Social programs aimed at poverty and food poverty in Mexico

One of the major social challenges faced by Mexico is reducing poverty (^{ENSANUT 2016}). According to CONEVAL, in the year 2018 poverty in Mexico afflicted 61.1 million people (48.8% of the total country population), of which 21 million were extremely poor, and at least 25.5 million (20.4%) lacked access to basic food (^{CONEVAL, 2019}). Nonetheless, another recent study reports that population reaching the food security range has contracted, from 50.1% in 2000 to 46.7% in 2018, which expresses that more than half of Mexico´s population present some level of food insecurity (^{Torres and Rojas 2020}).

In order to reduce poverty, social programs were developed and introduced by federal government, but they are focused mainly on economic welfare (25%), education (29%), health (17%) and only 4% is applied to reduce huger and the lack of food (^{CONEVAL, 2017}).

Social food policy in Mexico, dates back to the Mexican Revolution (^{Barquera et al, 2001}). The main objective has been to guarantee peoples’ access to food that allow them to satisfy their needs, and to fulfill the nutritional requirements necessary for a healthy living. (^{INSP, 2007: 91}). Throughout the history of social policy, different food programs had emerged. They can be classified in two groups: 1) programs focused in food security, subsides for productive activities and self-consumption; 2) strategies oriented to end poverty, transfer programs and subsides (^{Huesca et al. 2016}).

Since 2013, Mexican federal government changed its approach of solving poverty issues, by launching a comprehensive transversal strategy to combat poverty, malnutrition, and social marginalization (^{López and Gallardo 2015}) entitled National Crusade Against Hunger (CNCH by its Spanish acronym), in accordance with the United Nations´ Zero Hunger Challenge (^{Gills et al. 2015}). The overall strategy of the CNCH includes a set of articulated public programs attempting to address the problem of social policy fragmentation (^{Cejudo and Michel 2015}). This latter topic is an indispensable condition for public policies in order to solve complex problems as a part of the same sphere, such as poverty and food poverty in particular (^{Cejudo and Michel 2016}). In this sense, the CNCH is not a program but a strategy looking to achieve food security -and other several social goals-, through community participation and the cooperation of several government ministries (^{Gil et al., 2014}).

One of these CNCH programs was the so called Programa de Apoyo Alimentario (Food Support Program; PAL by its Spanish acronym), aimed at increasing access to basic food for beneficiary households by means of money transfers in order to compensate families´ low incomes, enhancing their economic welfare and nutrition. PAL is focus on families with a per capita income below poverty line, and it represents the transition to the Prospera (Prosper) program from the former Oportunidades (Opportunities) program, but its coverage is proportionately low, as ^{Morales-Ruan et al. (2013)} had estimated that by 2012, PAL was attending only 2% of households nationally.

PAL was operated using an electronic debit card for money transfers, and the program had two operating modalities: the conventional modality allowing beneficiaries to purchase items of their free choice at any shop or store with an electronic terminal (PAL-Cash hereafter), and a modality called PAL-Scheme Without Hunger (PAL-ESH by its Spanish acronym) exclusively allowing the usage of the electronic card for acquiring 19 basic food products at lower prices that distributed through the national network of Diconsa (with over 27,000 selling points). Diconsa was a federal public company dedicated to ensure food supply and other basic commodities in high and very high marginalized areas, and by the second half of 2013 was in charge of operating the PAL-ESH (^{Huesca et al., 2016}). Beneficiaries of PAL-ESH had received additional support to the PAL-Cash. As a reference, from the beginning of 2015, a regular family could receive annually from $ 454.27 US dollars for households with no children and up to $ 747.35 US dollars for families with three children under nine years old.

The structural equation models (SEM) and its application to calculate satisfaction models

The satisfaction index calculus arose from the necessity to explain the customer linkage to the goods and services supplier through their own satisfaction. Nowadays, there are several national satisfaction indexes that have been constructed for particular socioeconomic and cultural contexts. For instance, there is the Swedish Customer Satisfaction Barometer (SCSB), the American Customer Satisfaction Index (ACSI), the German Barometer, the Norwegian Barometer (NCSB), the Swiss Index (SWICS), the Korean (KCSI), the Malaysian (MCSI) and the Canadian (CCSI) (^{Johnson et al., 2001}). Countries such as Brazil, Argentina and Mexico are developing their own national consumer satisfaction index (^{Rodríguez, 2012}).

Usually, these index focus on perceptions and attitudes, which are abstract items that allow to include a measurement of the satisfaction of some goods and services (^{Hayes, 2012}). The knowledge of beneficiaries' perceptions and attitudes concerning the products and services offered, in this case by a social program such as the PAL-ESH, provides a solid basis to designers, operators, and program managers in general, to enhance the decision-making process applied to implementation and results of this program.

As a background of measuring beneficiaries´ satisfaction from social programs in Mexico, in 2011 ^{Lobato et al. (2011)} developed the Mexican User Satisfaction Index (IMSU), based on the American Consumer Satisfaction Index model (ACSI), and applied it to the social program "70 and above"; obtaining a satisfaction index of 79% using PLS. Similarly, ^{Rodríguez (2012)}, developed the Mexican Index of Beneficiary Satisfaction (IMSaB) applied to some social programs implemented by SEDESOL in Mexico, whose estimation was obtained by the maximum likelihood method. This index was applied empirically on four social programs: the program which obtained the highest valuation was the Liconsa, Social Supply Program of Milk (91.2%), and the lowest level of satisfaction was the Productive Options Program (65.9%). The variables "Quality of Benefit" and the "Quality of management” resulted the best ones explaining the variable "Satisfaction" (^{Rodríguez, 2012}). This result about the Liconsa program is consistent with ^{León and Christian (2010)}, and ^{Cogco et al. (2013}) whom report values of 91% and 91.2% respectively.

Finally, the method of estimating Partial Least Squares (PLS) applied to Structural Equation Modeling has become a standard tool for analyzing complex inter-relationships between observed and latent variables in social science research (^{Sarstedt et al., 2020}), and also has been recognized as one of the most effective methods to explain the residual variance of the LV (^{Esposito et al., 2010}), and when the structural model is complex and includes many constructs, indicators and / or model relationships (^{Hair et al., 2019}). In addition, it has the advantage of not requiring strong assumptions about the normal distribution of the data, size of the measurement scale, etc. (^{Esposito et al., 2010}; ^{Bonales et al., 2018}). However, these advantages must be contrasted with some disadvantages, for example, the absence of a global optimization criterion implies the lack of fit measures of the model as a whole (^{Hair et al., 2010}). Although the possible bias in the estimation decreases with the increase in sample size (^{Lobato et al., 2011}).

Generalities

The beneficiaries of the PAL-ESH program were considered as the study group. For comparative analysis, the results were contrasted between different types of beneficiaries according to two criteria: on the one hand, type of locality in which they live (i.e. rural or urban settlements), and on the other hand, the degree of marginalization (high or very high). These groups were identified from a proxy variable (i.e. the modality in which the governmental support is delivered) as follows: Group 1 includes beneficiaries from high marginalized rural areas (customers of Diconsa rural stores); Group 2 includes beneficiaries from very high marginalized rural localities (customers of Diconsa truck-stores); and Group 3 includes beneficiaries from high marginalized urban localities (customers of Diconsa CABES).

Type and size of sample

The total sample size for the national survey (n = 3,199) was the result of estimating an independent sample for each of the three compared groups, ensuring a level of confidence of 95% and a 3.2% margin of error, with the following sizes: Group 1 = 1,246 households, Group 2= 994 households, and Group 3= 959 households.

We applied the cluster sampling method in two stages: in the first stage, our reference was the rural store DICONSA selecting those cases that had the three groups of beneficiaries; in the second stage, for each group we chose the beneficiaries randomly, taking into consideration the required number of questionnaires (between 8 to 30 applications according to the group). The survey was conducted between the end of 2015 and early 2016, with a total of 211 locations covering 82 municipalities in 26 Mexican states. (Figure 1).

Source: Author’s own.

Figure 1 Sample profile and sampling points distribution 

Instrument

We used a Likert scale questionnaire with seven categories. The Reliability of the scale was assessed with Cronbach´s Alfa coefficient, obtaining a value of 0.863, which is highly acceptable for this kind of study (^{Hernández et al. 2010}). The main statistical descriptors for each sample are showed in Annex A1, including age, years of education, family income (monthly) and family food expenditure (weekly).

Data was analyzed by Statistical Program for Social Sciences (SPSS) version 22 for descriptive statistics and reliability tests. The structural equation modeling was made using the partial least squares method with the SmartPLS software package (Informer Technologies Inc.).

Stages for building the SEM

The structural equation model was based on ^{Hair and authors (2010)}, following five stages: (i) model specification: the theoretical and empirical knowledge upon beneficiaries´ satisfaction was applied to study the relationship of the causal effects of the manifest variables (observables) on the (unobservable) latent variables; (ii) model identification: we assured that they could be estimated model parameters, following identification rules, so-called conditions of order and identification range; (iii) model estimation: it consisted in obtaining the values of the parameters that fit best into the observed matrix; this was done by the Partial Least Squares method; (iv) model evaluation: it was applied a test of goodness of fit of X² as well as the reliability composite test (composite reliability) and the average variance extracted (average extracted variance); and (v) index construction: this index was calculated considering the estimated parameters for each case and proceeding to apply the corresponding measurements. The generalities for structural equation models are shown in Annex A2.

PAL-ESH beneficiaries’ satisfaction model

The model to assess beneficiaries’ satisfaction with PAL-ESH began with the conceptualization of latent variables shown in Table 1.

Model for calculating the ESH program beneficiary satisfaction index

The model used in this study was adapted from the European Consumer Satisfaction Index (ECSI) from previous studies made by the National Public Health Institute (^{INSP 2014}) and the Universidad Michoacana de San Nicolás de Hidalgo (^{Ayala, 2011}).

The ECSI indicates a direct influence on beneficiary satisfaction of beneficiary expectation, product quality perception, service quality perception, and social value perception. In addition, the beneficiary expectations have an impact on the social value of the PAL-ESH program. In this model the beneficiary expectations and the social value of the program have a direct influence on beneficiary satisfaction. Finally, beneficiary satisfaction has a direct effect on loyalty to PAL-ESH, as shown in Figure 2.

Source: Author’s own.

Figure 2 Causal model of Beneficiaries’ satisfaction index 

Operational definition

For the measurement and comparison of the latent variables, manifest variables were used (i.e. observed variables) and were included in the questionnaire for assessing beneficiary satisfaction by means of 16 items. Table 2 shows the manifest variables explaining the latent variables.

Beneficiary satisfaction model

We began by specifying the manifest variables and then it was possible to define the complete structural model including both, the latent variables (inner model) and the manifest variables (outer model). As an example, the path diagram for Beneficiary Satisfaction Model in Group 1 is shown in Figure 3. Path diagrams for model groups 2 and 3 feature the exact same structural conformation, varying only about the values obtained in each of the analyzed relationships. The comparative results for all groups are presented in the following tables.

Source: Author’s own.

Figure 3 Path diagram of beneficiary satisfaction showing the causal model for Group 1 

The first item to be analyzed was the predictive capacity of obtained inner as the outer models. The outer model was evaluated using three tests. Firstly, convergent validity was analyzed trough Average Variance Extracted (AVE). The criterion is: latent variables with an AVE at least 0.5 have the ability to explain at least half of the variance of those indicators on average (^{Hair et al., 2019}; ^{Sánchez et al. 2011}). Table 3 shows all the latent variables that meet this criterion, except the latent variable "Beneficiary expectations", that records a lesser value.

The composite reliability (CR) was a second test about the outer model, which describes internal consistency reliability (^{Hair et al., 2010}), i.e. how much the manifest variables are representing the latent construct (unobserved). A commonly accepted threshold value for accepting the reliability hypothesis is between 0.60 and 0.70 (“acceptable in exploratory research,”), meanwhile values between 0.70 and 0.90 range is considered “satisfactory to good (^{Hair et al., 2019}). In this case, Table 3 shows that all CR values are above this threshold, indicating a good model consistency (^{Sánchez et al., 2011}).

Once the outer model evaluation provided evidence of reliability and validity, we proceeded to examine the goodness of fit of the model. In this way, we use the R² statistic which indicates the percentage of the fit has been achieved with the linear model. For this case, the percentage of endogenous variation in S (Beneficiaries’ satisfaction), explained by the linear model, is estimated by the behavior of ξ1, ξ2 and ξ3 (i.e. perception on "Beneficiary expectations", "Service quality" and "Products quality"); thus, R² indicates the predictive behavior of the structural model (^{Henseler and Sarsted, 2013}).

Table 4 shows the R² values of the endogenous variable "Beneficiaries' Satisfaction" for the model according to the group, as well to the global case, which includes the evaluation as a whole with proportional information for each of the groups. Following ^{Hair et al. (2011)} "The judgment of which level of R² is high depends on the specific discipline of the research. While R² results of 0.20 are considered high in disciplines such as consumer behavior, R² of 0.75 would be perceived high in other studies". In the same sense, Chin (1999 in ^{Henseler and Sarstedt, 2013}) describes criteria for R²: Models estimated with PLS whose R² is around 0.67 are considered substantial, about 0.33 moderate and 0.19 weak (^{Henseler and Sarstedt 2013}; ^{Hair et al., 2019}).

As observed in Table 4, the R² values for groups 1 and 3 were moderate, and weak for group 2 and for the global case. Nevertheless, these last values fall within the range that ^{Ringle et al. (2010)} considered moderate. Sometimes, moderate R² values are valid whereas an endogenous latent variable is explained by few - two or three- latent exogenous variables (^{Henseler and Sarstedt, 2013}), as shown in this study.

Drivers of beneficiary satisfaction

Our central hypothesis stated that beneficiary satisfaction drivers and the level of beneficiary satisfaction ought to be differentiated according to the target group involved (i.e., BSI G1 ≠ BSI G2 ≠ BSI G3). For that purpose, standardized regression coefficients (beta) were used expressing the contribution of each exogenous variable for determining beneficiary satisfaction. In other words, the standardized regression coefficients determine the percentage of variance in beneficiary satisfaction which is partially explained by each of the causal variables. The results of this analysis are presented in Table 5 for each contrast group and for all groups (BSI Global).

Table 7 presents the correlations of each manifest variable (i.e. variables observable in the field) determining beneficiary satisfaction. Each group is discussed in the following sections.

Beneficiary satisfaction index

Once we identified the main latent variables associated with BSI and the manifest variables (causal) that largely determine it, we could analyze the specific satisfaction levels we found in this study (Table 6).

By that means we obtained an overall satisfaction rate of 84.8%. Comparing the three involved groups, Group 2 displayed a higher satisfaction rate of 87.6%, while the lower satisfaction rate of 82.4% corresponded to Group 3 (i.e. urban beneficiaries in high marginalization).

Drivers of beneficiary satisfaction in group 1 (rural, high marginalization)

The results of Group 1 including high marginalized rural localities showed that the main driver for beneficiary satisfaction was the latent variable of quality service, accounting for 45.8% of the explained variance, followed by social value of the PAL-ESH program (26.6%), and beneficiary expectations (22.9%; Table 5). In this group the main manifest variables correlating with each of the above-mentioned latent variables were perception about the helpfulness of store employees (r = 0.90), and perception about the promptness of attention by store employees (0.88). In the case of beneficiary expectations, the manifest variable perception about price dominated this correlation (r = 0.77), as shown in Table 7.

Drivers of beneficiary satisfaction in group 2 (rural localities with very high marginalization)

The analysis of Group 2, integrated by beneficiaries from very high marginalized rural areas, showed a greater contribution to BSI based on the perceived quality of products (37.7%; Table 5), which is mainly associated with the perception of overall price-quality relationship (including the 19 products in the PAL-ESH basket) (0.84). The second factor driving BSI in Group 2 was the perceived social value of the PAL-ESH program (32.5%), which correlated (r = 0.82) to the appreciation that the program allowed beneficiaries to buy cheaper products compared to other local stores (Table 7). Finally, BSI was incremented by quality of service (15.7%) and beneficiary expectations of the PAL-ESH program (13.9%).

Drivers of beneficiary satisfaction in group 3 (urban localities with high marginalization)

Table 5 shows that in Group 3, integrated by high marginalized urban localities with, the perception of the PAL-ESH program´s social value was the main factor, accounting for 46.9% of the explained variance in BSI, followed by perception of quality service (25.0%), and beneficiary expectations (17.7%). Among these latent variables, the manifest variables with the highest correlations with social value were the low prices offered (0.82) and the quality of offered products (0.77). Perception of quality service showed high correlations with the kindness of Diconsa employees (0.88) and, not statistically differing, with perception of promptness of attention by Diconsa´s employees (0.87) (Table 7).

Drivers of beneficiary satisfaction for all groups (BSI global)

Our results for the BSI value in the combined groups (Table 5) indicate that the variable that best explains BSI was the perception of social value of the PAL-ESH program (i.e., products´ quality, low prices, general satisfaction), a variable which accounted for 48.16% of the explained variance and was highly correlated (0.81) with the perception of low prices offered to the locality (Table 7). The second main determinant of BSI for the combined groups was the perception of quality service accounting for 30.57% of the explained variance. The two manifest variables included in the latter variable showed high correlation values about 0.87 for the perception of the helpfulness of Diconsa´s employees and 0.86 for the perception of promptness of attention by Diconsa´s employees (Table 7).