Determinants Affecting Public Intention to Use Micro-Vertical Farming: A Survey Investigation

Abstract

Vertical farming is a new branch of urban agriculture using indoor vertical space and soil-less cultivation technology to obtain agricultural products. Despite its many advantages over traditional farming, it still faces some challenges and obstacles, including high energy consumption and costs, as well as uncertainty and a lack of social acceptance. This study aims to investigate the influence of public acceptance on micro-vertical farming based on the deconstructed theory of planned behavior model. This model is adopted for statistical analysis to reveal the factors and their weights in influencing people’s behavioral intentions. The results indicate that the overall mean of the public’s behavioral intentions to use vertical farming is 3.9, which is above neutral (M = 3.00) but less than positive (M = 4.00). Differences in age, education level, and the living area of the public have significantly impacted behavioral intentions. Meanwhile, the statistical results support the hypotheses concerning the behavioral attitudes, subjective norms, and perceived behavioral control of the model, and also demonstrate that their decomposed belief structures considerably influence the public’s behavioral intentions to use vertical farming. Notably, perceived usefulness is the most critical driving factor in planting using vertical farming. The findings of this study contribute to better predictions of the effects of different elements of behavioral intention on vertical farming at the urban scale, which may provide a basis for decision making in the development of sustainable urban agriculture.

1. Introduction

The contemporary world, population expansion, and accelerating urbanization have become important reasons for the increase in the differences between agricultural land and construction land [1]. Apart from shrinking farmland, extreme weather caused by global climate change has had a huge impact on the traditional model of agricultural production, which has raised concerns about food security [2]. In 1999, Despommier introduced the concept of urban “vertical farming” and committed to promoting this new agricultural production mode to deal with the challenges of modern society [3].

The basic idea of vertical farming (VF) is to transform the original single-layer flat agricultural model into a multi-layer vertically stacked model that integrates modern agricultural technologies such as soil-less cultivation, artificial Light Emitting Diode (LED) sources, environmental control, and renewable energy to produce agriculture inside urban buildings [4]. Compared with traditional agriculture, the main advantages of VF include

• Utilizing land resources efficiently and alleviating the struggles between humans and the land [5].
• Improving production efficiency while avoiding the impact of extreme weather on agricultural production and ensuring food safety due to the controlled indoor environment [6].
• Reducing environmental pollution caused by agricultural production processes using renewable-energy-recycling technology and avoiding the use of pesticides [2].
• Establishing urban “local food supply chains” as a supplement and buffer to long-distance food supply chains [7].
• Providing a new employment model and jobs [8].

In addition, VF has certain ecological and environmental benefits. On the urban scale, large commercial VF is usually combined with resource reuse and recycling facilities, such as solid waste treatment, wastewater treatment, and waste heat recycling that enhance the sustainability of the city [9]. In terms of the indoor environment, the air purification effects of the main indoor air pollutants have been verified including carbon dioxide, particulate matter, and some volatile organic compounds [10]. VF also helps to improve the physical and psychological environment due to the visual relaxation effects of ecological designs [11]. In addition, it enhances indoor daylight distribution by providing shade and diffuse reflection in front of the windows [12].

Despite its potential advantages, the concept of VF has been met with skepticism and even criticism. First, due to cost-effective constraints, the types of plants suitable for vertical farming are limited. Leafy greens with high water content rather than crops high in carbohydrates are much more suitable plant species for VF since they grow much faster and consume fewer resources such as electricity and water [13,14]. Second, VF may bring about certain favorable or unfavorable effects on the indoor environment. For example, plant transpiration increases indoor relative humidity by 9–12% [10]. This may be favorable for indoor comfort in arid regions but could be disadvantageous in hot and humid regions [15]. Other considerations include the noise made by hydroponic systems [16], a minor chance of discomfort as a result of allergies [17], and energy consumption and heat dissipation caused by the artificial growth lights [18]. Finally, the exhaust gases, wastewater, and heat emitted from VF could affect the urban environment if not properly treated or recycled [16]. Currently, a series of solutions has been already tested and implemented to deal with the aforementioned challenges. For example, the cost-effectiveness of VF could be greatly enhanced by the careful selection of crop varieties and continuous improvements to the energy efficiency of the VF equipment and systems [19]. As for VF’s possible negative impacts on the environment, the integration of different systems and technologies is needed to reduce, reuse, or recycle the waste from VF [20].

Technology should not be an obstacle as it continues to improve. However, there are still uncertainties around the economic feasibility and social acceptance of VF. Shao et al. [21] developed the world’s first evaluation software to simulate VF’s economic feasibility and energy consumption considering the impact of multiple factors such as region, type, and technology level. The simulated results show that VF could be preferable in areas with high vegetable prices, low energy costs, and high labor costs, where the annual rate of return on investment can be up to 25%. Conversely, it could be easy to develop the project in a state of loss for a long time, thereby reducing commercial attractiveness. Zhang et al. [22] analyzed the economic feasibility of introducing VF to a university park. The results showed that the payback period of a 5000 m² vertical farm was 11.5 years and that the annual profit could reach $92,000. Avgoustaki et al. [23] analyzed and compared the internal rate of return and the net present value of VF and greenhouses for the same planting area (225 m²). The study showed that VF was economically more advantageous. Its internal rate of return was 34.74% with a payback period of 4 years. In contrast, Graff [24] estimated the return on investment for a 10-storey vertical farm in Toronto to be only about 8%, which was below the minimum acceptable rate of 10–12% for most investors.

Compared with an economic feasibility analysis of agriculture-oriented VF, studies on the social acceptance of small-scale mixed-use VF, especially family-run or office-based micro-VF, are scarcely documented. Mixed-use VF can be used not only for agricultural purposes, but also for other functions such as recreation [25], education [26], decoration, and exhibitions [27]. For this type of VF, agricultural production is usually considered a value-added function rather than the main function. In mixed-use VF, occupants are often more concerned with the improvements to the quality of the indoor environment brought about by vertical farming than the annual yield of agricultural produce. Mixed-use VF with the most market potential are family farms since the proportion of residential land is usually the largest in cities. However, the uncertainty lies in public acceptance and the preference for consumption. In order to research and develop this market, this study applies surveys and information systems to provide an important basis for market forecast analysis, taking a comprehensive consideration of objective conditions and subjective willingness [28].

In this study, the decomposed theory of planned behavior (DTPB) model was utilized to investigate the influence of public acceptance on the use of VF. Furthermore, partial least squares structural equation modeling (PLS-SEM) was adopted to test the hypothetical relationship in the theoretical framework. From a subjective perspective, an in-depth analysis of the factors that affect public acceptance and willingness to adopt VF was conducted in order to analyze the decision-making aspects of VF planting behavior and make tailored recommendations for its promotion and development.

2. Literature Review and Hypotheses

2.1. Theories and Models of Users’ Behavioral Intentions

In recent years, research on information systems applications has developed rapidly. The study of users’ behavioral intentions (BI) has become an important part of research on user acceptance [29]. Models and theories have been used to predict user behavior, including the Theory of Reasoned Action (TRA) [30], Theory of Planned Behavior (TPB) [31,32], and Decomposed Theory of Planned Behavior (DTPB) [33]. The TRA suggests that the execution of individual behavior is governed by BI [34], but this theory ignores the influence of objective reality on BI, resulting in its limited and questionable scope of application [34]. To compensate for the shortcomings of the TRA, Ajzen [35] incorporated Perceived Behavioral Control (PBC) into the theoretical model to form the TPB, which broadened the application area of the theory. However, as the research continued, many scholars found that the TPB has the same limitations as the TRA, and the belief dimensions of the two theories show multiple dimensions in many situations [36]. Accordingly, Taylor and Todd [37] combined the Diffusion of Innovation Theory (IDT) with the TPB in 1995 and proposed the DTPB using the second-order deconstruction of the three single linear structures in the TPB. Antecedent variables can be added or reduced in the model according to the specific research object and scenario, which is conducive to further explorations of the deeper psychological perceptual elements of individual behavior. Compared with the TRA and TPB, the DTPB has a broader perspective and greater explanatory power and applicability in multiple research fields, since a more stable belief structure can be built [38]. Therefore, although yet to be developed and applied in VF research, DTPB modeling could be an effective approach to investigating social acceptance and willingness to adopt VF.

2.2. Factors Influencing Acceptance and Willingness

From the perspective of behavioral theory, for individuals, acceptance and willingness to use new technologies are the result of the combined effects of various factors such as their socioeconomic attributes, social relationship norms, and behavioral control [39]. Studies have found that an individual’s attributes, such as gender, age, occupation, education level, income level, etc., have an impact on their willingness to use new technologies [40]. Studies have shown that people with higher income levels are more concerned about the quality of agricultural products than the price [41]. Younger groups are more receptive to and tolerant of new technologies. Women are more cautious in viewing and using new technologies than men [42]. As a new planting technology, the complexity of the planting process of vertical agriculture and the consumption of individual money, time, and energy in this process are also important factors that impact public planting intention. In addition, the advice and behavior of people around an individual are also important factors. Therefore, this paper first investigates the influence of an individual’s own socioeconomic attributes on planting intentions and then investigates the influence of other factors such as social relationship norms and behavior control.

There are two types of the disaggregate behavioral model of individual behavioral intentions [43]. RP (revealed preference) is an investigation of an individual’s completed choice behavior [44], and SP (stated preference) is an investigation of how an individual makes a choice and how he considers it under hypothetical conditions [45]. RP is for investigating the results and conditions of choices made by individuals in a certain actual state, mainly targeting actual or occurring schemes, that is, the contents of the survey are the observed behavioral choice data that have taken place [46]. SP is mainly used to investigate the factors and degree of influence that affect consumers’ acceptance of a product or service, and the main purpose is to obtain people’s subjective preferences for behavioral choices under assumed conditions that have not yet occurred [47]. Since there were no existing data on vertical agricultural planting behavior, the SP survey method was selected for this study.

2.3. Decomposed Theory of Planned Behavior

The DTPB is a type of SP survey method that uses behavior intention (BI) to measure public acceptance of and willingness to use new technologies. The DTPB suggests that BI can have a direct effect on an individual’s behavior. It means that a strong intention to perform a certain behavior indicates that an individual is more likely to perform that behavior [48]. Meanwhile, the DTPB assumes that BI is determined by three main constructs: behavioral attitude (BA), subjective norm (SN), and perceived behavioral control (PBC) [49]. These three components are expressed in behavioral decision contexts as behavioral beliefs, normative beliefs, and control beliefs. BA is an individual’s evaluation of a particular behavior [50]. SN specifically refers to the expectations that the individual perceives from the surrounding group, which may have a significant influence on the individual when making a certain behavioral decision [51]. PBC is a person’s perception of the ease or difficulty of performing a particular behavior. It is mainly expressed in the perceived ability of the individual to control the required resources and opportunities [52]. The relationship between BI and the three determinants, BA, SN, and PBC, has been supported by previous research [53]. In the context of the present study, positive feelings toward the use of VF can encourage the public to accept it. Being introduced to the concept by family, relatives, friends, or colleagues, or encouraged by elders or superiors, can influence the public’s behavior toward using VF. Similarly, skills and conditions that facilitate VF can influence the intention to adopt VF. Based on the DTPB model and evidence from the empirical literature, the following are hypothesized:

H1. 

Behavioral attitude (BA) has a positive influence on the public’s behavioral intention (BI) to accept and plant using VF.

H2. 

Subjective norm (SN) has a positive influence on the public’s behavioral intention (BI) to accept and plant using VF.

H3. 

Perceived behavioral control (PBC) has a positive influence on the public’s behavioral intention (BI) to accept and plant using VF.

Behavioral belief refers to the possible results of VF perceived by the public, which is the determinant of BA [54]. Davis used PU and PEU in the TAM, which theorizes that an individual’s perception of system usefulness and ease of use influence his or her attitude toward system usage as well as his or her behavioral intention, which in turn determines system acceptance and its usage [55]. He defined PU as “the degree to which a person believes that using a particular system would enhance his or her job performance” and PEU as “the degree to which a person believes that using a particular system would be free from effort” [55]. In the planting process of vertical agriculture, PU can be redefined as the extent to which people perceive that VF can help them increase their income and improve their indoor environment, and even the extent to which VF plays a role in sustainable urban development; PEU represents the ease of mastering the planting process and operating the VF structure, as well as the convenience of accessing and navigating the intelligent control system. As for the indoor physical environment, VF will also have some negative effects. As well as the benefits, growers should also weigh the adverse effects of VF on their living or working environments.

Based on this, this paper deconstructs the behavioral beliefs into three constructs, PU, PEU, and PR, and selects four indicators to measure planting intention: economic benefits, environmental benefits, operational difficulties, and risk assessment. The higher the benefit of the planting behavior or the easier the grower perceives the planting operation to be, the stronger the attitude of the grower to perform the planting behavior; at the same time, the lower the negative impact of the planting behavior, the more positive the attitude of the grower to perform the planting behavior. Based on this argument, the following hypotheses are proposed:

H1a. 

Perceived usefulness (PU) has a positive influence on the public’s behavioral attitude (BA) to accept and plant using VF.

H1b. 

Perceived ease of use (PEU) has a positive influence on the public’s behavioral attitude (BA) to accept and plant using VF.

H1c. 

Perceived risk (PR) has a negative influence on the public’s behavioral attitude (BA) to accept and plant using VF.

Normative beliefs refer to the public’s perceptions of the expectations of significant others or teams about whether they should participate in planting using VF, which is the determinant of SN [35]. In reality, when the public is unfamiliar with the various outcomes of planting using VF, they rely on the groups around them to form the basis for their own decisions by integrating the voices of all parties, which in turn facilitates the formation of SNs about the public’s willingness to accept and plant using VF. In this study, SN does not focus on social influences on decision making. SI comes from elders at home or leaders at work, whereas PI comes from family, friends, and colleagues [56].

Based on the above analysis, this paper deconstructs normative beliefs into two constructs, PI and SI, and selects five indicators, family, friends, colleagues, elders, and leaders, to measure planting willingness. The more positive encouragement from the surrounding groups, the greater the willingness of the public to engage in planting using VF. The above statement leads to the following hypotheses:

H2a. 

Peer influence (PI) has a positive influence on the public’s subjective norm (SN) to accept and plant VF.

H2b. 

Superior influence (SI) has a positive influence on the public’s subjective norm (SN) to accept and plant VF.

Control belief refers to the factors that the public can perceive during the process of planting VF, which may promote or hinder this behavior [57]. This perception of control consists of two aspects: the internal aspect comes from the public’s perceptions of their own abilities, such as money and time, and the external aspect comes from the influence of the surrounding external forces on the public’s own behavior. In the process of planting, this external force mainly refers to convenience, that is, the sufficient degree of the resources that the public possesses in order to realize planting behavior, such as the VF product sales platform and planting knowledge learning platform, etc.

According to this analysis, this study deconstructs control beliefs into two constructs, SE and FC, corresponding to six indicators of planting knowledge, reserve, risk-aversion ability, time cost–tolerance, financial cost–tolerance, planting information acquisition ability, and product-trading-platform construction. The greater the public’s recognition of their own abilities or the more favorable the external facilitation conditions, the stronger their willingness to plant. The following two hypotheses reflect this position:

H3a. 

Self-efficacy (SE) has a positive influence on the public’s perceived behavioral control (PBC) to accept and plant using VF.

H3b. 

Facilitating conditions (FC) have a positive influence on the public’s perceived behavioral control (PBC) to accept and plant using VF.

2.4. Partial Least Squares Structural Equation Modeling (PLS-SEM)

In recent years, as an important tool for multivariate data analysis, structural equation modeling (SEM) has become a mainstream method in the field of social science research [58]. Based on different research techniques, SEM can be divided into covariance-based structural equation modeling (CB-SEM) [59] and partial least squares structural equation modeling (PLS-SEM) [60]. Recently, PLS-SEM applications have expanded into marketing research and practice with the recognition that PLS-SEM’s distinctive methodological features make it a possible alternative to the more popular CB-SEM approaches [61]. Compared with CB-SEM, PLS-SEM has the following main advantages: (1) it requires fewer samples; (2) it can handle complex models with multiple latent variables; (3) it can handle both reflective and formative indicators; (4) it can handle non-constant information; and (5) it can reduce the impact of multicollinearity on modeling accuracy and reliability [62].

As shown in Figure 1, PLS-SEM consists of two parts. The first part is the measurement model, also known as the external model. The external model assesses the contribution of each indicator in representing its related underlying variables and evaluates the degree to which a set of indicators fits a potential variable. The second part is the structural model, also known as the internal model, which measures the direct and indirect relationships between latent variables [63]. Among them, X1… X4 and Y1… Y3 represent the observed variables, ξ1 and ξ2 are the latent exogenous variables, η1 represents the latent variables, and e1… e7 represent the error terms associated with the observed variables [64]. Among them, the relationships between ξ, X, and e and η, Y, and e can be expressed by Equations (1) and (2), respectively:

$$\xi =\lambda \cdot X+e$$

(1)

$$\eta =\lambda \cdot Y+e$$

(2)

where λ represents the path coefficient. In SEM, the relationship between the observed variable and the latent variable is a linear function; the meaning of the latent variable is reflected in the observed variable, and a change in the latent variable will lead to a change in the observed scalar.

As shown in Figure 2, the basic evaluation process of PLS-SEM is divided into two stages. The first stage evaluates the measurement model and the second stage evaluates the structural model. The measurement model assessment is mainly to test the quality of the theoretical models. Prior to beginning, researchers usually check the indicator loading of each variable. The minimum threshold value for the loadings is 0.70, which indicates that the construct explains over 50% of the indicators’ variances. The structural model assessment covers structural theory, including determining whether structural relationships are important and meaningful and then testing the hypotheses. Consistent with other analytical methods, PLS-SEM relies on empirical rules to evaluate the results estimated by the model. What follows is a discussion and description of the thresholds for each indicator.

2.4.1. Measurement Model Assessment

The first step in the measurement model evaluation involves the assessment of the constructs’ internal consistency reliability. When using PLS-SEM, internal consistency reliability is typically evaluated using composite reliability (CR) and Cronbach’s alpha (CA) [65]. The values of CA and CR of all variables exceeding 0.7 assured reliability in the internal consistency among these constructs. However, CR values higher than 0.95 are considered problematic as they indicate that the items are redundant, leading to issues such as inflated correlations among the indicator error terms [66].

In the second step, the convergent validity of the measured constructs is examined. Convergent validity measures the extent to which a construct converges in its indicators by explaining the items’ variances [67]. Convergent validity is assessed by the average variance extracted (AVE) for all items associated with each construct. In the case of indicator loading and CR higher than 0.7, the recommended value of AVE is 0.50 or higher [68] as it indicates that on average, the construct explains over 50% of the variances in its items.

Once the reliability and convergent validities of the reflective constructs are successfully established, the final step is to assess the discriminant validities of the constructs. Discriminant validity refers to the extent to which a measure is distinct from other measures from which it is supposed to differ [58]. The most conservative criterion recommended to evaluate discriminant validity is the Fornell and Larcker criterion [69]. The method compares each construct’s AVE value with the squared inter-construct correlation of that construct with all other constructs in the structural model [70]. The evaluation criterion is that a construct should not exhibit shared variance with any other construct that is greater than its AVE value [71]. Another approach to assessing discriminant validity is to examine the cross-loadings [71]. The evaluation criterion for this approach is that an indicator variable should exhibit a higher loading on its own construct than on any other construct included in the structural model [71]. If the loadings of the indicators are consistently highest on the construct with which they are associated, then the construct exhibits discriminant validity.

2.4.2. Structural Model Assessment

The next step after having established the constructs’ validities is to examine the causal relationship among the latent constructs. This covers structural theory, which includes determining whether the structural relationships are significant and meaningful and then testing the hypotheses [71]. Prior to this assessment, researchers must test for potential collinearity between the structural model’s predictors. When the values of all variance inflation factors (VIF) are lower than 3.3, it indicates that there is no collinearity problem in the model [72].

The first step of the structural model evaluation is to test the goodness of fit of the model. This study adopted several criteria to assess the PLS-SEM model’s fit, including the standardized root mean square residual (SRMR), the squared Euclidean distance (d-ULS) and the geodesic distance (d-G), and the Normed Fit Index (NFI) [73]. The highest threshold of SRMR is 0.08 [74], whereas the recommended value of NFI is 0.7 and above [75].

Next, one of the important parts of evaluating an SEM is testing the predictive accuracy and predictive relevance of the model [76]. The predictive accuracy is tested using the coefficient of determination (R² value), which presents the degree of variance explained in each endogenous construct. As explained by Hair et al. [77], a value of R² ranging between 0 and 1 with a higher value of R² indicated a higher level of predictive accuracy. Specifically, 0.19, 0.33, and 0.67, respectively, represent the three levels, small, medium, and large prediction accuracies [78]. Another means of assessing a model’s predictive relevance is the Q², also known as blindfolding [79]. This method is based on the blindfolding procedure of PLS-SEM 3.0. As a rule of thumb, Q² values greater than zero for a particular endogenous construct indicate that the model’s predictive accuracy is acceptable for that particular construct [71].

Finally, the strength and significance of the path coefficients are evaluated for the relationships hypothesized between the constructs. Generally, the direct and indirect relationships among the constructs are evaluated using regression coefficients (β) [59]. In terms of relevance, β values are standardized on a range from −1 to +1, with coefficients closer to +1 representing strong positive relationships and coefficients closer to −1 indicating strong negative relationships [71]. In addition, the bootstrap procedure was conducted to assess the significance of the β values in the indirect relationships among the constructs based on the t-value and p-value. When the t-value is higher than 1.96 or the p-value is less than 0.05, the path relationship is considered to be significant at the 95% significance level [80,81].

3. Materials and Methods

3.1. Research Framework

Figure 3 shows the workflow of the research that is divided into three parts: theoretical framework, survey, and hypothesis testing. In the first part, the theoretical framework includes three steps. Step A is to build a theoretical model by deconstructing the three dimensions, behavioral, normative, and control beliefs, in the DTPB based on the psychological characteristics of the public’s intentions to use VF. Step B is to determine the variables’ structures and define each variable. Step C is to formulate the hypotheses based on the operational logic of the theoretical model and the variable relationships. Step D focuses on the design, implementation, and data collection of the online survey. Finally, in Step E, the data are statistically analyzed to verify the hypotheses. This part includes the descriptive analysis, measurement model evaluation, and structural model evaluation.

The study was approved by the institutional review board of Nanjing Tech University, China. All participants provided written informed consent.

3.2. Theoretical Model

In view of the psychological basis and characteristics of the public’s intention to use vertical agricultural planting, this paper decomposed the three structures of BA, SN, and PBC into the DTPB model and divided them into seven levels: benefit, ease of use, risk, superior, partner, self, and outside. The specific model is shown in Figure 4. The measurement instrument was developed based on the original instrument used by Ahmed [82] for DTPB constructs and adapting the items to the context of VF. Among the constructs that affect BA, perceived risk (PR) was introduced into the model in conjunction with the specific context of VF, together with perceived usefulness (PU) and perceived ease of use (PEU) to constitute the behavioral beliefs [83]. Normative beliefs maintained the structure of the original model, retaining the two constructs of peer influence (PI) and superior influence (SI) [84]. PBC was divided into two constructs, self-efficacy (SE) and facilitating conditions (FC) [54].

As shown in Figure 4, the model framework established in this study is mainly based on the antecedent variables PU, PEU, PR, PI, SI, SE, and FC, the intermediary variables BA, SN, and PBC, and the outcome variable BI. Each variable and construct were defined, as summarized in

Table 1. Definition of the constructs and measurement items.

Constructs	Definition	Item	Measurement	References
Perceived Usefulness (PU)	People perceive how much benefit can be obtained by planting using VF.	PU1	I think that planting using VF can bring a high yield of vegetables and fruits.	[85]
		PU2	I think that planting using VF is beneficial to the sustainable development of the city.
		PU3	I think that indoor VF is not only pleasant to view but also improves the indoor environment.
		PU4	I think that VF is cost effective.
Perceived Ease of Use (PEU)	People’s perceptions of the complexity of the VF planting process.	PEU1	The settings of the automatic control system on the VF equipment are easy for me.	[86]
		PEU2	The VF planting process is easy for me.
Perceived Risk (PR)	People’s perceptions of the unforeseen and uncertain consequences involved in the process of planting using VF.	PR1	I am worried that the noise generated during the operation of the VF water circulation system will affect work or rest.	[87]
		PR2	I am worried that plant transpiration will release water vapor and adversely affect indoor humidity.
		PR3	I am worried that VF planting will increase the number of indoor mosquitoes.
		PR4	I am worried that VF will affect the normal use of indoor spaces.
Peer Influence (PI)	The influence of the opinions of family, friends, and colleagues on people’s planting behavior.	PI1	My family should have a positive attitude toward planting using VF.	[88]
		PI2	My colleagues should have a positive attitude toward planting using VF.
		PI3	Most of my friends should have a positive attitude toward VF planting.
Superior Influence (SI)	The influence of the opinions of elders and leaders on people’s planting behavior.	SI1	The elders in my family should have a positive attitude toward planting using VF.	[89]
		SI2	The leaders of our unit should have a positive attitude toward planting using VF.
		SI3	Under normal circumstances, I will do what my elders and leaders agree to.
Self-Efficacy (SE)	People’s self-judgment on planting using VF.	SE1	I think I understand the knowledge and processes related to planting using VF.	[90]
		SE2	I think I have the ability to deal with the possible negative effects of VF.
		SE3	I think I have enough time to plant using VF.
		SE4	I think I can afford the initial costs and operating expenses of VF.
Facilitating Conditions (FC)	People’s perceptions of the availability of resources required for planting using VF.	FC1	I think that VF can be grown easily using smart control systems such as mobile phones.	[91]
		FC2	I can easily purchase VF-related equipment from major e-commerce platforms.
		FC3	I can easily understand and learn VF-related knowledge from the Internet.
Behavior Attitude (BA)	People’s positive or negative evaluations of planting using VF.	BA1	I think that planting using VF is worth promoting in the residential environment.	[92]
		BA2	I think that planting using VF is worth promoting in the office environment.
		BA3	I think that VF should be used in living or working environments.
Subjective Norms (SN)	People can feel social pressures about whether to plant using VF.	SN1	The people I value (colleagues or family members) approve of my practice of planting using VF in the office or at home.	[93]
		SN2	People who have an influence on me (elders or leaders) agree that I plant using VF in the office or at home.
Perceived Behavioral Control (PBC)	People’s perceptions of the mastery of planting using VF.	PBC1	I have the necessary conditions for VF.	[94]
		PBC2	I have the ability to bear the negative effects and risks brought about by VF.
		PBC3	I think that there are many convenient reasons for planting using VF.
Behavioral Intention (BI)	Intensity of people’s willingness to plant using VF.	BI1	I am willing to try VF within the next 12 months.	[95]
		BI2	I am willing to share relevant knowledge of VF with people around me and will continue to pay attention to its development.

.

3.3. Survey

The questionnaire consisted of four parts. Since VF is a relatively new concept to most consumers, to avoid any uncertainty or ambiguity about the term, information about it was provided in the first part of the questionnaire, including the definition, types, advantages, current problems, and cost-effectiveness. The second part was designed to check whether respondents fully understand the information given in the first part by asking a few clear questions as a preliminary test. Only if a respondent answered all the questions correctly were his questionnaire results retained for subsequent analysis. In the third part, respondents’ sociodemographic characteristics were collected including gender, age, education level, and annual household income. The last part of the questionnaire mainly focused on consumer attitudes and willingness to adopt VF, including acceptance, willingness to plant, and willingness to purchase. It consisted of 33 questions (see Table 1) corresponding to the 11 variables in the theoretical model. The questionnaire structure was measured using a 5-point Likert scale [96] ranging from strongly disagree (1) to strongly agree (5) to record respondents’ responses.

Before the large-scale questionnaire, three experts were pre-tested to determine the wording and order of the questions. After some modifications, a pilot survey was conducted with 30 respondents to ensure the clarity of the questions. After that, an online survey was initiated. In order to encourage respondents to complete the survey, they were informed that they could enter a lucky draw with a cash bonus after answering all the questions.

The questionnaire was distributed online from 10 February to 16 February 2021, and a total of 4065 questionnaires were returned, of which 987 passed the preliminary test and 61.5% had heard of the concept of VF. After excluding incomplete questionnaires and those with a high repetition rate of answers, the number of valid questionnaires was 740, and the effectiveness of the questionnaire was calculated to be 75.0%.

3.4. Data Analysis

SPSS22.0 was used for statistical analysis of the data collected from the survey. The data analysis process was divided into two parts: (1) descriptive analysis and (2) structural equation model analysis. In terms of the descriptive analysis, it mainly investigated the differences in the acceptance and willingness to adopt VF among people of different groups categorized by gender, age, educational level, annual household income, and living area. The main methods included descriptive statistics, variance analysis, and Least Significant Difference (LSD) [97]. In terms of the structural equation model analysis, the DTPB, which contains a third-order structure, was selected as the theoretical model. In order to reduce the influence of multicollinearity on modeling accuracy and reliability, the PLS-SEM was applied. SmartPLS 3.0 was used to evaluate the measurement model and the structural model, and the main evaluation method was the path analysis method. The correlation between different variables can be measured by the path correlation coefficient, or the mutual influence or causal effect relationship between the different variables can be directly judged using the calculation of the path correlation coefficient. The workflow of the PLS-SEM has been shown in Figure 2 and Section 2.3.

4. Results

4.1. Descriptive Statistics

The sociodemographic characteristics of the respondents are shown in

Table 2. Survey respondent characteristics.

Profile	Characteristics	Frequency	Percentage
Age	Below 18	26	3.5
	18–24	251	33.9
	25–30	255	34.5
	31–40	153	20.7
	41–50	41	5.5
	Above 50	14	1.9
Gender	Male	450	60.8
	Female	290	39.2
Education	Secondary school and below	0	0
	Junior middle school	21	2.9
	Senior middle school	113	15.3
	Junior college	128	17.3
	Bachelor’s degree	340	45.9
	Master’s degree and above	138	18.6
Annual family income (CNY ten thousand)	Below 3	70	9.5
	3–8	201	27.2
	8–30	332	44.8
	30–100	120	16.2
	Above 100	17	2.3
Living area	Metropolis	224	30.3
	Ordinary citie	400	54.0
	Towns and villages	116	15.7

. First, the ratio of male to female respondents is close to 3:2, with more respondents being male (60.8%). Second, more than 70% of people are under 30 years old and more than 90% are under 40 years old. The surveyed population has a relatively high level of education, with 64.5% of them having a bachelor’s degree or above. In addition, the survey shows that 63.3% of respondents have an annual household income of more than CNY 80,000, reaching the level of well-off and above.

4.1.1. The Overall Level of the Public’s Intentions to Plant Using VF

The overall mean of the public’s intentions to plant using VF is 3.90, and the means (M) of the three dimensions are between 3.49 and 3.83, all of which are above neutral (M = 3.00) but not relatively positive (M = 4.00). The mean values of each dimension from highest to lowest are “behavioral attitude” (M = 3.83), “subjective norm” (M = 3.73), and “perceived behavioral control” (M = 3.49). This shows that the general public has a positive attitude toward planting using VF but that there are also some concerns.

After further analysis of the internal indicators of each dimension, it is found that the public’s intentions to plant using VF are not balanced in each dimension. First of all, in terms of “behavioral attitude”, the average value of “perceived usefulness” (M = 4.26) is higher than the average of the “behavioral attitude” dimension (M = 3.83), and the overall average of “planting intention” (M = 3.90); The mean values of the indicators of “perceived ease of use” (M = 3.77) and “perceived risk” (M = 3.43) were lower than the mean of the dimension of “behavioral attitude” (M = 3.83) and also lower than the overall mean of “planting intention” (M = 3.90). This shows that public attitudes toward the usefulness of VF are very positive, which is an important factor in people’s willingness to engage in planting behavior. On the contrary, the VF planting process is considered to be complex and its complexity is also an important factor that restricts people’s planting behavior. In addition, people’s perceptions of the negative effects of VF are not significant.

Second, in terms of “subjective norms”, the mean values of “peer influence” (M = 3.83) and “superior influence” (M = 3.82) are both higher than the mean value of the “subjective norm” dimension (M = 3.73) but lower than the overall mean value of “planting intention” (M = 3.90). This suggests that the opinions of surrounding people can influence, but not determine, whether or not people plant using VF.

Finally, the mean of the “perceived behavioral control” dimension (M = 3.49) was significantly lower than the other two dimensions and the overall mean of “planting intention” (M = 3.90). Among them, the mean values of “self-efficacy” and “facilitating conditions” are 3.43 and 3.77, respectively. This indicates that “self-efficacy”, that is, people’s evaluations of their own abilities such as time and money, is an important factor that restricts their intention to plant using VF.

4.1.2. Multidimensional Differences in the Public’s Intentions to Use VF

(1)

Gender: The independent sample t-test results (

Table 3. Comparison of different gender groups’ intentions to use VF.

	Gender (Mean ± SD)		F-Value	p-Value
	Male (n = 450)	Female (n = 290)
BI	3.90 ± 0.87	3.89 ± 0.86	0.006	0.937
BA	3.85 ± 0.86	3.79 ± 0.82	0.749	0.387
SN	3.75 ± 0.97	3.71 ± 1.01	0.272	0.602
PBC	3.57 ± 1.02	3.37 ± 1.07	6.41	0.012 *

Note: *: p < 0.05.

) show that there are no significant differences between the different gender groups in behavioral intention, behavioral attitude, and subjective norms. Although there is a statistically significant difference in perceived behavioral control, the differences in the mean values are small.

(2)

Age: There are significant differences among the different age groups. The LSD multiple comparison method was used for post hoc testing and the specific data are shown in

Table 4. Comparison of different age groups’ intentions to use VF.

	Age (Mean ± SD)						F-Value	LSD
	Below 18 (A) (n = 26)	18–24 (B) (n = 251)	25–30 (C) (n = 255)	31–40 (D) (n = 153)	41–50 (E) (n = 41)	Above 50 (F) (n = 14)
BI	3.63 ± 0.91	3.83 ± 0.90	3.91 ± 0.87	4.06 ± 0.80	3.94 ± 0.75	3.50 ± 0.90	2.495 *	D > A, D > B, D > F
BA	3.68 ± 0.93	3.73 ± 0.86	3.81 ± 0.84	4.04 ± 0.79	3.88 ± 0.82	3.55 ± 0.80	3.211 **	D > A, D > B, D > C, D > E, D > F
SN	3.54 ± 0.94	3.65 ± 0.98	3.79 ± 0.95	3.90 ± 0.95	3.51 ± 1.15	3.18 ± 1.25	2.966 *	D > B, C > F, D > E, D > F
PBC	3.27 ± 0.98	3.32 ± 1.06	3.56 ± 1.02	3.76 ± 0.96	3.48 ± 1.10	2.88 ± 1.20	5.002 ***	D > A, C > B, D > B, C > F, D > F

Note: *: p < 0.05, **: p < 0.01, ***: p < 0.001.

. Generally, the intention to use VF is an approximate normal distribution, with the highest value in the 31–40 aged group.

(3)

Education: The results show statistically significant differences among people with different educational levels (

Table 5. Comparison of intentions of groups with different education levels to use VF.

	Education (Mean ± SD)					F-Value	LSD
	Junior Middle School (B) (n = 21)	Senior Middle School (C) (n = 113)	Junior College (D) (n = 128)	Bachelor’s (E) (n = 340)	Master’s and Above (F) (n = 138)
BI	3.83 ± 0.75	3.86 ± 0.87	4.08 ± 0.73	3.95 ± 0.87	3.64 ± 0.95	5.032 **	D > C, C > F, D > F, E > F
BA	3.98 ± 0.81	3.87 ± 0.83	4.05 ± 0.68	3.88 ± 0.84	3.43 ± 0.90	10.95 ***	B > F, C > F, D > F, E > F
SN	3.88 ± 1.08	3.91 ± 0.92	4.01 ± 0.76	3.72 ± 1.00	3.32 ± 1.04	9.959 ***	B > F, C > F, D > E, D > F, E > F
PBC	3.56 ± 1.07	3.74 ± 0.84	3.84 ± 0.79	3.53 ± 1.06	2.87 ± 1.09	19.599 ***	B > F, C > E, C > F, D > E, D > F, E > F

Note: **: p < 0.01, ***: p < 0.001.

). The planting intentions of the group with a junior college degree, followed closely by the group with bachelor’s degrees, are slightly higher than the other groups.

(4)

Annual household income: The results (

Table 6. Comparison of intentions of groups with different annual household incomes to use VF.

	Annual Household Income (Mean ± SD) (CNY Ten Thousand)					F-Value	p-Value
	Below 3 (A) (n = 70)	3–8 (B) (n = 201)	8–30 (C) (n = 332)	30–100 (D) (n = 120)	Above 100 (E) (n = 17)
BI	3.74 ± 0.93	3.87 ± 0.94	3.95 ± 0.78	3.89 ± 0.93	3.94 ± 0.88	0.915	0.455
BA	3.67 ± 0.83	3.81 ± 0.94	3.84 ± 0.80	3.89 ± 0.80	4.04 ± 0.72	1.089	0.361
SN	3.63 ± 0.95	3.75 ± 1.03	3.74 ± 0.95	3.72 ± 1.07	3.85 ± 0.61	0.28	0.891
PBC	3.39 ± 0.99	3.46 ± 1.08	3.50 ± 0.99	3.56 ± 1.17	3.76 ± 0.80	0.636	0.637

) show that there is no significant difference among people with different annual household income levels.

(5)

Living area: The results show significant differences among people in different living areas (

Table 7. Comparison of intentions of groups in different living areas to use VF.

	Living Area (Mean ± SD)			F-Value	LSD
	Metropolis (A) (n = 224)	Ordinary Cities (B) (n = 400)	Towns and Villages (C) (n = 116)
BI	4.06 ± 0.77	3.86 ± 0.91	3.71 ± 0.83	7.181 **	A > B, A > C
BA	3.96 ± 0.69	3.78 ± 0.92	3.72 ± 0.83	4.561 *	A > B, A > C
SN	3.96 ± 0.85	3.66 ± 1.03	3.52 ± 0.98	10.276 ***	A > B, A > C
PBC	3.74 ± 0.91	3.41 ± 1.10	3.28 ± 0.97	10.273 ***	A > B, A > C

Note: *: p < 0.05, **: p < 0.01, ***: p < 0.001.

). People living in the metropolis show higher intentions than those living in ordinary cities, towns, and villages. This indicates that there is a positive relationship between the density of the living environment and people’s intentions to use VF.

4.2. Measurement Model Assessment Results

First, the internal consistency between the components in each latent variable is checked using Cronbach’s alpha (CA) and composite reliability (CR).

Table 8. Construct validities.

Constructs	Item	Loadings	Cronbach’s Alpha	AVE	CR
Perceived Usefulness	PU1	0.827	0.805	0.628	0.871
	PU2	0.768
	PU3	0.785
	PU4	0.789
Perceived Ease of Use	PEU1	0.905	0.764	0.809	0.895
	PEU2	0.894
Perceived Risk	PR1	0.714	0.841	0.646	0.879
	PR2	0.778
	PR3	0.886
	PR4	0.827
Peer Influence	PI1	0.856	0.836	0.753	0.902
	PI2	0.858
	PI3	0.889
Superior Influence	SI1	0.857	0.752	0.669	0.858
	SI2	0.840
	SI3	0.753
Self-Efficacy	SE1	0.819	0.866	0.714	0.909
	SE2	0.816
	SE3	0.882
	SE4	0.861
Facilitating Conditions	FC1	0.802	0.780	0.694	0.872
	FC2	0.876
	FC3	0.820
Behavior Attitude	BA1	0.848	0.734	0.653	0.850
	BA2	0.808
	BA3	0.767
Subjective Norms	SN1	0.928	0.835	0.858	0.924
	SN2	0.925
Perceived Behavioral Control	PBC1	0.910	0.862	0.785	0.916
	PBC2	0.900
	PBC3	0.846
Behavioral Intention	BI1	0.902	0.758	0.805	0.892
	BI2	0.892

shows that the CA values of all constructs are between 0.734 and 0.866, which are all higher than the minimum threshold of 0.70 proposed by Nunnally [98]. The CR values of all latent variables are between 0.85 and 0.924, which exceeds the minimum threshold of 0.70 and is lower than the maximum threshold of 0.95. The results from this work meet the minimum thresholds and do not exceed the highest threshold, indicating good convergent validity.

Next, the convergence validity of the questionnaire was checked. As shown in Table 3, the lowest indicator loading of 0.714, the lowest CR value of 0.85, and the lowest AVE value of 0.628 in this study satisfy the minimum threshold, indicating good convergent validity.

Finally, the discriminant validity of the questionnaire was checked. As shown in

Table 9. Fornell–Larcker criterion.

	FC	BA	BI	PBC	PEU	PI	PR	PU	SE	SI	SN
FC	0.833
BA	0.649	0.808
BI	0.584	0.698	0.897
PBC	0.656	0.651	0.663	0.886
PEU	0.456	0.404	0.356	0.456	0.9
PI	0.532	0.679	0.607	0.632	0.428	0.868
PR	0.092	−0.07	−0.031	0.004	0.151	−0.013	0.803
PU	0.369	0.514	0.42	0.383	0.419	0.451	−0.058	0.792
SE	0.702	0.636	0.562	0.79	0.512	0.607	0.045	0.363	0.845
SI	0.562	0.653	0.600	0.619	0.431	0.747	−0.027	0.473	0.645	0.818
SN	0.617	0.709	0.631	0.697	0.351	0.647	−0.031	0.396	0.667	0.668	0.926

Note: The diagonal bold is the radical value of AVE.

, the square root of the average variance extracted (AVE) for each latent variable is higher than the inter-construct correlation, providing sufficient evidence of discriminant validity for the constructs. In addition, the results of cross-loading are shown in

Table 10. Cross-loadings of measured items.

	FC	BA	BI	PBC	PEU	PI	PR	PU	SE	SI	SN
FC1	0.802	0.563	0.488	0.496	0.350	0.467	0.093	0.324	0.574	0.437	0.490
FC2	0.876	0.549	0.499	0.617	0.390	0.450	0.076	0.326	0.658	0.497	0.556
FC3	0.820	0.514	0.475	0.516	0.401	0.417	0.063	0.271	0.515	0.470	0.492
BA1	0.560	0.848	0.642	0.544	0.300	0.536	−0.068	0.436	0.503	0.526	0.605
BA2	0.518	0.808	0.568	0.495	0.265	0.538	−0.085	0.360	0.469	0.502	0.624
BA3	0.492	0.767	0.474	0.539	0.420	0.576	−0.016	0.450	0.574	0.557	0.488
BI1	0.520	0.625	0.902	0.628	0.322	0.559	−0.026	0.382	0.529	0.558	0.578
BI2	0.528	0.627	0.892	0.561	0.317	0.530	−0.030	0.373	0.478	0.518	0.554
PBC1	0.572	0.577	0.588	0.910	0.410	0.565	0.019	0.333	0.742	0.541	0.632
PBC2	0.616	0.557	0.565	0.900	0.439	0.546	0.015	0.319	0.745	0.548	0.615
PBC3	0.553	0.599	0.613	0.846	0.360	0.569	−0.026	0.368	0.606	0.558	0.604
PEU1	0.418	0.373	0.336	0.414	0.905	0.379	0.136	0.391	0.456	0.390	0.336
PEU2	0.403	0.353	0.304	0.406	0.894	0.392	0.135	0.362	0.465	0.385	0.294
PI1	0.471	0.596	0.528	0.589	0.410	0.856	−0.026	0.415	0.583	0.664	0.573
PI2	0.440	0.591	0.501	0.505	0.366	0.858	−0.014	0.377	0.499	0.638	0.540
PI3	0.474	0.581	0.551	0.548	0.338	0.889	0.007	0.381	0.498	0.642	0.572
PR1	0.141	−0.007	0.045	0.086	0.248	0.031	0.714	0.011	0.146	0.039	0.071
PR2	0.095	−0.041	0.009	0.056	0.179	0.038	0.778	−0.021	0.120	0.040	0.021
PR3	0.039	−0.076	−0.050	−0.056	0.097	−0.038	0.886	−0.052	−0.047	−0.082	−0.073
PR4	0.109	−0.054	−0.027	0.036	0.110	−0.014	0.827	−0.072	0.082	0.006	−0.009
PU1	0.319	0.422	0.345	0.331	0.326	0.360	−0.073	0.827	0.301	0.380	0.322
PU2	0.191	0.322	0.243	0.138	0.253	0.261	−0.026	0.768	0.115	0.283	0.205
PU3	0.234	0.378	0.347	0.255	0.272	0.327	−0.047	0.785	0.233	0.329	0.271
PU4	0.385	0.478	0.375	0.431	0.439	0.447	−0.035	0.789	0.437	0.470	0.416
SE1	0.542	0.502	0.457	0.620	0.428	0.495	0.029	0.279	0.819	0.539	0.526
SE2	0.555	0.503	0.442	0.596	0.424	0.486	0.072	0.301	0.816	0.520	0.540
SE3	0.617	0.604	0.522	0.726	0.431	0.566	0.007	0.343	0.882	0.579	0.617
SE4	0.652	0.534	0.474	0.715	0.450	0.502	0.050	0.301	0.861	0.541	0.566
SI1	0.471	0.583	0.516	0.565	0.390	0.659	−0.032	0.405	0.593	0.857	0.594
SI2	0.496	0.559	0.525	0.531	0.365	0.665	−0.044	0.384	0.554	0.840	0.568
SI3	0.408	0.448	0.424	0.408	0.295	0.492	0.016	0.373	0.419	0.753	0.465
SN1	0.551	0.653	0.594	0.646	0.306	0.602	−0.054	0.366	0.583	0.616	0.928
SN2	0.593	0.660	0.574	0.644	0.345	0.598	−0.004	0.368	0.653	0.621	0.925

. All the loadings of the indicators are consistently highest on the construct with which they are associated. Therefore, the evaluation of discriminant validity among the constructs satisfied the requirements.

In summary, the evaluations of the measurement model regarding internal consistency reliability, convergent validity, and discriminant validity all met their requirements, empirically validating the suitability of the measurement model in this study.

4.3. Structural Model Evaluation Results

After verifying the reliability and validity of the measurement model, the next step was to analyze the structural model to test the hypothetical relationship. First, to ensure that there was no multicollinearity problem, the VIF of the model was checked. The results show that the tolerance level in the measurement model construction is between 1.338 and 2.697, which is lower than the critical level of 3.3, indicating that no multicollinearity problem exists.

Next, the model’s goodness-of-fit was checked. The results show that the values of the selected model’s fit indicators basically satisfy the recommended values. For example, the SRMR = 0.058, which is lower than the threshold value of 0.08, and the NFI = 0.778, which is higher than the threshold value of 0.7. In addition, the d-ULS = 1.897 and the d-G = 0.714, which indicate that the goodness of fit of the structural model meets the requirements.

The results of the assessment of the prediction correlation are shown in Figure 5. The R² value of BA is 0.314 (>0.19), which indicates an average prediction accuracy. Considering the high number of potential variables of BA, its prediction accuracy is acceptable; the R² values of SN, PBC, and BI range from 0.496 to 0.644 (>0.33), which indicates a high level of prediction accuracy. Another method for assessing the predictive relevance is Q² and the results are shown in

Table 11. Evaluation of predictive accuracy and predictive relevance.

Constructs	R2	SSO	SSE	Q2 (=1 − SSE/SSO)
BA	0.314	2220	1776.276	0.2
SN	0.496	1480	858.223	0.42
PBC	0.644	2220	1108.75	0.501
BI	0.569	1480	811.811	0.451

, where the Q² values of all the constructs are greater than zero, which indicates that the model proposed in this study has sufficient predictive relevance.

The final step in the structural model evaluation is to evaluate the path relationships.

Table 12. Results of direct effects among constructs.

Path	Hypothesis	β	SE	t-Value	p-Value	Results
PU→BA	H1a	0.406	0.038	10.765	<0.001	Supported
PEU→BA	H1b	0.247	0.037	6.702	<0.001	Supported
PR→BA	H1c	−0.084	0.04	2.067	0.039	Supported
PI→SN	H2a	0.336	0.051	6.611	<0.001	Supported
SI→SN	H2b	0.416	0.044	9.565	<0.001	Supported
SE→PBC	H3a	0.651	0.036	18.082	<0.001	Supported
FC→PBC	H3b	0.199	0.038	5.171	<0.001	Supported
BA→BI	H1	0.405	0.043	9.334	<0.001	Supported
SN→BI	H2	0.127	0.044	2.907	0.004	Supported
PBC→BI	H3	0.311	0.046	6.812	<0.001	Supported

Note: β = Path coefficient.

shows that all the proposed hypotheses presenting direct relationships among the constructs were empirically supported, with all t-values being more than 2.067 at the significance level of 0.05. The results indicate that the exogenous variables have strong relationships with the endogenous variables, i.e., the paths were statistically significant and support the model.

5. Discussions

5.1. Conclusion of the Data Analysis

In terms of the descriptive analysis, the general public’s overall intention to use VF is generally above ‘neutral’ but slightly below ‘positive’. At the same time, differences in the age, education level, and living area of the public significantly affect their intentions to use VF. However, there are no significant differences in the intentions to use VF among people of different genders and annual household incomes. Overall intention is the highest among those aged 31–40 with a college degree and living in metropolitan areas. Among them, intention is approximately normally distributed with age and education level, and the middle-aged and middle-educated groups have the strongest intentions. Urban residents show higher enthusiasm toward VF than those living in the countryside. At the same time, the results of this part of the analysis show that perceived behavioral control, that is, people’s assessments of their own abilities, are the key factor limiting their intentions. In this survey, 77.7% of the 4065 returned questionnaires had heard of the concept of vertical farming. However, only 789 questionnaires passed the preliminary test. This suggests that people’s knowledge of VF is not rich enough to play a key role in their planting decisions. Better publicity and popularization of VF could help to improve public acceptance and willingness to use VF.

In terms of the structural equation model analysis, the results show the applicability of the DTPB in research into the intention to use VF. The evaluations of the measurement model and the structural model showed the theoretical adequacy of the hypothetical model. In the proposed model, BI has three direct predictors: behavioral attitude (BA), subjective norm (SN), and perceived behavioral control (PBC). The results show that all these predictors have a significant impact on BI with an R² value of 0.569. R² represents the prediction accuracy of the model, and higher values indicate a higher prediction accuracy. As mentioned above, the values 0.19, 0.33, and 0.67 represent the three levels of low, medium, and high prediction accuracies, respectively. This means that BA, SN, and PBC have 56.9% explanatory power for BI with an above-medium prediction accuracy [78]. The results of the path analysis showed that BA (β = 0.405), SN (β = 0.127), and PBC (β = 0.311) have significant effects on behavioral intention. Among them, BA has the most important influence on BI, followed by PBC, whereas SN has a weaker effect.

In the BA dimension, perceived usefulness (PU) is the strongest factors in determining BA (β = 0.406), followed by perceived ease of use (PEU) (β = 0.247) and perceived risk (PR) (β = −0.084). This indicates that people think that VF has more advantages than disadvantages for daily life and that they are not worried about the side effects of VF. For PU, the path coefficients of the two variables PU1 and PU4 representing the economic benefits are 0.827 and 0.789, respectively, which are higher than the path coefficients of PU2 (0.768) and PU3 (0.785) representing the environmental benefits. This suggests that compared with the environmental benefits, the economic benefits of VF are more attractive. When planting behavior can bring more economic benefits, people’s planting intentions can be improved. From this perspective, although micro-VF in mixed-use spaces could bring a lot of added value, how to improve the yield and reduce the costs are still the first things that people will consider. In addition, PEU also has a significant impact on people’s attitudes. Simplifying the planting process by equipping it with intelligent monitoring and control systems and even developing VF using an automatic mode in the future, could attract more people to participate in VF activities.

In the SN dimension, the two factors of SN, superior influence (SI) and peer influence (PI), explained 49.6% of the variance in SN, which is statistically significant. The path analysis results for each of the determinants of SN revealed that both SI (β = 0.336) and PI (β = 0.416) have a significant positive effect on SN. This indicates that public acceptance of VF is influenced by the opinions of the people around them and that the opinions of superiors are more influential for them to make decisions. In PI, the path coefficients reflecting the importance of the opinions of family members (PI1), colleagues (PI2), and friends (PI3) are all greater than 0.85, indicating that the public has a certain herd mentality when making decisions Similarly, in the influence of superiors, the results of the path analysis showed that both elders (SI1, β = 0.857) and leaders (SI2, β = 0.840) had a significant influence on the decision-making behavior of the public.

Finally, in the PBC dimension, the two factors, facilitating conditions (FC) and self-efficacy (SE), were statistically significant (64.4%). The path analysis results show that SE (β = 0.651) has a greater influence than FC (β = 0.199). It shows that in the decision-making process, the public are more obviously affected by their own knowledge and skills related to VF. In terms of FC, the least trusted was the intelligent control capability of VF (FC1, β = 0.802). This indicates that the public considers that the intelligent control system of VF is still not reliable and user friendly enough for them. Further studies and technical improvements to the control system are needed.

5.2. Comparison with Previous Related Studies

It may be a huge challenge for urban VF extension research to predict public acceptance and willingness to use VF, and few substantive attempts have been made to date. The most relevant study is by Jürkenbeck et al. [99], which used TAM to investigate consumer acceptance of the three VF systems. The results indicated that perceived sustainability is the major driver of German consumers’ acceptance of VF systems and that the “indoor vertical farm” is the most acceptable system for people. In contrast, in this study, perceived usefulness and facilitating conditions were the most important driving forces affecting Chinese consumers. Grebitus et al. [100] investigated the likelihood of consumers buying or growing food on urban farms. The study showed that subjective knowledge about urban agriculture and overall favorable attitudes toward urban farms increased the possibility that consumers would buy and grow produce on urban farms. In addition, women and older consumers are more likely to grow on urban farms. In this study, gender has little effect on acceptance and people aged 31–40 were more inclined to plant using VF. Suárez-Cáceres et al. [101] studied the knowledge, attitudes, and willingness to pay for aquaponics products among consumers in Spain and Latin America. The study showed that the quality, taste, and absence of pesticides or chemical residues of the products for aquaponics systems were the main purchasing motivations for consumers. In contrast, household income, concern for the environment, and a priori knowledge of aquaponics had a significant effect on the willingness to pay higher prices. However, in this study, the effect of household annual income on VF intention is not significant. At present, consumers from different countries and regions have different understandings, views, and concerns about VF. Future studies on this issue are also important for the worldwide research and development of VF.

5.3. Limitations and Future Research Direction

This study is a new attempt in the research of VF and there are still some limitations in the following aspects. First, the survey was conducted within mainland China; in comparison, consumer behavior may vary across countries and regions. Future researches need to be expanded to investigate aspects specific to other geographical locations and cultural regions to gain more insight into this topic. Second, this study used the behavior-based DTPB model to predict public acceptance and willingness to use VF. Apart from the DTPB model, there are other higher-order models, such as the Task-Technology Fit (TTF) [102] or the Unified Theory of Acceptance and Use of Technology (UTAUT) [103], to be explored. In addition, research on the driving forces of consumers to buy food is also of great significance. Research showed that local food production, naturalness, ethics, and contextual factors, e.g., price and availability, are important aspects that consumers consider when purchasing food [104]. Flores et al. [105] reported that consumers are more willing to pay for locally produced food than for products that are produced further away or do not have any clear origin markings. Better quality and taste, safety, and easier traceability are important driving factors for people in choosing to purchase local food. This study does not address these factors and future research could investigate these issues.

6. Conclusions

In this paper, public acceptance and willingness to plant using VF were investigated using the DTPB model. This study selected behavioral intention (BI) to assess to public willingness to use VF and hypothesized that behavioral attitude (BA), subjective norm (SN), and perceived behavioral control (PBC) directly influence BI. Then BA, SN, and PBC were deconstructed separately and it was hypothesized that the deconstructed factors have indirect effects on BI. Finally, the proposed hypotheses were tested by evaluating the measurement model and structural model using PLS-SEM as the data analysis method. The key findings are given as follows:

• The level of the public’s attitudes and behavioral intentions toward micro-VF is better than “neutral” but slightly less than “positive”.
• The results of the structural model evaluation showed that the proposed original hypotheses were all supported, that is, that the factors that influence public acceptance and willingness to plant using VF are behavioral attitude (perceived usefulness, perceived ease of use, and perceived risk), subjective norm (peer influence, superior influence), and perceived behavioral control (self-Efficacy, facilitating conditions). Among them, the most critical factors are the cost-effectiveness of VF and the consumption of time and money required for the planting process.
• The biggest motivation for the public to engage in micro-VF is the possible economic benefits (β = 0.406). On the contrary, the biggest obstacle is the general lack of knowledge about VF and the unfamiliarity with its growing process and other expertise.
• Further improving the cost-effectiveness of VF, simplifying the growing process, and enhancing the dissemination and popularization of professional knowledge are important research directions for the subsequent promotion of VF.