Consumption of low pesticides food: implications for producers and policymakers. Results from a multi-attribute analysis

2.1 Data

Data was collected using a web-based survey tool via an online questionnaire (Qualtrics, Provo, UT, USA). The structure of the questionnaire follows that prepared by a previous research of a Horizon project (H2020, EUCLID, 2015–2018) in which experts belonging to naturalistic and economic disciplines took part. They were asked to review the questionnaire and improve its reliability and validity. The questionnaire consists of 35 questions, each of which represents an indicator. The study was conducted over an 8-week period (1 March 2021 to 30 April 2021). Given the specificity of the topic, a virtual snowball sampling technique was used (Goodman, 1961). Using the authors' networks, the interviewees were asked to forward the survey link to family, friends, colleagues who they believe involved in food purchase choices. More than 3,000 consumers were involved in the survey, of which 1,103 were completed in all parts. The completion rate is approximately 37%. The first note concerns the greater participation of women and the higher educational qualifications of the interviewees, as if to indicate greater sensitivity to the issues of sustainable food consumption of women and people with higher educational qualifications.

The 35 questions (attributes) of the questionnaire were partitioned into three blocks, B_i (i = 1,2,3):

1. Basic characteristics (B₁): These questions were used to collect data on age, gender, education level, income range, place of residence, nationality, number of people in the family and profession (9 questions).

2. Eating and grocery shopping habits (B₂): These questions were used to collect data on participants' diets, what part of their diets consisted of horticultural products, whether the decision maker was responsible for food choices in her/his household, where food was regularly purchased and what factors determined the choice (7 questions).

3. Perception of and attitudes toward IPM products (B₃): These questions were used to assess the familiarity of participants with the IPM method and their knowledge of differences between IPM and organic agriculture, and to determine whether they could distinguish the IPM method from others commonly discussed in the media, such as genetically modified organisms (GMOs) and post-harvest measures (19 questions).

2.2 Method: DRSA

To provide suggestions that can support policymakers, farmers or companies, it is necessary to produce informative and easy-to-understand results, in which evaluation paths are transparent and based on indicators that can be updated; thus, the model should be dynamic and, for this reason, suitable for use as a continuous evaluation tool.

The method we propose consists of the following steps:

1. Definition of the information system IS=(U,A), where U is the (Universe) set of all 1,103 survey participants and A=⋃i=13Bi is the set of all information gathered in the questionnaire.

2. The definition of the disjoint sets of the condition and decision attributes C and D, respectively, such that the set of all information A is partitioned as A=C∪D. The choice of considering a question as a condition attribute (i.e. an attribute characterising the sample) or as a decision attribute (i.e. information marking the category of the sample) can change during the analysis. This is due to two reasons: (1) because our aim was to develop a general framework for the analysis of consumer food choices, it was necessary to consider different interrelations between the collected information. (2) Multiple answers and nested questions were presented in the questionnaire.

3. Definition of the coefficient matrix (different choices of the partitioning condition C and decision D attributes give origin to different coefficient matrices) with reference to the particular partition of A under consideration: f(xi,aj)∈Vaj denotes the answer given by i-consumer (i=1,…,1,103) with reference to j-question. The set Vaj is called the domain set and contains all the values (i.e. given answers) assumed by attribute (i.e. question) aj; in some cases, a fixed-point scale answer was attached to the question.

4. Extraction of DRSA decision rules: in the form of if … then … sentences, the conditions that characterise the rule and the resulting decision class assignment can be explicitly displayed.

5. Sensitivity analysis and suggestions: by explicitly representing the dependence between condition and decision criteria (conditions → decision), decision rules provide exhaustive and easy descriptions of patterns in the data and, therefore, are the best means to analyse the results and to communicate them to the operators.

In choosing the framework to model our decision support system, the possibility of considering both numerical and categorical data, the non-use of statistical assumptions on the distribution of data and the absence of the need for structures that collect data (e.g. functions or equations) were all factors that guided our analysis. We were also interested in a simple method for the description of schemes that were exhibited by the data, using a useful and flexible tool that can be easily updated and is capable of capturing the fundamental characteristics of the information system in a multifaceted representation (see case studies conducted by Celotto et al., 2015; Zolin et al., 2017; Ferretti et al., 2020; Roma et al., 2020).

The key idea of a universe U partitioned into equivalence classes is not sufficient when objects are described by attributes with domains that are preference orders, given that inconsistencies can be generated by violations of the dominance principle (Greco et al., 2001, 2002). However, in the analysis of multi-criteria decisions, it is possible that some of the attribute domains are ordered; therefore, it is necessary to explicitly consider preference relations in attribute domains.

In DRSA, the main assumption is that each domain Vaj is completely preordered by an outranking relation ≿ai with the following meaning: x≿ajy when x is at least as good as y with respect to criterion aj. If each domain Vaj is real valued, then it is x≿ajy if and only if f(x,aj) ≥f(y,aj).

In the case of decision criterion d, each element in U is assigned one class Clt (t∈N) such that the classes are preference-ordered, that is, when r>s, each element in Clr is preferred to each element in Cls. Suppose that m denotes the finite number of decision classes. Preferences can be either strict or weak. More precisely, if r>s and it is x ϵ Clr and y ϵ Cls, then ′x is at least as good as y’ and not ′y is at least as good as x’ with respect to the decision criterion. Equivalently, given the decision criterion d, under the same assumptions, it is f(x,d) ≥f(y,d) while is not f(y,d) ≥f(x,d).

Related to the previous assumption are the definitions of the upward and downward unions of classes Cls: Clt≥=⋃s≥tCls; Clt≤=⋃s≤tCls. Therefore, x ϵ Clt≥ implies that x belongs to at least class Clt, while x ϵ Clt≤ means that x belongs to at most class Clt.

With reference to condition criteria in C, it is possible to define a partial preordering DP (i.e. reflexive and transitive) for each condition criterion in P⊆C, with the following meaning: x DPy if x≿aiy for each criterion ai∈P. In this manner, each element x in U is related to two sets: the P-dominating set and the P-dominated set, respectively

If x dominates y on all condition criteria in P⊆C, it also dominates y on the decision (i.e. element x should be assigned to at least as good a decision class as y). In our study, decision makers satisfying the dominance principle were consistent, while those violating the dominance principle were classified as inconsistent.

The P-dominating and P-dominated sets represent the basis of knowledge, in fact, knowledge is approximated by the upward and downward unions of decision classes; the P-lower approximation of Clt≥ with respect to P⊆C is P (Clt≥)={x∈U:DP+(x)⊆Clt≥}, while the P-upper approximation of Clt≥ with respect to P⊆C is P¯ (Clt≥)={x∈U:DP−(x)∩Clt≥≠∅}.

Analogous are the definitions for the P-lower and P-upper approximations of Clt≤ with respect to P⊆C:

Accordingly, the lower approximations are composed of elements belonging to the upward and downward unions of classes, whereas the upper approximations contain elements that can belong to the upward and downward unions of classes.

Finally, the upward and downward unions of classes are related to the upward and downward unions of the decision classes by the set inclusion relations:

The related decision rules, that is, logical statements given by a composed condition component (if …) and a decision component (then …), in DRSA are formalised as D≤–decision rules, D≥–decision rules and D≥≤–decision rules, given that the information system can be described by decision rules based on dominance relations and associated approximations, and the related rules can be certain, possible or approximate. Certain rules follow from lower approximations, possible rules are linked to upper approximations and the approximate rules refer to the boundary regions.

Step 4 was performed using the VC-DomLEM algorithm implemented with the jMAF software (Błaszczyński et al., 2013) developed by the Laboratory of Intelligent Decision Support Systems at the Poznan University of Technology (http://wwwidss.cs.put.poznan.pl).

5.1 Segment 1 green consumer

The decision attributes d1 (Views IPM products as healthier than conventional), d2 (Environmentally conscious) and d8 (Purchased organic products in the past six months) describe the green consumer.

According to the decision attribute d1, the consumer is most likely to consider IPM products to be healthier than conventional food products if he or she is responsible for purchasing food, is indifferent to the price and prefers to buy cereals obtained with IPM methods. The second rule states that IPM products are considered healthier than conventional ones if the environment is the most important factor in purchasing decisions and if the preference is for IPM grapes. Third, the situation described by the decision attribute occurs if the purchasing decisions are not based on price, the consumer is familiar with IPM products and if he or she prefers IPM grapes.

The decision rules selected and related to the decision attribute “Environmentally conscious” share one conditional attribute: the price is not among the most important elements. The strongest decision rule describes a consumer who does not consider the price important, does not consider the appearance of the product and packaging, considers organic products to be high-quality and is willing to pay a higher price for organic grapes. The subsequent decision rules add more information. If the price and appearance are not taken into account, the product does not contain GMOs, and the consumer has correct information about IPM methods, then the consumer is more likely to be environmentally conscious. Furthermore, if the price is not decisive and the consumer resides in a Southern European country and has a monthly income equal to or greater than 5 thousand euros, then he or she is more likely to be environmentally conscious.

Moreover, consumers who have purchased organic products within the past six months are decision-makers who think that environmentally friendly production methods play an important role in choosing food items; moreover, they live alone. Similarly, a worker who purchased IPM products in the past six months was likely to have purchased organic products in the same period.

Policy implications: This is an informed market segment with low price elasticity and relatively high income, where extrinsic characteristics are of marginal importance. The products that this market segment considers important if obtained with low environmental impact processes are grapes and wine. Moreover, consumers of organic products are prevalent. Consequently, producers should equip themselves with tools (labels) capable of allowing these consumers to recognise production processes with a lower environmental impact (European Commission, 2007, 2020). From the perspective of public decision makers, dissemination campaigns using digital tools (Demartini et al., 2018) on the health and environmental benefits of consuming products with lower pesticide content could further strengthen this segment.

5.2 Segment 2 IPM informed and active consumer

The IPM informed and active consumer is outlined by the decision attributes d4 (Familiar with IPM products) and d6 (Purchased IPM products in the past six months).

Consumers are likely to be familiar with IPM products if they are aware that IPM methods require more pesticides than organic ones, if they have purchased them in the past six months and curiously are not available for training and/or additional information. Furthermore, he or she is familiar with IPM products if residing in a Southern European country, has purchased them in the past six months and is willing to pay double the price of organic leafy vegetables.

The rules' structure is very simple; only one condition attribute appears (unique case in all considered decisions): the decision maker who is not influenced by food labelling, has no restrictions in terms of the type of food production method, or thinks that IPM foods are healthier is likely to have purchased IPM food in the past six months.

Policy implications: This consumer segment has strong similarities to Segment 1 (green consumer). The rules describe an informed consumer who knows how to distinguish between different production processes and is indifferent to price and extrinsic characteristics. The strategies that can be adopted in this case are similar to those of segment 1 (labelling and digitalisation) and are attributable to tools for consolidating loyalty by both policymakers and producers, focusing on the intrinsic characteristics of IPM products.

5.3 Segment 3 potential IPM/low-pesticide or organic food consumer

The potential consumers of IPM or organic food are described by the decision attributes: d5 (Unfamiliar with IPM products), d7 (Did not purchase IPM products in the past six months), d9 (Did not purchase organic products in the past six months) and d3 (Price conscious).

It is the most important segment of our sample and describes a consumer who is misinformed, who has not purchased IPM or organic foods in the last six months and who is influenced, in purchasing decisions, by prices.

In the case of unfamiliarity with IPM, the decision rule with the highest support shows an influence of income (up to 2000 euros per month). The place of origin is not important, and he or she has not bought IPM products in the past six months because he or she is unaware of them and does not know that they are safer in terms of health. The other rules add further information on condition attributes, such as geographic area (Northern Europe), gender and other food aspects in purchasing.

The situation of not purchasing IPM products in the past six months is characterised by very high support rules (between 354 and 418); the decision-maker is not totally dedicated to work or studying, or he or she pays attention to the ways in which food is presented, or he or she believes that high-quality food undergoes a process that involves the least quantity of pesticides.

According to the decision rules, the consumer did not buy organic products in the past six months because of his or her lack of knowledge. In addition, this consumer is available to pay more for organic wine, ignores that IPM is safe to eat or is convinced that high quality means food purchased directly from the producer/farmer.

An analysis of the rules related to price consciousness reveals a consumer for whom price plays a key role: he or she did not buy organic food because it was too expensive.

Moreover, in the first decision rule, he or she declared a willingness to buy imported IPM products if cheaper, and if there is a discount. The second decision rule shows that consumers are more inclined to worry about price if they are omnivorous and if they prefer food with fewer pesticides. The last decision rule confirms the previous one, moreover he or she is not interested in IPM training information initiatives. The consumer described by the rules has high price elasticity and very moderate attention to environmental issues.

Policy implications: This is the most important segment and is numerically predominant. The decision-making rules selected highlight the lack of information of these consumers, to whom a specific campaign should be addressed to disseminate information on the environmental and health benefits derived from the consumption of food with little or no use of pesticides by policy makers. Producers of food with low environmental impacts should contribute to strengthening the knowledge of various production processes. In this context, multinational and national retail companies can assume a crucial role, which, when in direct contact with consumers, can influence their choices.

Given the importance of the public food service sector, green public procurement schemes (De Almeida Ferreira Neto and De Oliveira Gama Caldas, 2018; European Commission, 2016, 2020) in public tenders and catering services represent an indispensable tool to increase the knowledge and consumption of food with lower pesticide content.

This segment has a high elasticity with respect to price, which is decisive in consumption decisions. To influence the choices of this segment of consumers, producers should propose loyalty campaigns based on low prices, combined with tools to raise awareness of environmental protection. For this consumer segment, the extrinsic characteristics are of marginal importance.