**3. Materials and Methods**

Previous studies [60] were mainly related to primary empirical research. Specifically, they allowed the identification of the potential blocks of environmental determinants a ffecting a country's food security, such as: (1) Measures concerning natural resource availability and usage; (2) energy production and consumption items; (3) fertilizer usage; (4) greenhouse gas emissions by agricultural enterprises; (5) parameters of agribusiness yield. In turn, as a result of this literature review, a set of 37 environmental determinants was collected from the World Bank DataBank [61] and the United Nations Environment Program Data Explorer [62]. Correlation analysis helped to select the most influential

factors and eliminate multicollinearity problems. It allowed the choosing of 14 out of 37 environmental factors. Additionally, two of these 14 variables were eliminated because they had negative influences on regression model quality parameters. Therefore, previous research [60] helped to clarify a set of environmental factors that do have an impact on a country's food security.

The realization of this research task implied the need for several stages: (1) Construction of the comprehensive food security indicator; (2) identification of certain ecological factors influencing food security in the short and long run.

In general terms, the research was based on data collected from public sources (the World Bank DataBank [61], the United Nations Environment Programme Data Explorer [62], and the Food and Agriculture Organization of the United Nations database (FAOSTAT) [63]) for 28 post-socialistic countries (Albania, Armenia, Azerbaijan, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Georgia, Hungary, Kazakhstan, Kyrgyz Republic, Latvia, Lithuania, Macedonia, Moldova, Montenegro, Poland, Romania, Russia, Serbia, Slovak Republic, Slovenia, Tajikistan, Turkmenistan, Ukraine, and Uzbekistan) from 2000 to 2016.

As for the first stage, it might be noted that The Economist in cooperation with the FAO have developed the Global Food Security Index, which consists of 28 measurement indicators of a ffordability, availability, quality, and safety of food. Nevertheless, this index has been calculated from 2012, which is too small a period for gaining reliable modeling results. That is why the Food Security Index (FSI) was constructed. The FSI consists of 19 indicators of food availability, food access, food stability, and food utilization. The FAO o fficially identifies these parameters as measures of food security. The descriptions of the indicators used for the FSI's construction are in Table 1.


**Table 1.** Measurement indicators of the Food Security Index (FSI).

The FAO does not clarify a certain algorithm for aggregation of food availability, food access, food stability, and food utilization indicators. Therefore, the Principal Component Analysis (PCA) in Stata software was used to realize this particular task. Namely, the eigenvalues of the first principal component were used as weighted coe fficients for the FSI's construction. It is worth noting that we use the PCA method rather than the Analytic Hierarchy Process (AHP) because it is a rather complicated task for realizing pairwise judgments to prioritize measures of food security on a scale of 1 to 9. Thus, we decided to apply not a subjective, but a more objective method (PCA), which aimed at clarification of data trends and identification of weight coe fficients based on them [64]. In addition, before applying the PCA, all of the above-mentioned indicators were primarily normalized considering their stimulating or unstimulating influence on the state of countries' food security. The normalization process allows us to arrange them from 0 to 1.

In turn, the second stage of the research is focused on the identification of environmental determinants influencing a country's food security in short- and long-run perspectives. As the research sample includes rather huge number of observations, both in terms of periods, countries, and independent variables (panel data sample), a pooled mean-group (PMG) estimator, developed by Pesaran, Shin, and Smith [65], was used. Traditionally, in research based on panel data with a large number of cross-sections but a small number of time observations, fixed e ffects are applied, as well as random e ffects estimators or generalized method of moments. However, an increase in the number of time observations might result in non-stationarity. As this research covers a rather large number of cross-sectional observations and time observations, it is better to apply the PMG estimator. Moreover, this research method allows us to manage the problem of non-stationarity and better fits heterogeneous panels. In addition, the PMG estimator considers both pooling and averaging approaches (it allows short-run coe fficients to di ffer across countries, but long-run coe fficients might be equal for the whole panel). Thus, it helps to mix some technical aspects from the mean group estimator and fixed e ffects estimator [66].

The PMG estimator allows testing of the hypothesis about the existence of influence on food security (specifically, the FSI) in the long-term and short-term perspectives of the following environmental indicators: X1—access to clean fuels and technologies for cooking (% of population); X2—access to electricity in rural areas (% of rural population); X3—agricultural methane emissions (% of total); X4—agricultural nitrous oxide emissions (% of total); X5—arable land (% of land area); X6—cereal yield (kg per hectare); X7—CO2 emissions (metric tons per capita); X8—electric power transmission and distribution losses (% of output); X9—electricity production from renewable sources, excluding hydroelectric (% of total); X10—fertilizer consumption (kilograms per hectare of arable land); X11—forest area (% of land area); X12—renewable electricity output (% of total electricity output). The summative statistics for the set of dependent and independent variables are in Table 2.


**Table 2.** Summative statistics for the set of variables.

Notes: X1—access to clean fuels and technologies for cooking (% of population); X2—access to electricity in rural areas (% of rural population); X3—agricultural methane emissions (% of total); X4—agricultural nitrous oxide emissions (% of total); X5—arable land (% of land area); X6—cereal yield (kg per hectare); X7—CO2 emissions (metric tons per capita); X8—electric power transmission and distribution losses (% of output); X9—electricity production from renewable sources, excluding hydroelectric (% of total); X10—fertilizer consumption (kilograms per hectare of arable land); X11—forest area (% of land area); X12—renewable electricity output (% of total electricity output); Obs—amount of observations; Std. Dev.—Standard deviation.

Based on the results presented in Table 2, it should be noted that the number of observations di ffers for some variables. Nevertheless, the panel is strongly balanced, which allows us to ge<sup>t</sup> reliable and significant empirical research results.
