Summit in 1996, an innovative definition of food security was developed: “Food security, at the individual, household, national, regional and global levels [is achieved] when all people, at all times, have physical and economic access to sufficient, safe and nutritious food to meet their dietary needs and food preferences for an active and healthy life” (FAO 1996). By explicitly acknowledging the four interdependent pillars of food availability, access, utilisation and stability, the World Food Summit (FAO 1996) marked a milestone contribution in the analysis of food security, which, until then, used to be identified only by food availability, that typically refers to Countries rather than on individuals.
The financial, economic and food crisis in 2008 fostered the need for data on food insecurity at individual and household levels, as many studies demonstrated that households that were more vulnerable before the financial crisis saw a worsened effect in terms of food insecurity with the crisis (Bloem 2010; Vilar-Compte et al. 2015). Disparities also increased among men and women, poor and not poor households, educated or uneducated people (d’Errico et al. 2018; FAO 2010), and smaller households performed much better than households with more members (Lokosang et al. 2016). These direct measures are intended to capture a household’s or individual’s reported experience of the problem through responses to validated survey items that are transformed into a scale (Coates et al. 2006). FIES data can be analysed at the individual level, allowing the analysis of inequalities in access to food by gender, taking into account several personal and household characteristics, and recognising that households do not necessarily distribute resources equitably and should not be conceived as a unique entity (Brunelli and Viviani 2014).
FIES measures the severity of food insecurity based on eight questions that ask people directly about having to compromise the quality and quantity of food, due to limited money or other resources to obtain food.
The FIES score—e.g. the number of positive answers to the FIES module—is a sufficient statistic to represent the severity of food insecurity of the respondents on an ordinal scale (Grimaccia and Naccarato 2019a; Nord 2014; Nord et al. 2016). This means that the FIES score is a quantitative, discrete and ordinal variable, with values ranging from zero (no events of food insecurity) to 8 (all events of insecurity), and it is a sufficient statistic for the latent trait that is being measured: experienced food insecurity.
Descriptive Analysis of FIES and Related Covariates
Together with the FIES module, other meaningful covariates have been surveyed. These variables are closely related to food insecurity, and they represent geographical, demographic, economic and social features. The variables considered are gender, education, household income, household composition (couples, lone parents, with or without children) and age of the respondents.
Other than the already mentioned factor “gender”, another important demographic characteristic related to food security that has been analysed in previous researches is age (Nord 2003; Strickhouser et al. 2014).
Considering poverty (measured by international poverty lines), three quarters of the developing world’s poor still live in rural areas. But, the share of the poor living in urban areas is rising, and more rapidly than for the population as a whole (Ravallion et al. 2007). As reported by the World Bank (2017), poverty rates are falling in both urban and rural areas, but they are lower in urban areas. Nevertheless, poverty is becoming “more urban” in more urbanised regions (Olinto et al. 2013).
As called for by eminent scholars (Champion and Graeme 2003), FIES data allow the classical distinction between rural and urban dwelling to be overcome. In the FIES data, it is possible to distinguish different kinds of settlements such as “a rural area or on a farm”, “a small town or village”, “a large city”, or “a suburb of a large city”.
The analysis takes into account other meaningful covariates. First of all, the level of education of the respondents has been acknowledged as an important driver against food insecurity (Bartfeld et al. 2006; d’Errico et al. 2018; Nord and Hopwood 2008): education is a good proxy of social status, and it is related to employment.
In the dataset, education is classified according to three levels: elementary, corresponding to having completed elementary education or less (up to 8 years of basic education); secondary, or having completed some education beyond secondary education (9 to 15 years of education); and tertiary, if the respondent has completed 4 years of education beyond “high school” and/or received a 4-year college degree (Gallup 2017).
In FIES dataset, globally the share of women who experience food insecurity is larger than men: 45% of the female population present at least one symptom of food insecurity compared with the 43.3% of men (Fig. 1).
The gender difference is widespread across every continent, including Europe. In order to evaluate the significance of these differences, we rely on the estimation of a micro-econometric model, that benefits from the large sample.
The Model
We analyse the individual FIES data on experienced food insecurity depending upon exogenous characteristics such as gender, age, level of education and household economic and social covariates. The analysis of results of the model offers a gain in knowledge on the gender differences in food insecurity in different territorial areas of Europe. In this way, we provide new evidence concerning the complex link between food insecurity and gender, controlling for other meaningful covariates.
An extensive analysis of the drivers of experience-based individual food insecurity show that the variables significantly impacting food insecurity are related to social, demographic and economic characteristics of the population, such as education, household income, household composition (couples, lone parents, with or without children) and age (Grimaccia and Naccarato 2018, 2019b).
In this work, first of all, we have analysed how FIES was related to the socio-economic status of the respondents across the globe, in order to highlight the different impact of each covariates to food insecurity in Europe, and—for comparison purposes—in other world continents. Furthermore, we evaluate more specifically the effects of such drivers separately for the female and male population in Europe. Finally, we extended our results analysing the differences in the determinants of food insecurity in different areas of Europe (Northern, Western, Southen, Eastern), that—according to the literature—present different socio-economic conditions.
To verify whether the observed differences (Sect. 2.2) are significant, we analysed—through an Ordered Logistic Model (OLM)—the relationships among FIES and the covariates described below. All the categorical variables, in order to be included in the model, have been transformed in dummy variables, thus allowing the estimation of a coefficient for each value of the variable. Among the observable individual characteristics, we take into account two dummies (male and female) for the dichotomous variable related to gender, age (both the linear and the quadratic relationships), three dummies for marital status (single, married or in a relationship, widows, divorced and separated), and level of education (primary, secondary and tertiary). The household economic and social covariates that we included in the model are two dummies for living in extreme poverty or not, the number of children in the household, and four dummies related to the kind of settlements where the respondents live (a farm or a rural location; small town; big city; or the suburb of a big city). A territorial specification of the model has been included, accordingly to the geographical extension of the model: a dummy for each region has been included to consider in the model a characterisation of the different territorial specificities (territorial fixed effects).
Then, the multivariate set-up we rely upon for our model is:
$$ \begin{aligned} y & = {\text{ologit}}\left( {\text{y}} \right) = \alpha {\text{c}} + \beta_{1} \;{\text{male}} + \beta_{2} \;{\text{female }} + \beta_{3 } \;{\text{age }} + \beta_{4} \;{\text{age}}^{2} + \beta_{5} \;{\text{single }} + \beta_{6 } \;{\text{married}} \\ & \quad + {\kern 1pt} \beta_{7} \;{\text{widow}}\;{\text{separated}}\;{\text{divorced}} + \beta_{8} \;{\text{primary}}\;{\text{education}} + \beta_{9} \;{\text{secondary}}\;{\text{education}} \\ & \quad + {\kern 1pt} \beta_{10 } \;{\text{tertiary}}\;{\text{education}} + \beta_{11} \;{\text{extreme}}\;{\text{poverty}} + \beta_{12} \;{\text{no}}\;{\text{extreme}}\;{\text{poverty}} \\ & \quad + {\kern 1pt} \beta_{13} \;{\text{rural}}\;{\text{area}}\;{\text{or}}\;{\text{farm}} + \beta_{14} \;{\text{small}}\;{\text{town}}\;{\text{or}}\;{\text{village}} + \beta_{15} \;{\text{large}}\;{\text{city }} + \beta_{16} \;{\text{suburb}}\;{\text{of}}\;{\text{a}}\;{\text{large}}\;{\text{city}} \\ & \quad + {\kern 1pt} \beta_{17} \;{\text{geographical}}\;{\text{fixed}}\;{\text{effects}} + \varepsilon \\ \end{aligned} $$
(1)
where the endogenous variable y is the sum of the affirmative answers to the FIES module.
Estimates were conducted using an OLM. As the variable measuring the experienced food insecurity is an ordinal variable (Sect. 2.1), a non-linear model has been preferred (Espinosa and Hennig 2019; Grimaccia and Naccarato 2019b; Agresti 2010). That is because we cannot state that the distances between categories are the same for all of them. For instance, we cannot affirm that the difference between two and zero is twice as important as the difference between five and four (Wooldridge 2012).
An OLM for an ordinal response Yi with C categories is defined by a set of C-1 equations where the cumulative probabilities \( g_{ci} = \Pr \left( {Y_{i} \le y_{c} |x_{i} } \right) \) are related to a linear predictor \( \beta^{{\prime x_{i} }} = \beta_{0 } + \beta_{1} x_{1i} + \beta_{2} x_{2i} + \beta_{3} x_{3i} + \cdots \) through the logit function.
$$ logit\left( {g_{ci} } \right) = \log \left( {\frac{{g_{ci} }}{{1 - g_{ci} }}} \right) = \alpha_{c} + \beta 'x_{i} ,\;\;c = 1,2, \ldots ,C - 1 $$
(2)
The parameters αc, called thresholds or cut-points, are in increasing order (c1 < c2 < …), and C = 1, 2,…,C − 1, where C is the number of categories of the ordinal variable. These cut-points reflect the predicted cumulative probabilities considering the covariates are all equal to zero. The last cumulative probability is necessarily equal to 1, so the model specifies only C-1 cumulative probabilities. When there are only two outcomes (zero and one), we set the single cut-point to zero and estimate the intercept; this approach leads to the standard logit model (Wooldridge 2012).
First of all, we estimate model (1) for the whole sample in order to underline that food insecurity exists also in Europe. Moreover, compared with other continents, Europe presents a significant gender difference. Results of the estimation of the model are shown in Table 1, column (1).
Table 1 Coefficients and standard errors for the determinants of FIES score across the globe—2014 As the significant gender coefficient in Europe (Table 2), in order to highlight whether the impact of exogenous variables varies by gender, we estimate model 1 separately for men and women (Table 2). This allows to obtain coefficients that indicate which driver is significant for each gender separately (Agresti 2010; Wooldridge 2010), and then to emphasise the difference of the impact of each covariate for men and women.
Table 2 Coefficients and standard errors for the determinants of FIES score in Europe by gender—2014 Subareas of Europe present very different characteristics of the population in relation with income per capita, distribution by age and level of education (Eurostat 2019). In order to give an account of these specific features, we include in the model dummies (areas’ fixed effects) corresponding to more homogeneous areas with respect to the geopolitical usual distinction. In this way, the estimated coefficients evaluate the impact of living in each European area on the probability of experiencing food insecurity. The corresponding coefficients are significant. Therefore, we estimate the model for the five areas separately in order to verify if any differences exist in the main drivers (Table 3).
Table 3 Coefficients and standard errors for the determinants of FIES score in Europe by gender including European main regions effects—2014 The obtained coefficients are comparable since the structure of the survey, the definition of the variables and their categories are the same in each and every territorial area, as in Chen et al. (2019), Espinosa and Hennig (2019), Williams (2016). The sample is, indeed, representative for each country and for both sexes (Gallup 2017). This allows to compare the results of the estimations and the significance and the sign of the coefficients in every subsample.