2.2.1 Contextual factors influencing growing food at CGs.

Educational and technical assistance for CG projects are often cited as facilitators of successful community gardening [38–40]. For instance, examining the resources needed for successful urban agricultural projects in New York City, Cohen and Reynolds [38] found that technical assistance to practitioners is a crucial factor. Specific forms of technical assistance can include provision of gardening advice, demonstration gardens, and guidance by volunteers and staff to new gardeners to ensure a continued participation in CGs [38, 41, 42]. While previous studies indicated the importance of addressing such lack of assistance based on qualitative interviews, they did not test how a specific type of assistance determines consumers’ preferences for CGs quantitatively. In this study, we provide a quantitative, empirical analysis of whether and how provision of guidance to garden influences consumers’ preferences for CGs.

Costs to use gardens are consistently associated with the success of CG projects. In a survey conducted to assess shared challenges among 445 CGs in the U.S. and Canada, Drake and Lawson [43] found that lack of funding is one of the key challenges, and Cohen and Reynolds [38] reported similar challenges. Literature review-based studies also identified costs to use CGs as one of the impediments to successful operation [10, 44]. In addition, these studies show that poor access to appropriate tools (soil, compost, seeds, etc.) places a limitation on participants’ ability to grow food continually in CGs [10, 44]. Again, the results are based on qualitative studies, which limits a quantitative determination whether and how costs to use CGs, and access to equipment influence participation. Our study directly addresses these issues by investigating them quantitatively.

Another barrier to a successful development of CGs is lack of stakeholder engagement [10]. Drake and Lawson [43] discussed that the time and commitment required to pursue gardening activities may deter some people from participation. On the other hand, lack of knowledge and awareness can also be an obstacle [39]. One strategy to increase community engagement is promoting networking and learning across social communities [10]. This may involve organization of social events to increase integration and knowledge exchange [39]. Nevertheless, there is a lack of deeper understanding of how social events influence decisions to grow foods. In this study, we contribute to the literature by empirically analyzing how social events affect this decision.

Finally, distance to a garden site is listed as another barrier to effective implementation of CGs [40]. This barrier is associated with the transportation needs to get to CGs, especially for those that are not within walking and bicycling distance [45]. While the previous literature observed the negative effects of distance on CG participation, empirical analysis regarding the effects of transportation modes is non-existent. In this study, we consider three modes of transportation (on foot, by bicycle, and by car) to shed light on how they influence consumers’ decisions regarding small-scale urban agriculture.

2.2.2 Intrapsychic factors influencing growing food at CGs.

The role of intrapsychic factors, such as, attitudes and subjective knowledge, in explaining behavior in various contexts has been recognized in the social psychology and marketing literature. In this section, we review the relevant literature to inform a theoretical framework demonstrating the roles of attitudes, subjective knowledge and reasons in driving consumer behavior.

Attitudes are one of the most studied intrapsychic factors in the literature. Ajzen [46, pp. 188] defined attitudes as “the degree to which a person has a favorable or unfavorable evaluation or appraisal of the behavior in question”, stating that attitudes are antecedents of behavior in that they determine the intention to the intended behavior. Related to CGs, environmental psychologists identified attitudes as antecedents of pro-environmental behavior [37]. Other researchers showed the same for energy conservation behaviors [36], and healthy eating behaviors [47]. Most of the existing studies contend that attitudes are a key impetus for behavior changing interventions. Hence, the success of gardening interventions can hinge on consumers’ attitudes. In this regard, Somerset and Markwell [48] investigated the impact of school-based food gardening programs on students’ attitudes towards vegetables and fruits consumption. They found positive results. Lee and Matarrita-Cascante [18], and Grebitus, Printezis [49], considered attitudes as drivers of urban agriculture participation and found attitudes to be important. We test the effect of attitudes on CG participation.

Knowledge is also linked to consumers’ behavior. It can be conceptualized in three different ways: subjective knowledge, objective knowledge and usage experience [50, 51]. Accordingly, the first one is related to the individual’s self-perceived knowledge based on experience while the second one represents the individuals’ actual knowledge based on accurate information stored in long-term memory. Subjective knowledge can play a greater role than objective knowledge in consumer decision-making process, as it can determine the level of uncertainty depending on the strength of one’s self-belief and experience. In fact, the previous literature provides empirical results that support this argument in various contexts including sustainable food consumption [52]. Hence, we test the role of knowledge on CG participation.

In addition, “reasons” can be considered important with regards to intrapsychic factors. Modern rational choice theory postulates that “a rational agent has beliefs and desires, and acts so as to satisfy his or her desires in accordance with his or her beliefs” [53, pp. 1]. The same authors asserted that this theory overlooks the role of reasons in preference formation, mainly because the theory ignores the fact that individuals can modify or revise their preferences, for example, as a result of responding to reasons. Recognizing the role of reasons, these authors developed “a reason-based choice theory”. Earlier philosophical works also recognize the importance of reasons in rational decision-making. Some examples of such works are cited in Dietrich and List [53] including Parfit [54], Wallace, Pettit [55] and Raz [56]. In line with this literature, we seek to understand the role of reasons related to decision making using the example of growing foods in CGs. Here, we anticipate that consumers’ preferences for CGs depend on various reasons, which can act as motivating or constraining factors. Thus, we examine whether reasons for CG participation influence consumer preferences for growing foods.

In Fig 1, we provide a summary of our theoretical framework showing pathways for the linkages and interactions between consumers’ choices for growing foods at CGs, and contextual and intrapsychic factors. The framework also shows the possibility of combining choice data with data representing intrapsychic factors to provide a deeper understanding of consumers’ behavior in the context of growing foods at CGs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Conceptual framework showing the relationships between contextual and intrapsychic factors and the choice for growing foods in CGs.

Dotted lines represent our contribution to the literature where there are no studies, yet, investigating these interactions using latent class analysis.

https://doi.org/10.1371/journal.pone.0243949.g001

3.5.1 Discrete choice modeling.

Using data from the DCE we employ discrete choice modeling which is rooted in utility theory [21]. Specifically, we assume that individuals participate in CGs if their utility from participation is maximized. We further assume that individuals maximize their utility by making choices among alternatives of CGs described by different contextual attributes. Therefore, the utility, U, of an individual, n, among, g, alternatives of CGs in choice situation, t, can be formulated as: (1) Where V_ngt is the deterministic part of utility, and ε_ngt is the random part of utility. This recognizes the randomness of utility because the analyst cannot observe all the factors that can influence individuals’ utility. The analyst can only observe the factors that the individual faces [67]. In the case of choices of CGs to grow foods, the analyst only observes a limited set of contextual factors that affect participation. Because utility is fully known to the individual making choice decisions, she chooses alternative, g, if the utility obtained from this alternative is strictly greater than the utility obtained from alternative, k. In addition, because utility is random, analysts can only predict choices of individuals using probabilistic models. Thus, the probability of choosing alternative, g, can generally be expressed as: (2) Therefore, for alternative g to be chosen, the difference between the error terms should be less than the difference between the deterministic parts of the utility for all ∀g ≠ k. Estimation requires the specification of the cumulative distribution of the probability function over the density of the error terms. (3) Different probabilistic models can be specified depending on the assumptions placed on the distribution of the error terms [67].

3.5.2 Latent class models.

Following the framework discussed in 3.5.1, the probability, p, that an individual, n, chooses alternative, g, over alterative, k, can be modeled as: (4) where β_nX_ngt represents the deterministic part of utility, V_ngt, expressed as the product of the vectors of explanatory variables, X_ngt, and the corresponding vectors of the estimated coefficients, β_n. Eq (4) is the simple logit model, which has important limitations related to handling preference heterogeneity [67]. Latent class (LC) models can address these limitations by accounting for unobserved heterogeneity across segments of consumers [68]. Specifically, LC models allow for the identification of latent groups, which are allocated to different classes. Members of consumers across classes have heterogeneous preferences whereas those within a class have homogeneous preference. The peculiar feature of the LC models is that they allow us to use not only consumer characteristics but also intrapsychic factors to determine class membership. This feature makes them popular in the literature concerned with using intrapsychic factors to explain consumers’ preferences. We also employ these models as our study focuses on linking consumers’ preferences for CGs to their intrapsychic constructs.

In the case of LC models, the utility function in Eq (1) can be decomposed into a choice model and a membership model. Thus, according to Boxall and Adamowicz [68], we specify a joint probability that an individual, n, is allocated to class, c, and chooses alternative, g, as: (5) where ω_c denotes the class-specific vector of estimated parameters, and Z_n represents consumers’ characteristics and intrapsychic variables. These variables include Age (age in years); Household size (number of persons in a household); Gender (dummy variable 1 if female, otherwise 0); Attitudes (attitudes towards growing foods in CGs); Knowledge (familiarity and experience with growing foods in CGs); Reasons_Food_Health (food and health related reasons for CG participation); Reasons_Social_Emotion (social and emotional related reasons for CG participation). For the latter four variables, scores are calculated for each respondent based on weights obtained from factor analyses.

There is a standard procedure of normalizing the estimated membership coefficients of one of the classes to zero to identify membership parameters for the other classes, which should be described relative to the normalized class [68]. The optimal number of classes is commonly determined based on the minimum statistical information criteria such as Akaike Information Criteria (AIC), Corrected-AIC (CAIC), and Bayesian Information Criteria (BIC) [69]. We implement the same approach in the present study. The LC models are estimated using the maximum likelihood procedure [67]. Once the estimated coefficients in β_n are obtained, willingness-to-pay (WTP) can be calculated for each class by dividing the negative of the coefficients of non-monetary attributes by the coefficients of the cost attribute.

3.5.3 Factor analysis.

As stated in section 3.4, we measure the different intrapsychic variables based on ordinal scales resulting in ordinal responses. These responses can be used as explanatory variables in econometric models dealing with consumer behavior. However, the question of how to treat them is unsettled in the literature. Some researchers collapse ordered responses into binary variables by using arbitrary cutoff points. One shortcoming of this approach is the possibility of losing important information [70]. Another approach includes using factor analysis [71], which enables us to condense a large set of ordinally scaled and correlated items into a few independent common factors underlying a latent construct. We employ such analysis with varimax rotation to extract factors underlying the latent intrapsychic constructs including participants’ attitudes, subjective knowledge, and reasons. We determine the reliability of the extracted factors using Cronbach’s alpha. Factor scores are calculated for each respondent based on predicted values of the factors, yielding the weighted sum of standardized scores. These scores are subsequently used as explanatory variables in the analysis.

4.2.1 Estimation results from latent class models.

We estimate up to 10 classes with and without covariates to determine the optimal number of classes for the LC model. The log-likelihood and pseudo R-squared values reveal improved model performances as the number of classes increase, suggesting the existence of multiple latent groups in the samples. Searching for the optimal number of classes with covariates suggests five or six classes depending on the statistical information criteria. The BIC, CAIC and BIC (L²) reveal six classes for both samples in Phoenix and Detroit. The results from this estimation are presented in Tables 5 and 6 for the samples from Phoenix and Detroit, respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Estimation results from latent class models–Phoenix.

https://doi.org/10.1371/journal.pone.0243949.t005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Estimation results from latent class models–Detroit.

https://doi.org/10.1371/journal.pone.0243949.t006

We start by describing results for the Phoenix sample. Looking at Table 5, the probability that a consumer belongs to class one, class two, class three, class four, class five, and class six is 24%, 19%, 18%, 16%, 12% and 11%, respectively. Consumers in class one (24%) attach positive utility to most of the characteristics of CGs. Specifically, they derive positive utility from participating in CGs where gardening tools, guidance to garden and social events are provided. The same is true for CGs where the cost to use them is low and which are accessible by bike within ten minutes from home. The estimated coefficient for the no CG option is highly negative as compared to the estimates in other classes. This option explains around 28% of the choice variances next to the cost attribute, which explains 30% of the variances (see S4 Appendix). Members of this class are more likely to be male and young. They are associated with favorable attitudes and subjective knowledge about growing foods in CGs. Their participation in CGs is likely to be motivated by reasons related to improved food choice, personal health, social engagement, and emotions. Thus, we name this class as proponents of growing foods.

Class two (19%) contains consumers who are more likely to be young. Non-monetary attributes are not significant in this class but the estimated parameter for the no CG option is significantly negative suggesting a preference for participation in CGs. This can be associated with their subjective knowledge about growing foods in CGs and their reasons for CG participation, which are tied to social and emotional benefits. In fact, the relative importance of the no CG option is approximately 65% indicating that the total variability in choice was sensitive to this option. However, the only attribute, which affects their preferences for CG participation significantly, is cost. This attribute is also the second most important attribute in this class explaining 18% of the choice variances. Thus, we name this class as content with price.

Consumers allocated to class three (18%) are likely to be represented by male consumers with a large household. They derive positive utility from provision of tools and guidance to garden. However, their preferences for traveling by car and by bike are significantly negative, thus they are labelled as prefer to walk consumers. These consumers derive negative utility from the no CG option, which accounts for 19% of the choice variances. The cost attribute explains half of the choice variance (50%) in this class. These preferences for growing foods in CGs can be associated with their positive attitudes, subjective knowledge, and beneficial food choice and personal health related to CG participation.

Class four (16%) is devoted to consumers who are more likely to be represented by male consumers with a large household. They derive positive utility from provision of tools and guidance to garden. However, most of the variability in choices is explained by the cost attribute (79%). In addition, the absolute value of the estimated cost coefficient is the highest in comparison to the estimated coefficients in the other classes, which indicates that increasing costs generates the highest disutility for CG participation. Thus, we label this class as price sensitive. Although preferences in this class are positively linked to favorable subjective knowledge and attitude, these consumers grow foods in CGs only if the cost of doing so is very low.

Class five (12%) belongs to consumers who are likely to be represented by relatively young consumers. They have positive preferences for the no CG option, which represents non-participation in CGs. However, they obtain positive utility from participating in CGs where tools and guidance to garden are provided. This positive utility can be explained by their reasons for CG participation, which include benefits related to improved food choice, health, and social and emotional feelings. In terms of attribute importance, 24%, 21% 19%, and 17% of the choice variances are explained by the attributes representing costs, no CG option, whether tools, and guidance to garden are provided or not, respectively. What makes members of this class distinct is their significantly positive preference for traveling by car or by bike to CGs. Thus, we label consumers in this class as prefer to use car and bike consumers.

The last class, class six (11%) contains consumers who are more likely to be characterized as being females, older, and small in household size in comparison to consumers in other classes. These consumers can be regarded as opponents of growing foods in CGs for the following reasons. They derive positive utility from provisions of tools and guidance to garden but they have a highly positive preference for the no CG option. As also indicated in section 3.5.2, class membership results are interpreted relative to this class. Given that the other classes are positively associated with two or more intra-psychic factors, members of this class have the opposite intrapsychic orientations in terms of unfavorable attitudes and low subjective knowledge. Further, they are more likely to be negatively associated with the reasons for CG participation. Thus, the tendency of non-participation is likely explained by their negative intrapsychic constructs. The cost attribute explains 41% of the choice variance followed by the no CG option (23%) and the attribute indicating whether tools are provided or not (11%).

Table 6 presents the estimation results for the sample obtained from Detroit. The probability that a consumer belongs to class one, class two, class three, class four, class five, and class six is 28%, 22%, 18%, 15%, 9% and 8%, respectively. Class one (28%) includes consumers who are likely to be overrepresented by young individuals in large households. They react positively to all non-price attributes. In addition, they have the highest negative reaction to the no CG option. As a result, we label these consumers as proponents of growing foods. Their positive preferences for CGs can be associated with a high subjective knowledge and reasons tied to the benefits of CG participation in terms of improved food choice and personal health. While the no CG option and the cost attributes explain much of the choice variance (24% and 30%, respectively), members of this class attach the highest utility to tool-provision and guidance to garden. These attributes explain 20% and 13% of the choice variance, respectively (see S5 Appendix).

Consumers in class two (22%) are also likely to be young with a large household. They differ from consumers in the other classes in that they seem to be price insensitive albeit the estimated coefficient for the cost parameter is insignificantly small. It accounts for only 6% of the choice variance. Therefore, we call this class price insensitive. Members of this class derive higher utility from tools provision, guidance to garden, and social events. They have significantly negative preferences for the no CG option suggesting a preference for participation in CGs. They are likely to be interested in the benefits of CGs associated with improved social engagement and emotions, which can explain their negative preference for non-participation in CGs. In terms of relative importance, the no CG option explains 58% of the choice variance while the attributes social events organized and tools provided account for 12% of the variance each.

Class three (18%) entails consumers who derive positive utility from participating in CGs where tools and guidance to garden are available. They also derive positive utility from using cars to get to the CG. While they do not show any distinctive preferences in comparison to the other classes, they can be identified by their strong preferences for gardening tools and guidance to garden attributes. These attributes explain 20% and 16% of the choice variance, respectively. Thus, we refer to these consumers as content with tools and guidance. The results related to intrapsychic constructs in the membership models suggest that these preference structures are likely to be explained by favorable attitudes about growing foods in CGs and beliefs about the food choice and personal health benefits.

Class four (15%) is more likely to be represented by consumers with a large household size. This class is labelled as price sensitive, as its members are likely to derive highly negative utility from the cost attribute compared to the other classes. In addition, this attribute accounts for the greatest proportion of the total choice variance (72%). The only non-monetary attribute that has a significantly positive influence on consumers’ utility in this class is the guidance to garden. However, members derive negative utility from the no CG option indicating their preference for CG participation. Their participation in CGs is likely to be motivated by reasons related to improved food choice and personal health.

Class five (9%) is likely to be overrepresented by female consumers with a large household. They obtain significantly higher utility from receiving gardening tools and traveling by car. These positive preferences can be related to their motivation for CG participation, which is the ability to build social relationships and enjoy positive emotional attachment to their neighborhoods. Their strongest positive preferences for traveling by car distinguishes them from the other classes. This attribute comes first in terms of relative importance explaining much of the choice variance (25%). Thus, we label members in this class as prefer to use car consumers. Another notable result in this class is the positive preference for the no CG option, which is the second most important attribute (24%) in this class. Thus, despite their socially-oriented motivation to participate in CGs, members in this class may support non-participation particularly if gardening tools are unavailable and if the gardens are not accessible by car.

Class six (8%) can be labelled as opponents of growing foods because they have the strongest positive preference for the no CG option. These consumers mainly consist of older individuals with fewer household members. Membership in this class is likely to be based on unfavorable intrapsychic orientations. These negative orientations explain the strongest positive preferences for non-participation. In fact, this class does not show positively significant preferences for the non-monetary attributes suggesting their skepticism about CG participation. The no CG option is the third most important attribute accounting for 12% of the choice variance next to the cost and traveling by bike attributes, which explain 33% and 31% of this variance, respectively.

4.2.2 Segment-specific consumer valuation of CG characteristics.

To directly compare consumers’ preferences in Phoenix and Detroit, we report segment-specific WTP values for each attribute in Table 7. Though we translate the value into WTP in U.S. Dollars, we interpret this as a relative value rather than an absolute value given it is derived by dividing the coefficients of the non-monetary attributes by the negative coefficient of the cost attribute. Results reveal differences in consumers’ valuations of CG characteristics across segments and samples. In both samples, the “proponents of growing food” consumers in class one are willing to pay more for most of the CG characteristics while the “Opponents” in class six have insignificant WTP values albeit those in Phoenix have positive WTP for the provisions of tools. The former attaches the highest values to tools, guidance and social events in the order of ranking. However, the only transportation mode that received significantly positive valuation by these consumers is traveling by bike in Phoenix. The “Price insensitive” consumers in class two have insignificant WTP values in Detroit while this is true for the “Content with price” consumers in class two in Phoenix. In both samples, the highest WTP values are found for the “Prefer to use car and bike” consumers in Phoenix and for the “Prefer to use car” consumers in Detroit. While the former consumers in Phoenix are willing to pay more for the CG characteristics except social events, the latter consumers do this only for using a car to travel to a CG and provisions of tools. We find negative WTP values for the Phoenix sample in the classes “Opponents of growing food” and “Prefer to walk”, which indicates that consumers in these classes are willing to accept compensation to travel by car or by bike to offset the disutility from using such transportation modes. In Detroit, similar preference structures are observed for the “Price sensitive” consumers. These consumers have the lowest WTP values in both sample sites due to their strong sensitivity to costs to use CGs but they are only willing to pay more for CGs where guidance to garden is provided in Detroit.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. WTP values ($).

https://doi.org/10.1371/journal.pone.0243949.t007