Construction and Validation of an Instrument to Measure Perceptions and Attitudes towards Genetically Modified Organisms in the Mexican Urban Population

Introduction: this paper presents the development and validation of an instrument to measure the perceptions and attitudes about the production and consumption of the Mexican urban consumers towards genetically modified organisms (GMOs). Method: The proposed questionnaire contains 63 questions that encompassed 11 latent factors that was applied to 14,720 people of the Mexican urban population aged between 18 to 65 years. This measuring questionnaire was validated using confirmatory factor analysis (CFA). Results: we found that it is acceptable the proposed instrument for measuring perceptions and attitudes towards genetically modified organisms (PAGMOs) for the urban Mexican population. Discussion and Conclusion: The same instrument (construct) it is suitable for each of the 8 regions studied since with the multi-group CFA performed we found evidence that this is valid for each of the regions under study. Also, the analysis of the questions of the proposed instrument revealed that participants have a low general knowledge, a high distrust towards GMOs, want labeling of genetically modified (GM) products and not perceive their social values and positive health effects beyond increasing agricultural productivity.


Introduction
Currently in the world several Genetic Engineering techniques are being developed and applied in plants and animals to increase the production of food that is made with Genetically Modified Organisms (GMOs) in order to increase production and extend safety levels and reduce malnutrition in the population in poor and developing countries (Kimani & Zennah, 2019). With the application of the genetic engineering methods in animals such as goats, cows, rabbits, and birds, the production costs of recombinant proteins used in medicine have been reduced through the use of milk, egg whites, blood and other body fluids (Woodfint, Hamlin, & Lee, 2018). Consumer behavior has a higher level of distrust, in relation to the production and processing of foods with genetic engineering, considering the large number of diseases that are currently known. The buyer is afraid to consume food that can affect his family in the medium and long terms, as well as, the effects on the environment (Gatto & Smoller, 2018). In the period from 1996 to 2019, crops with genetic engineering have been adopted by the commercial industry, by small and large farmers in industrialized and developing countries, however, this type of products and food are immersed in a great social controversy (Kamle M. , Kumar, Patra, & Bajpai, 2017). In the last 23 years, more than 2.15 million hectares of biotech crops have been commercially cultivated in the world. Cultivating primarily: soybean, corn, cotton and canola, for a growing world population, it is estimated that by 2050 the world population will be 9,800 million. Therefore, it is required that a greater amount of nutritious and safe food be produced for good human nutrition (The International Service for the Acquisition of Agri-biotech Applications (ISAAA), 2017). Since 1996, when the cultivation of GMOs began, a great deal of controversy has been generated among the population around the world, with a large number of reports available in the media such as books, magazines, television, radio, newspapers and social networks. Some present the benefits of GMOs and others argue that have negative effects on the family and on future generations (Bardin, Perrissol, Facca, & Smeding, 2017). In May 2016, the National Academies of Science, Engineering and Medicine (NASEM) published their report, "Genetic Engineering Crops". Participants in this meeting analyzed the risks, health benefits of humans and the effects on the environment of the GMOs (Landrum, Hallman, & Jamieson, 2019). The conclusions of this meeting were as follows: there was not enough evidence of risks to human health and the environment between conventional crops and those of Genetic Engineering, these conclusions also were published in The New York Times (Pollack, 2016). However, a group of more than 300 independent scientists and academics disagrees with the conclusions in which some groups of researchers have stated doubts about the safety of crops and products made with genetic engineering, they mention that the conclusions lack scientific foundations, since no independent arbitrators have been reviewed (Hilbeck,et al., 2015). Knowledge of perceptions and attitudes towards biotechnology products are lacking in México, althougth various studies have been conducted in a large number of countries. Most surveys have an academic intention and the conclusions are only valid for specific regions (Pino, Amatulli, De Angelis, & Peluso, 2016). Some important questionnaires to measure attitudes towards the production and consumption of GMOs have been constructed after a systematic literature review for specific regions as those proposed by Costa-Font andGil (2009), Herodotou et al. (2012), Sorgo, Ambrozic-Dolinsek, Usak and Özel (2011) and Sorgo and Ambrozic-Dolinsek (2010). It is important to note, that those instruments were mostly designed for specific populations with a high skepticism towards transgenic and a moderate knowledge of biotechnology issues, which is characteristic of the European Community (Eurobarometer, 2010). In addition, there are also questionnaires that have been applied in contexts where populations have sympathy towards GMOs, such as Ma (2015). However, due to the peculiar characteristics of Mexican people, we propose to construct and validate a new questionnaire in order to measure their perceptions and attitudes toward GMOs. In Mexico, corn represents a biological and cultural heritage, for this reason GMOs introduction to México have generated a great controversy in the whole society causing mobilization of broad sectors of society (rural and urban) in defense of native seeds. Such protests have arisen not only because of the threat that agricultural biotechnology represents for native corn, but also because there is a need of more fair and sustainable options. These experiences seek to strengthen efforts to protect native maize. On the other hand, there are several opinions, which highlights the lack of knowledge about the advantages and disadvantages in terms of genetic, economic, social, cultural, public and ecological health (Reséndiz-Ramírez, López-Santillán, Briones-Encinia, Mendoza-Castillo, & Varella-Fuentes, 2014).Therefore, it is important to construct and validate an instrument (1) to be able to understand the perceptions and attitudes of Mexican urban population about the production and consumption of GMOs, and (2) to study and assist decision-making about the introduction of transgenic in Mexico through a comprehensive analysis. Since there is no records about a questionnaire for measuring such perceptions and attitudes in Mexico, according to Sánchez and Echeverri (2004), in order to develop measurement scales, which are helpful in cases of complex measurement and with diffuse features, a process to validate an original questionnaire, which was based on existing instrument used in other countries, was implemented using confirmatory factor analysis (CFA), that is, a statistical technique used to assess measurement models, which represent hypotheses about relationships between indicators and factors.

Method
In this study our target population was the Mexican urban population aged between 18 to 65 years and the sample size at each region was obtained (Cochran, 1990;Olaiz-Fernández et al., 2006) ) ℎ) This formula was used to estimate with a 90% confidence proportions close to 13% with a maximum relative error of about 17%. The design effect (DEFF), expected response rate (RR), average number of people per household (ℎ) assumed were 2.84, 75%, and 1.29 respectively. denotes the households sample size, the proportion to estimate, / the quantile of a normal distribution, denotes the allowed relative error. With this formula we got a sample size of 1840 households per region. The size of the sample by region (1,840 households) was distributed proportional to the basic geostatistical area units (AGEBs) that make up the region.
The selection of primary sampling units (PSU) is made up of the AGEBs listed in the 2010 Mexican census and the AGEBs of the 2010 census that are not listed in the 2010 census. Therefore, the selection of sample units was in several stages, at the first stage, the AGEBs within each location was selected, then blocks within each AGEB, then homes within each block and finally individuals within the households. The application of the questionnaire was done from May 21 to July 21, 2015 using questionnaire given in Table 2. Table 1 gives the distribution of the 14,720 surveys taken at national level. An adult aged between 18 to 65 years was selected at each household. Also, in each block was selected a household from each cardinal point. We found that 50.96% and 49.03% of the surveyed people were men and women, respectively. The largest participation was of people between 30 and 44 years (35.37%), and the lowest of people between 45 and 54 (18.03%).

Instrument
The questionnaire was built on studies conducted in different countries, and contain factors that measure perceptions and attitudes towards the production and consumption of GMOs. It is composed of 11 such factors: Trust, Knowledge, Perceived Risks, Benefits, Attitude towards Technology, Religion, Attitude toward Gene Technology, Labelling, Attitude towards Buying, Societal Values, and Promotion.
First, we made a list of 84 items related to the above mentioned 11 dimensions (latent factors); measured with a Likert-type scale of 5-points (1 = strongly disagree, 2 = disagree, 3 = neither agree nor disagree, 4 = agree, 5 = strongly agree). Then we conducted a pilot study with this questionnaire with a group of 90 people over 18 years of age from Colima, México. The study showed that respondents found it difficult to answer the survey items. Also, we found a high percentage of respondents who knew nothing about the use of GMOs, and most of them opted for answering option 3 (neither agree nor disagree). Also, we noted that the questionnaire was too long as it took 30 minutes on average to answer it, and respondents were eager to complete the questionnaire. Based on the results of the pilot survey, the problematic items were rewritten to improve the style and content. In this way, the questionnaire was reduced from 84 to 60 questions and the response options were simplified to a dichotomous scale. We also added questions aimed at gathering demographic information (occupation, educational level, gender and age) and a section to record aspects of geo-referencing. This instrument was tested with 1000 people in Colima, Mexico, and we only found small details that needed to be corrected. Finally, we end up with the questionnaire given on  Attitude towards Gene AGT1 Do you think that the production of transgenic products to increase the amount of food among ijbm.ccsenet.org International Journal of Business and Management Vol. 15, No. 5; Technology (AGT) Mexicans is a contribution?

AGT2
Do you consider morally acceptable the production of transgenic products for consumption by the Mexicans?
AGT3 Do you agree with the production and consumption of transgenic products for the Mexican population?
AGT4 Do you think that transgenic products have higher nutritional content than organic products?

AGT5
Do you think the consumption of transgenic products will increase life expectancy of Mexican society?
AGT6 Do you agree in promoting transgenic products for family consumption?
Does your religion is in favor of the development of transgenic products for human consumption?

REL2
Does your religion prohibit the consumption of transgenic products?

REL3
Does your religion consider for moral reasons that you should not eat genetically modified products?

REL4
Does your religion consider morally incorrect the processing of transgenic products?

REL5
Is it right for your religion that scientists genetically modify plants and animals for human consumption?
Do you have the habit of reading the labels of the products that your family consumes in the diet before buying?

LA2
Do you think that transgenic products must display on its label if they contain genetically modified ingredients?
LA3 Do you think that in advertising of genetically modified products is vital that the consumer is informed about the content of the product in question?

AB1
Would you buy transgenic products if they contain less fat than conventional products?

AB2
Would you buy transgenic products if they were cheaper than organic products?

AB3
Would you buy transgenic products if these were grown under similar environmental conditions to the organic production?

AB4
Would you buy transgenic products if the price is equal to the organic products? The hypothetical relationship between indicators and the factors given in Table 2 is shown in Figure 1. The first coefficient (load) of each factor was set to 1 to ensure an over-identified model (with more observations than parameters to estimate). This model also allows correlated factors. Fifty five correlations or covariance´s were established because it has eleven factors. This model is over-identified (more observations than parameters to estimate) since the number of observations is larger than the number of parameters. Sixty three indicators gives 63 ((63 + 1)) / 2 = 2016 number of observations. There were 175 parameters (49 loads factors, 11 variances of the facto freedom t each of th categoric in Figure   Figur Source: ( for testing invariance and next we explain these models: Configural invariance (Model 1). This is tested by constraining the factorial structure to be the same across groups and it is satisfied if the basic model structure is invariant across groups, indicating that participants from different groups conceptualize the constructs in the same way. This model is the first step to establish measurement invariance. Also, by running individual CFAs in each group configural invariance can be tested, but usually this approach is not used since it is still required to run this step in MGCFA, since it serves as the baseline model for subsequent tests.
Metric invariance (Model 2). This model is tested by constraining all factor loadings to be the same across groups. If this is satisfied, the ratings can be compared across groups and observed item differences will indicate group differences in the underlying latent construct. This model tests if the strengths of the relations between specific scale items and their respective underlying construct are the same across groups. At least partial metric invariance must be established before continuing in the sequence of tests (Vandenberg and Lance, 2000).
Scalar invariance (Model 3). To compare (latent) means it is required the scalar or intercept invariance. This model is evaluated by constraining the intercepts of items to be the same across groups. Establishing scalar invariance means that observed scores are related to the latent scores; that is, that regardless of their group membership individuals who have the same score on the latent construct would obtain the same score on the observed variable. To compare scores across groups this is the last model necessary. The additional tests are not mandatory since are not meaningful in all contexts.
Error variance invariance (Model 4). Here all error variances are constrained to be equal across groups to be able to test if the same level of measurement error is present for each item between groups.
Factor variance invariance (Model 5). This model is evaluated by constraining all factor variances to be the same across groups. This invariance means that the range of scores on a latent factor do not vary across groups.
Factor covariance invariance (Model 6). This model is evaluated by constraining all factor covariance's to be the same across groups. The stability of the latent factor relationships across groups is assessed in this model, which implies that all latent variables have the same relationship in all groups.
Factor mean invariance (Model 7). This model is evaluated by constraining the means to be the same across groups, which indicates that groups differ on the underlying construct(s).
The first 4 models evaluated measurement invariance because these models test relationships between measured variables and latent constructs. While models 5, 6 and 7 test structural invariance and this models concern only the latent variables. The factor mean invariance can be tested (see Model 7 below) immediately after testing Model 3 (scalar invariance or intercept) since these models are not necessarily nested (Kline 2011). The CFA analysis was conducted in the R statistical package (R Core Team, 2015) with the library lavaan (Rosseel, 2012).

Goodness-of-Fit Indices
CFA model fit can be tested through chi-square analysis. A non-signicant chi-square is indication of good fit, that is, a non-significant chi-square value indicate a good fit. However, chi-square test is more sensitive with a larger sample size. Some authors suggest that the weighted root mean square residual (WRMR) index can be used to fit categorical data (WRMR<1.0 indicates a good fit) instead of the Standardized Root Mean Square Residual (SRMR) (Yu and Muthén, 2002). Hu and Bentler (1999) recommended using several fit indices. For this reason, in this paper we used three indices. We used the Tucker-Lewis index (TLI), the root mean square error of approximation (RMSEA) and the comparative fit index (CFI). The CFI and TLI evaluates the adequacy of the specified model in relation to the baseline model (Wade et al. 1996). Better fits are indicated by higher values (to a maximum of 1). Therefore, acceptable adjustments are considered with values ≥ 0.96 for CFI and TLI (Yu, 2002). Compared to a perfect or saturated model the RMSEA estimates the lack of fit in a model (Steiger and Lind, 1980). A value below 0.06 for RMSEA provides an acceptable model fit (Browne and Cudeck, 1993).
Three specific incremental indices, based on the differences in the CFI, for testing measurement invariance has been suggested (Cheung and Rensvold, 2002) these are: (a) the Steiger's (1989) index (b) the gamma hat (GH) index; and the (b) McDonald's (1989) non-centrality index (NCI) that is obtained when comparing nested models.
The more restrictive model should be rejected If in the sequence of the invariance tests, two nested models show a decrease in the value of CFI, GH and NCI greater than or equal to .01, .01, and .02 in magnitude, respectively (Cheung and Rensvold, 2002). Here, we used the CFI, TLI, RMSEA and only the differences in the CFI since the other two indices are not given as output in the lavaan package we used for fitting our models.

Logistic Regression
The logistic regression was used to test the factor mean invariance. For this reason, in each factor each item was considered as binary dependend variable (where 0=no and 1=yes) and the 8 regions under study were considered the independent variable. This associative analysis was key to see differences between regions on the perceptions and attitudes toward GMOs. We used the logistic regression since our dependent response variable is binary (see Stroup, 2012 for details of logistic regression).

Model Fit of the Single Group (at National Level) with CFA
To validate the proposed questionnaire given in Figure 1 and Table 2, we performed a CFA for binary responses. The overall fit was satisfactory, because the CFI = 0.989 (> 0.96), TLI = 0.988 (> 0.96) and RMSEA = 0.055 (<0.06). Table 3 shows the factor loadings and thresholds for the model using CFA. All factor loadings were statistically significant (p-value <0.05) and only one of the thresholds PN1 was not significant (p-value = 0.574).   Table 4 shows the estimated factor loadings and thresholds for multi-group model by regions. In this case all factor loadings were statistically significant and only one of the thresholds turned out to be not significant (

Configural Invariance
Due to the good fit of the 11-factor structure for each group earlier, one could expect that configural invariance would be supported and the fit indexes confirmed this. Table 5 shows that Model 1 provided a good fit to the data (CFI, TLI and RMSEA yielded satisfactory values: 0.991, 0.99 and 0.055, respectively) indicating that there is evidence the factorial structure of the construct is equal across regions. All this also was supported by ΔCFI criteria which support configural invariance when the ΔCFI<0.01.

Invariance of Factorial Loadings, Thresholds and Intercept
Since we found evidence of configural invariance of the factors among regions, we continued to seek evidence to identify the invariance of factor loadings and thresholds among regions. According with the fit indices CFI, TLI and RMSEA we have a good fit (0.990, 0.990 and 0.056 respectively). Also, in this case the ΔCFI = 0.001 (<0.01) index provided evidence in favor loadings of the factor and thresholds (Table 5).

Invariance of Co-Variance Error
Invariance of co-variance error between regions also is reasonable since we got a good fit for CFI, TLI and RMSEA (0.988, 0.988 and 0.061respectively) (Table 5). Also, according with the ΔCFI = 0.002 (<0.01) index gives evidence in favor of invariance of co-variance error, implying Covariance's) of errors are invariant between regions (Table 5).

Latent Variance and Invariance of Covariance's
Since we have shown that it is satisfied the assumption of invariance of the (co)variance´s between regions, here we tested the latent variance and invariance of covariance's. Also, this invariance assumption is reasonable since On the other extreme, we found that high perceived risks are associated with low perceived benefits, a negative attitude towards gene technology and a negative attitude towards the promotion.
Regarding items of each factor measured in different regions, a similar pattern among them were observed. In the case of knowledge factor, a general lack of such factor was observed, not only in the GMOs definition, but also in their applications, and its laws and regulations. With respect to the trust factor, its result was low (< 50%), however Mexican people have greater trust for universities and scientists than for pharmaceutical companies working with transgenic products. As for benefits and perceived risks, Mexicans perceive high risk in the use of transgenic products. Most individuals in all regions consider that the use of GMOs can bring consequences which range from several diseases and environmental damage that can produce adverse effects on future generations. Nevertheless, excepting gene technology, test-takers have a positive attitude towards science and technology as they consider both are vital for social development.
In the case of the religion factor, surveyed people are not agree with GMOs production, in fact, they do consider immoral plants and animals modifications. With regard to the labelling, it was observed that only 63% of test-takers read labels on the products they consume. Additionally, they show a positive attitude towards the need to label genetically modified products. Meanwhile, surveyed people showed preference to buy GMOs product only if such products are cheaper, with low fat content, and have been grown in similar environmental conditions as organic products. If prices are similar, participants prefer to buy organic products such as maize and beans. However, they are not agree, about the use of technology to produce transgenic produces for human consumption, even that they recognize the GMOs potential, but they disagree with their application.
Finally, with regard to promotion, a slightly positive attitude among surveyed people was observed, toward that the Mexican government provide funding for research with transgenic products to generate new drugs. However, participants showed their disagreement about that the government could fund private companies to produce transgenic products in Mexico or to import these products for consumption. Also, it is important to point out that the proposed instrument is very important to measure the perceptions and attitudes of the Mexican population and we hope can be used for future studies since can measure 11 latent factors really important to have a clear picture of these perceptions and attitudes.