Tanzanian Farmers’ Intention to Adopt Improved Maize Technology: Analyzing Influencing Factors Using SEM and fsQCA Methods

Abstract

The China–Tanzania Village-based Learning Center for Poverty Reduction project aims to demonstrate China’s experience in poverty reduction through developing smallholder agriculture at the village level, and through the promotion of improved technologies with the purpose of increasing agricultural productivity and improving village-level agricultural production. However, to promote technology application effectively, a better understanding of farmers’ behavioral intention toward improved maize technology is needed. This study uses microdata from 282 Tanzanian maize farmers. Compared with previous research, the innovation of our study is that the methods of structural equation model (SEM) and fuzzy-set qualitative comparative analysis (fsQCA) are applied to study the influencing factors of farmers’ intention to adopt improved maize technology and the combination paths that affect farmers’ intention. The analysis results show that farmers’ behavior perceptions and compatibility perceptions positively impact their intention to choose agricultural technology. Three modes can trigger farmers’ intention to adopt technology: “behavior perception, compatibility perception, non-self-efficacy, facilitation conditions,” “behavior perception, social impact, non-self-efficacy, facilitation conditions,” and “compatibility perception, social impact, self-efficacy, facilitation conditions.” To increase maize yield and promote the extension of improved agricultural technology through the China–Tanzania Village-based Learning Center for Poverty Reduction project, it is necessary to choose an effective combination path to influence farmers’ intention to adopt the proposed changes. If farmers’ intention to adopt improved maize technology can be increased to a greater extent, it can accelerate the improvement of agricultural technology in Tanzania, thereby increasing agricultural productivity, improving agricultural production at the village level, and reducing poverty.

1. Introduction

Agriculture is an essential economic pillar of the United Republic of Tanzania and a key driver of its rural development. Agriculture employs 78% of Tanzania’s population, provides approximately 95% of the country’s food, ensures livelihood for more than 70% of the population, and makes up half of the gross domestic product (GDP) and export income for the country [1]. Despite this, the malnutrition rate in Tanzania is still very high. Food security in Tanzania remains an urgent problem to be solved as more than 34% of children under five years old are stunted, and nearly 45% of women of childbearing age are anemic [2].

Tanzania has always had an important partnership with China regarding agricultural cooperation in Africa. China’s assistance to Tanzania’s agricultural and rural development can be traced back to the 1960s [3]. The first Forum on China-Africa Cooperation (FOCAC) was held in Beijing in October 2000, marking the beginning of a framework for Sino-African engagement in the 21st century [4]. The China–Tanzania Village-based Learning Center for Poverty Reduction project was founded in 2009 [5]. The project is co-managed by the International Poverty Reduction Center in China (IPRCC), and the President’s Office, Planning Commission of the United Republic of Tanzania. Peapea village is responsible for specific implementation, and technical support is provided by the Research Center for International Development at China Agricultural University and Sisal Farm of the China–Africa Agricultural Investment Co. [5]. The project aims to promote China’s experience in poverty reduction through the development of village-level smallholder agriculture, especially promoting the use of labor-intensive technologies to improve agricultural productivity at village level [5,6].

Maize is the main food crop in most of Tanzania [3]. According to the Food and Agriculture Organisation (FAO) data, Tanzania’s total cereal production in 2020 was 12.50 million tons, of which maize production was 6.71 million tons, accounting for 53.72% of the total cereal production. Maize growth has many requirements regarding altitude, temperature, and precipitation, so while it is widely planted in Tanzania, the yield is low [3]. According to the FAO data in 2020, the unit yield of maize in Tanzania was 1.81 tons per hectare, while the unit yield of maize in China reached 6.32 tons per hectare in the same year. The unit yield of 1.81 tons per hectare is the level of China’s maize yield in the 1960s, indicating that there is a significant technical gap between Tanzania’s maize planting technology and that of China, and improved maize technology has great reference significance for the development of Tanzania’s technology. Peapea Village in Rudewa Town, Kilosa District, Morogoro Region, Tanzania, where the Village-based Learning Center is located, is a major maize-growing area. The previous experience of agricultural technology transfer indicates that the transfer of agricultural technology must meet local needs and be consistent with local agricultural development [7,8]. Therefore, the Village-based Learning Center for Poverty Reduction project mainly adopts the concept of “small technology, big harvest” and promotes the selection of improved varieties, reasonable dense planting, maize and pigeon pea interplanting technology, and other simple and easy-to-use improved technologies. This is done to adopt improved agricultural technology to drive local development, promote local people’s leading role and self-development abilities, and encourage local farmers to increase production and income [5]. Currently, the Chinese embassy in Tanzania, with the support of the local government, will gradually promote the village-level poverty reduction model by sharing labor-intensive and applicable improved technologies in parallel with 10 villages in the Morogoro Province [9]. Therefore, from the perspective of continuing development, improving local agricultural technology, and improving production efficiency, it is necessary to analyze the factors and the combination paths that affect farmers’ intention to choose improved agricultural technology in maize production.

Research on the adoption of maize technology by farmers in Tanzania has mainly focused, firstly, on whether the production endowment of farmers, such as farm scales, farmers’ education levels, farmers’ income, experience of producers, age of producers, and family size, has an impact on the adoption of a series of new maize technologies, including improved maize seeds and pest control [10,11,12,13]. Tanzanian farmers with more fertile land are more likely to adopt new and improved maize varieties [14]. Second, from the perspective of economic, social, and environmental factors, scholars have found that the adoption of new maize production technologies by Tanzanian farmers is affected by the economic benefits of adopting the technology, visits of extension personnel, links between research and extension, and rural credit systems and information dissemination [13,15,16,17,18,19]. For example, Magrini et al. [17] found that two technologies, improved seeds and inorganic fertilizers, had a positive and significant impact on maize farmers by increasing maize productivity. The use of hybrid maize seeds can increase both maize yield per acre and total maize production, while increased yields and profits correspond to high adoption rates of hybrid maize seeds [18]. Asset holdings, access to extension, access to information, and community participation can significantly determine the adoption of improved maize varieties by farmers in Tanzania [19]. From the perspective of farmers’ technology cognition, Jha et al. [20] argued that the adoption of improved agricultural techniques by smallholder farmers in Tanzania can be effectively explained by integrating variables related to individual, household, socioeconomic, and farmer perceptions. Doss et al. [21] have proposed that maize breeding research should be closer to farmers’ preferences. Furthermore, Midega et al. [22] used semi-structured questionnaires to assess farmers’ perceptions of technology benefits, which were used to evaluate farmers’ further adoption of maize technology for the potential control of weeds and pests.

The study of improved agricultural technology demonstrated by China in Africa has primarily focused on agricultural technology experts and their personal abilities and technology transfer intentions through interviews or questionnaires [8,23,24,25]. Regarding the technology adoption of African farmers, mainly in macro-technical cooperation research and case studies, it is believed that China’s agricultural technology can increase crop yield, improve economic effects, and has advantages over local methods of pest resistance [26,27,28,29,30]. However, Makundi [31] posits that African farmers’ intention to adopt improved agricultural technology demonstrated by China depends not only on whether the crop yield increases but also other factors. Other studies have shown that agricultural technology experts in agricultural technology demonstration centers lack interaction and communication with farmers participating in training and contact with other local farmers [24,25]. Moreover, there is no clear plan for interaction and communication between these demonstration centers and local farmers, which affects farmers’ intentions to adopt new technology and strategies [6,32].

However, the existing literature has some shortcomings. First, there is a lack of research on farmers’ intention to adopt new maize technologies in Tanzania. The farmers’ behavioral intention is a psychological, cognitive process that should be further investigated [33]. Second, the literature regarding the agricultural technology transfer between China and Tanzania lacks an analysis of farmers’ intention to adopt improved maize technology and the key influencing factors thereof based on micro-farmer household survey data. Third, the existing quantitative analyses in the literature mostly focus on the linear relationship between the influencing factors and the farmer’s intention; however, the “linkage effect” between the factors is often ignored [34].

Our study attempts to remedy these shortcomings. First, we build a theoretical framework to analyze farmers’ intention to adopt improved maize technology and investigate the factors influencing this intention from three dimensions: behavior attitude, subjective norms, and perceived behavioral control. Second, through field research, we collect microdata on maize farmers in Tanzania. Here, we reveal the intentions at the individual farmer level to study the factors that affect farmers’ intention to adopt improved maize technology. Third, to analyze the complex causal relationship formed by the interdependence of multiple antecedent conditions, various combinations of influencing factors of farmers’ intention to adopt improved maize technology are studied through a fuzzy-set qualitative comparative analysis (fsQCA) method.

2. Methods and Materials

2.1. Theory of Planned Behavior (TPB) and Structural Equation Model (SEM)

The theory of planned behavior (TPB) is suggested based on the theory of reasoned action (TRA), which proposes that behavior intention directly determines behavior and is influenced by behavior attitude and subjective norms [35]. The TRA assumes that individual behavior is controlled by will; however, Ajzen [36] holds that human behavior is not completely voluntary, but rather under control, which restricts the wide application of the TRA. Based on the TRA, Ajzen [36] proposes the TPB by adding perceptual behavior control variables to the model. The TPB postulates three conceptually independent determinants of intention, which are attitude toward the behavior, subjective norm and perceived behavioral control [36]. Specifically, attitude toward the behavior refers to an individual’s favorable or unfavorable evaluation, cognitions or expectations of performing a certain behavior. The social pressure perceived by an individual when performing a certain behavior is called a subjective norm [36,37]. Perceived behavioral control refers to the perceived ease or difficulty of performing a certain behavior, which reflects the past experience as well as anticipated impediments and obstacles that an individual has overcome to perform a certain behavior, such as constraints of money, time, and skills [36,38].

The TPB has obvious advantages in explaining people’s behavior motivations and intentions and has been widely used in the field of social science research [39]. Tanzanian farmers’ intention to adopt improved agricultural technology essentially reflects the cognition of technology, influence of social groups, and cognition of control ability. To promote improved agricultural technology in Tanzania, it is of practical significance to study farmers’ intentions and behaviors toward adopting such technology. An analysis based on the TPB can explain the determinants of farmers’ adoption of improved maize technology.

This study constructs a theoretical model based on the above theoretical basis and research hypothesis (see Figure 1). First, according to the TPB, we divide the determinants of farmers’ intention to adopt improved maize technology into three dimensions: behavior attitude, subjective norm and perceived behavioral control. Firstly, there are two influencing factors in the dimension of behavior attitude, which are behavior perception and compatibility perception [39,40]. We aim to study the influencing factors of adoption intention on improved maize technology by evaluating farmers’ behavior perception and compatibility perception. Second, under the dimension of subjective norm, social impact is set as an influencing factor [40], which is used to study the perceived social pressure to adopt or not to adopt the improved maize technology. Third, the dimension of perceptual behavioral control includes two influencing factors, namely self-efficacy and facilitation conditions [39,40,41]. We aim to study the experience as well as anticipated impediments affecting farmers’ adoption intention from the perspective of self-efficacy and facilitation conditions.

Based on TPB and our theoretical model, we choose to use the Structural Equation Model (SEM) approach. SEM analysis can comprehensively consider multiple dependent variables, estimate the structure and relationship of latent variables simultaneously, and analyze the action path of latent variables and observation variables in a system [42]. Furthermore, SEM analysis can reflect the influence effect of latent variables and observation variables on a system and allow measurement errors in parameter estimation [42]. Compared with methods such as multiple regression, path analysis and simultaneous equations in econometrics, SEM has unique advantages [33,43]. SEM does not have strict assumptions and constraints, and allows maximum error between exogenous variables and endogenous variables, providing the possibility to analyze the structural relationship between latent variables, which is its greatest advantage [33,43]. Our analysis of farmers’ intention to adopt improved maize technology is mainly based on the latent variables determined by TPB. These latent variables are not directly observed but inferred from other observed variables. Therefore, SEM is more advantageous in our study.

SEM includes two parts: a measurement model and a structural model. A measurement model depicts the relationship between latent and observation variables, whereas a structural model refers to the relationship between latent variables [44]. This study collected the observation variables through scale design and a questionnaire survey, reflecting the latent variables that cannot be observed directly through the data of observed variables [42,44].

The matrix equation form of the specific SEM is as follows [33,40]:

Measurement model:

$$Y={\mathsf{\Lambda }}_y\eta +ϵ$$

(1)

$$X={\mathsf{\Lambda }}_x\xi +\sigma$$

(2)

Structural model:

$$\eta =b\eta +\gamma \xi +\zeta$$

(3)

Here, $Y$ denotes the observation variables of endogenous latent variables. $X$ is the observation variable of exogenous latent variables. ${\mathsf{\Lambda }}_y$ and ${\mathsf{\Lambda }}_x$ are the correlation coefficient matrices of the exogenous latent variables and endogenous latent variables with their respective observed variables, respectively. $\eta$ is an endogenous latent variable, $\xi$ denotes exogenous latent variables, and $b$ and $\gamma$ are the coefficient matrices of corresponding variables. $ϵ$, $\sigma$ and $\zeta$ are error variables.

The influencing factors of farmers’ intention to adopt improved maize technology that we selected based on TPB are: behavior perception, compatibility perception, social impact, self-efficacy, facilitation conditions and behavior intention. Among them, behavior intention is an endogenous latent variable, while the others are exogenous latent variables. The analysis of these latent variables is based on the observation questions in the farmer field survey questionnaire.

2.2. Configuration Theory and Fuzzy-Set Qualitative Comparative Analysis (fsQCA)

The generation of configuration analysis perspectives and methods stems from scholars’ interest in configuration problems and the limitations of traditional linear regression analysis methods in solving these problems [45]. Configuration theory is based on causal complexity, arguing that the concurrency of first-order elements and the formation of various higher-order configurations may have equivalent effects on an outcome [46,47]. Configuration studies are usually exploratory in that they do not assume (and often cannot foresee) that certain configurations are necessarily better unless theory provides sufficient inferences [47]. The configuration perspective is based on holism, which means that one thing cannot be analyzed and explained in isolation, and requires analysis of the impact of the combination of condition variables on the outcome variables, emphasizing the causal complexity between variables [48]. Configuration theory focuses on how multiple conditional variables are related to each other and affect the results in a specific environment, and does not study the net effect of each conditional variable separately [49]. According to configuration theory, only by studying the combination of multiple conditional variables can we better explore the commonness or characteristics of the combination [50]. Configuration theory has three distinctive features: equivalence, asymmetry, and multiple concurrency [51,52]. Equivalence means that different condition variables can be combined in a certain way to produce the same result in the end [51,52]. Asymmetry means that the reasons for the appearance and non-appearance of the results are different; hence, the existence of a variable may lead to a certain result, but the absence of this variable may not lead to the loss or negation of the results [51,52]. Multiple concurrency refers to the influence of a conditional variable on the results, which depends on its interaction with other conditional variables in different configurations [46].

An epistemologically and ontologically derived approach is the Qualitative Comparative Analysis (QCA) method developed by the sociologist Ragin [53], which is presented as a third way between case-study methodology and quantitative statistical techniques. Through comparative analysis, each case is classified into a set of cause conditions, and the relationship between case-level configuration and outcome variables is analyzed [47,53,54]. The core of the QCA method is causal interpretation, visualization, and analysis of causal complexity as well as principles of logical minimization [55]. The basic elements of the new configuration perspective are summarized into four aspects: (1) the case is regarded as a configuration of the set theory; (2) calibration and ensemble membership analysis; (3) sub-set relationship analysis of sufficient and necessary conditions; and (4) counterfactual analysis [56]. The basic intuition underlying QCA is that cases are best understood as configurations of attributes resembling overall types and that comparing cases can allow a researcher to remove attributes unrelated to the outcome in question [57]. It assumes that the independent variables are interdependent and work together, and there is no optimal equilibrium state; conversely, QCA posits that there are multiple equivalent paths or solutions [47,53,54,57]. QCA was initially used to analyze and intermediate number of cases (5–100) [47], but it has also been used more in large sample studies in recent years [34].

There are three main types of QCA methods for now: Crisp-set QCA (csQCA), Multi-value QCA (mvQCA) and Fuzzy-set QCA (fsQCA). CsQCA is the earliest research method and the basis of other analysis technologies. Its main feature is data dichotomy; that is, if a condition exists, the code is 1, and if a condition is missing, the code is 0 [55]. MvQCA is an extension of csQCA, but both methods lead to some missing information [50]. Ragin [47] then introduced the concept of membership into QCA and proposed a qualitative comparative analysis method based on fuzzy sets. FsQCA sets the membership score of the conditional variable [47], which effectively avoids the absolute limitation of binary data. FsQCA is uniquely suited for analyzing causal processes in typologies because it is based on a configurational understanding of how factors combine to generate outcomes and can handle significant levels of causal and interaction relationships [47,58]. This approach is an analytical technique based on set theory that allows for a detailed analysis of how causal conditions affect the outcome in question [57]. Due to the qualitative comparative analysis tool of fsQCA produced by Boolean algebra and set analysis, fsQCA can identify the configuration relationships between various factors as well as the various paths that achieve a specific goal [57,58]. This greatly improves the matching of configuration analysis theory and methods. Due to the respective advantages of qualitative and quantitative analysis, fsQCA has been widely used in various fields of management disciplines in recent years [46]. Therefore, we choose the method of fsQCA for our study, and the key steps are as follows:

(1) Selection of the research cases and specification of research variables [34,50,55]. The case in our study is a sample of farmers surveyed by field questionnaires. Determined variables refer to the determination of condition variables and result variables [54]. The conditional variables are the variables selected according to TPB in Section 2.1: behavior perception, compatibility perception, social impact, self-efficacy and facilitation conditions. The aim of our study is to find the influencing factors of farmers’ adoption of improved maize technology, so the result variable is behavior intention (see Figure 1). (2) Calibration of the variables. Calibration variables mainly refer to the transformation of data into fuzzy-sets [55]. FsQCA divides the data according to the membership degree, and uses the continuous interval of to represent the membership degree of variables [47]. There are mainly three membership degrees: complete membership (fully in), incomplete membership and complete non membership (fully out) [47,59]. (3) Analysis for necessary and sufficient conditions [55]. The main purpose of this process is to judge the interpretation of the final configuration in relation to the result variables, and to analyze whether the condition variables in the final configuration are the core conditions according to the results of necessity analysis [60,61]. The interpretation of the results mainly depends on consistency and coverage [50]:

$$Consistency\left(X_i\le Y_i\right)=\sum \left[min\left(X_i,Y_i\right)\right]/\sum \left(X_i\right)$$

(4)

$$Coverage\left(X_i\le Y_i\right)=\sum \left[min\left(X_i,Y_i\right)\right]/\sum \left(Y_i\right)$$

(5)

$X_i$ is the calibrated value of the condition variables, and $Y_i$ is the calibrated value of the result variable.

(4) Boolean simplification to form a truth table. Information for a case is displayed in a truth table, where each row shows one of the 2^k logically possible combinations of variables (k), and the observed value describing the case for one of the 2^k variables [62]. The truth table shows combinations of conditions for which a particular outcome occurs or does not occur, and cases with the same conditions and outcomes are presented in the same row of the truth table and are analytically identical [55]. (5) Finally, the combination of conditions where the result variable occurs or does not occur is obtained, which is the configuration of the result variable [45,50]. Our study ultimately wants to find the configuration that triggers farmers’ intention to adopt improved maize technology, that is, the combination of selected condition variables.

Based on the above theories and methods, the theoretical framework finally established is shown in Figure 1. Many studies involving farmers’ behavioral intentions have adopted SEM to study the relationship between latent variables [33,38,39]. SEM analysis focuses on the net effect of a single factor on the results, that is, the impact of a single variable on the results, thus ignoring the correlation between the variables [34,50]. Therefore, we choose to use fsQCA to analyze whether farmers’ intention to adopt improved maize technology is affected by the combination of multiple factors. We expect to analyze the single influencing factor and combinations of influencing factors that affect farmers’ intention to adoption.

2.3. Research Hypotheses

Behavioral attitude refers to the degree to which the behavior subject likes or dislikes the implementation of specific behavior, reflecting the tendency of the behavior [63]. In the relationship between attitude and behavioral intention, attitude is regarded as a very important determinant, and farmers’ attitudes toward farmland options can be reflected by expected returns [33,36,41]. Regardless of whether the new technology that potential adopters are exposed to is simple to operate and easy to learn, or rather cumbersome to operate, it requires a lot of energy and time to understand and evaluate the ease of use of new technologies [64]. After learning information regarding improved maize technology, farmers will first make a preliminary judgment from their own perceptions [41]. When they perceive that adopting the improved maize technology can improve their maize yield, quality, or income or that the technology itself is easy to learn and apply, farmers will hold a more positive attitude toward adopting the technology [65]. Additionally, as a foreign technology, the local adaptation degree, that is, farmers’ perceptions of the compatibility between improved maize technology and traditional maize technology, as well as their own values, practical experience, and environment, will affect their intention to choose such technology [39]. Therefore, this study proposes the following hypotheses:

Hypothesis 1 (H1).

Behavior perception positively impacts farmers’ intention to adopt improved maize technology.

Hypothesis 2 (H2).

Compatibility perception positively impacts farmers’ intention to adopt improved maize technology.

Subjective norms refer to the social pressure individuals perceive when deciding whether to take a certain behavior, which reflects a person’s degree of restraint on their behavior decisions [36,66,67]. The motivation for farmers’ adoption of agricultural technology comes from the psychological process of internalization or identification, which is integrated into their cognitive system, and then internalized into their views on agricultural technology [68,69]. Farmers can refer to the effects of other farmers’ adoption of agricultural technology. If the effect is obvious, farmers tend to identify with agricultural technology [68,69]. Improved maize technology is still in its infancy in Tanzania, and local farmers do not know much about the technology, so their intention to adopt will be influenced by the views or practices of surrounding groups [40]. When farmers are uncertain about the pros and cons of adopting improved maize technology, the opinions of friends and neighbors and the importance attached by rural organizations to the training and popularization of such technology will affect farmers’ decision-making [39]. The more positive the social impact is, the greater the social pressure farmers perceive [64]. Based on this, this study proposes the following hypothesis:

Hypothesis 3 (H3).

Social impact positively affects farmers’ intention to adopt improved maize technology.

Farmers’ perceived behavioral control is mainly affected by their self-efficacy and resource convenience. The self-confidence and resource convenience of somatosensory mastery of new technologies reflect the degree of resource constraints of individuals’ perceived mastery of new technologies [70]. The more resources farmers have and the fewer obstacles they expect, the stronger their perceptual and behavioral control over the adoption of technology, and the stronger their behavioral intention [39]. Self-efficacy belongs to the internal control category of behavior control, whereas convenience belongs to the external control category [66]. The higher a farmer’s self-recognition, that is, thinking they have enough knowledge and skills to learn and apply improved maize technology and having the confidence to solve problems encountered in the adoption process, the higher their self-efficacy [38,40]. Facilitation conditions encompass mainly whether the technology extension department can provide sufficient conditions to supply the necessary support to farmers [71]. The more favorable the external conditions, the more positive the impact on farmers’ decision-making regarding improved maize technology adoption [33,39]. Simultaneously, the more favorable the external facilitation conditions are, the more farmers’ positive feelings regarding technology adoption behavior control are stimulated, and consequently, the stronger the force of perceived behavioral control on the outcome variables [33,39]. Therefore, this study proposes the following hypotheses:

Hypothesis 4 (H4).

Self-efficacy positively impacts farmers’ intention to adopt improved maize technology.

Hypothesis 5 (H5).

Facilitation conditions positively impact farmers’ intention to adopt improved maize technology.

2.4. Setting

The China–Tanzania Village-based Learning Center for Poverty Reduction project is located in Peapea Village, Rudewa Town, in the Kilosa District, Morogoro Region; the distance from Peapea Village to Morogoro Town is 120 km. Peapea Village comprises four sub-villages (or natural villages) and has no irrigation facilities, thus the agricultural production is completely dependent on rainwater and the main crop is maize; consequently, the yield per unit area is very low [5]. The project was planned in 2009 and constructed in 2011. The maize technology training and demonstrations began in 2012, and the number of demonstration households increased from 10 in 2012 to 53 in 2015 [5]. In the village, 75% of farmers have adopted improved technology to grow maize, and their yields have increased by 50% to 180% [9]. By 2021, the China–Tanzania Village-based Learning Center, China-Tanzania Agricultural Development Joint Research Center, and “Thousand-House Ten Thousand-mu Maize Yield Increase Demonstration Project” jointly shared labor-intensive and applicable technologies in 10 villages in the Morogoro province to gradually promote the village-level poverty reduction model [72]. After investigation, the Village-based Learning Center found that technologies, such as irrigated agriculture and mechanized farms are not suitable for local development [5,73]. For example, few local farmers can afford the cost of tractors, which are also inconvenient to maintain after use, and many cannot afford pesticides and fertilizers. Therefore, local farmers need simple and easy-to-use agricultural technologies, such as improved seed selection, reasonable dense planting, and maize and pigeon pea intercropping technologies [5,9].

At the beginning of the construction of the Village-based Learning Center in 2011, improved agricultural experts found through field research that one of the main reasons for the low maize yield in Tanzania was insufficient planting density [5,73]. About 15,000 plants were planted per hectare of land, while more than 60,000 should ideally be planted per hectare of land [5,9]. Based on the experience of intensive farming in China, experts have designed a complete set of labor-intensive maize dense planting programs with reasonable dense planting as the core, supplemented by thinning and replenishing seedlings, inter-cultivation, weeding, moisture conservation, and soil cultivation [9]. Furthermore, experts have designed simple and cheap auxiliary seeding tools such as the “sowing rope” [9]. The specific method to use the sowing rope is to tie a ribbon on a rope every 30 cm to mark the point of sowing and insert stakes in the field. The farmer should then sow seeds in the same positions, as indicated by the rope, so that the precise plant spacing is determined; the row spacing is then determined by placing a 75 cm long wooden stick perpendicular to the “seeding rope”. After a row is planted, farmers move the stake to the other end of the stick on the ground, and so on [9]. This method of accurately determining the plant spacing and row spacing not only increases the planting density but also greatly facilitates the subsequent weeding work [9]. Furthermore, the intercropping of maize and beans can effectively control weeds’ growth, reduce soil water’s evaporation, and improve drought resistance [74]. The complementarity between the two crops’ morphological and physiological functions improves the utilization efficiency of nutrients and, simultaneously, avoids diseases and insect pests caused by single crop planting, thereby reducing soil pollution [74].

2.5. Data Collection

Due to the COVID-19 pandemic, the research group members were unable to travel to Tanzania for field research. Therefore, we adopted the method of entrusted research, whereby local investigators who have long-term contacts with the China–Tanzania Village-based Learning Center for Poverty Reduction were assigned to collect data. After training, the investigators conducted one-on-one interviews with maize farmers to complete the survey questionnaire. The survey adopted a two-stage method. The first stage was to carry out a small-scale preliminary survey to collect preliminary information, revise the questionnaire, and improve the sampling strategy; the next stage was to conduct a formal large-scale survey to obtain the final data. The preliminary survey period was from 25 January to 8 February 2022. A random sampling of 20 households was conducted in the four sub-villages under Peapea Village, with five maize farmers in each sub-village. After organizing and analyzing the preliminary survey results, the improved maize technology demonstrated by the China–Tanzania Village-based Learning Center considered in this study was determined. This technical package comprised nine main promotion techniques, including the selection of improved varieties, “sowing rope”, maize and pigeon pea interplanting, soil cultivation, plowing, weeding, use of chemical fertilizers, and use of pesticides. The final selected research areas were Peapea Village, where the Village-based Learning Center is located, and two villages sharing labor-intensive applicable technologies in parallel, located in Batini Village and Mbuyuni Village, Rudewa Town, in the Kilosa District; these villages are adjacent to Peapea Village by distances of 1.8 km and 2.3 km, respectively. From each of the 11 sub-villages in these three villages, 20–30 maize farmers were randomly selected.

The official survey was conducted from March to May 2022. Data were obtained through face-to-face interviews with local investigators familiar with local agricultural systems, traditions, values, and the production environment of participating farmers. The farmers signed a consent form before the interview. The main content was to inform participants about the interview content, anonymity, and voluntariness. A total of 300 questionnaires were distributed, and 295 questionnaires were returned. Due to missing important variables and inconsistent answers, 13 questionnaires were invalid and excluded, and a total of 282 valid questionnaires were collected (effective response rate of 94%). This total meets the basic research requirements regarding sample size and structural equation modeling (SEM) analysis [42]. This questionnaire included several sections. The first section collected information on basic household characteristics (e.g., family size, labor size, land, and production assets). A five-point Likert scale was used to measure the affecting factors [33,42] in which 1 = strongly disagree, 2 = disagree, 3 = neutral, 4 = agree, and 5 = strongly agree (see

Table 1. Latent variables, observed variables, and source of observed variable.

Latent Variable	Observed Variable	Source
Behavior perception (BP)	Adopting improved maize technology can improve maize yield. (BP1)	[36,41,63,64,65]
	Adopting improved maize technology can improve maize quality. (BP2)
	Adopting improved maize technology can achieve higher income. (BP3)
	It is easy for me to acquire knowledge of improved maize technology. (BP4)
	Through training, I can easily master the key points of improved maize technology. (BP5)
Compatibility perception (CP)	Improved maize technology is suitable for local farming methods. (CP1)	[36,39,63]
	Improved maize technology can be popularized locally. (CP2)
	I can accept the adoption of improved maize technology. (CP3)
	I think improved maize technology is more suitable for me than traditional technology. (CP4)
Social impact (SI)	I will consider the opinions of my friends and neighbors regarding the adoption of improved maize technology. (SI1)	[36,64,66,67]
	Friends and neighbors have tried improved maize technology and improved their production efficiency. (SI2)
	Rural organizations actively publicize the training and popularization of improved maize technology. (SI3)
	I will be influenced by the propaganda in the village to adopt improved maize technology. (SI4)
Self-efficacy (SE)	It is up to me to decide whether to adopt improved maize technology. (SE1)	[33,39,40,66]
	I have the knowledge and basic skills to adopt improved maize technology. (SE2)
	I do not think it is difficult to learn and adopt improved maize technology. (SE3)
	I can cope with the technical difficulties encountered in the process of adopting improved maize technology. (SE4)
Facilitation conditions (FC)	I can obtain relevant knowledge and training on improved maize technology adoption. (FC1)	[39,67]
	I can get the necessary help and support in the process of adopting improved maize technology (FC2)
Behavior Intention (BI)	I am willing to adopt improved maize technology. (BI1)	[33,40,63,67]
	I will continue to pay attention to the relevant information of improved maize technology. (BI2)
	I am willing to participate in the training and promotion of improved maize technology. (BI3)
	I would like to recommend improved maize technology to my friends and neighbors. (BI4)

).

2.6. Variable Description

The statistical analysis in

Table 2. Basic characteristics of farmers.

Item	Category	Proportion (%)
Gender	Male	54.04%
	Female	45.96%
Age	≤30	3.86%
	31–40	28.42%
	41–50	39.65%
	51–60	23.51%
	>60	4.56%
Education level	None (illiterate)	4.21%
	Basic (can write and read)	2.46%
	Lower primary (1–4 years)	8.07%
	Upper primary (5–7 years)	75.44%
	Secondary (9–12 years)	9.82%
Experience in farming activities (years)	≤10	16.14%
	11–20	37.19%
	21–30	37.19%
	>30	9.47%
Family agricultural labor (person)	1	11.58%
	2	52.63%
	3	19.30%
	>3	16.49%
Maize garden area (acre)	≤1	18.25%
	1 < acres ≤ 2	52.28%
	2 < acres ≤ 3	18.25%
	>3	11.23%

shows that more than half of the interviewed farmers (all family decision-makers) were men, accounting for 54.04% of the total sample. The predominat age group of the farmers was 41–50 years old, accounting for a total of 39.65%. The educational level was generally not high, and the proportion of farmers who had received seven years of formal school education (upper primary) was 75.44%. The farmers interviewed were experienced in farming activities and had mostly engaged in farming for 11–20 years (37.19%) and 21–30 years (37.19%). Family agricultural labor comprised mostly two persons (52.63%). Small farmers were the mainstay, and the maize garden area was mostly 1 < acres ≤ 2 (52.28%) (See Table 2).

3. Results

3.1. SEM Reliability and Validity Tests

The quality of scale design in the questionnaire is directly related to the reliability and validity of the final research conclusion, so it is necessary to test the reliability and validity of the scale. This study uses SPSS 26.0 and AMOS 26.0 for data processing and model estimation.

The most commonly used method to test the construction validity of data is to determine the size of the Kaiser–Meyer–Olkin (KMO) value [75] combined with the significance of Bartlett’s spherical test [42,44]. As shown in the first row of

Table 3. Test results using KMO and Bartlett.

Project		Test Value
Kaiser–Meyer–Olkin Measure of Sampling Adequacy		0.769
Bartlett’s sphericity test	Approximate chi-square	2171.392
	Degree of freedom (df)	253
	Significant (Sig.)	0.000

, the KMO measurement value is 0.769, greater than 0.7. Bartlett’s spherical test approximate chi-square value is 2171.392, and the degree of freedom is 253 with a p value of 0.000 (Table 3). Therefore, the scale data is very suitable for factor analysis [40].

Cronbach’s Alpha coefficient method was chosen to test the reliability of the latent variables. Cronbach’s Alpha value generally needs a score of at least 0.6, and a score of above 0.7 represents high reliability of data [76]. The test results show (

Table 4. Reliability test results and confirmatory factor analysis.

Latent Variable	Observed Variable	Cronbach’s Alpha Value	Factor Load
Behavior perception	BP1	0.768	0.718
	BP2		0.746
	BP3		0.796
	BP4		0.549
	BP5		0.594
Compatibility perception	CP1	0.720	0.654
	CP2		0.643
	CP3		0.749
	CP4		0.754
Social impact	SI1	0.732	0.769
	SI2		0.770
	SI3		0.723
	SI4		0.509
Self-efficacy	SE1	0.758	0.758
	SE2		0.735
	SE3		0.764
	SE4		0.694
Facilitation conditions	FC1	0.707	0.802
	FC2		0.770
Behavior intention	BI1	0.741	0.845
	BI2		0.752
	BI3		0.559
	BI4		0.636

) that Cronbach’s Alpha values of the six latent variables are all above 0.7, indicating good internal consistency and high reliability of the data.

As the collection of the questionnaires in this study is based on the self-evaluation of farmers, there may be homologous variance. Therefore, the Harman factor analysis method was used to test for the problem of homologous variance bias, and exploratory factor analysis was conducted [77]. There are six factors whose initial eigenvalues are greater than 1, explaining 60.646% of the total variance. Of this variance, the first factor explains 21.939%, which is less than 40% of the standard, indicating that the homoscedasticity problem in this study is not serious and within an acceptable range [40,42]. To investigate the measurement effect of the observed variables on latent variables, the factor load of each observed variable was obtained by factor analysis [42]. According to the rotating component matrix, the factor attribution of each observed variable can be judged, and the factor load is generally required to be greater than 0.6. The analysis result shows that two measured items of behavior perception should be eliminated, which are “It is easy for me to learn the knowledge of improved maize technology. (BP4)” and “Through training, I can easily master the key points of improved maize technology. (BP5).” One measured item of social impact should be eliminated, “I will be influenced by the propaganda in the village to adopt improved maize technology. (SI4),” as well as one measure item of behavior intention, “I am willing to participate in the training and promotion of improved maize technology. (BI3)” (see Table 4).

A construction validity test is divided into convergence validity and discrimination validity tests. Convergence validity generally requires that the average variance extraction (AVE) is greater than 0.4 and the combined reliability (CR) is greater than 0.7 [42]. The results in

Table 5. Test results of convergence validity of the scale.

Latent Variable	Average Variance Variation Extraction	Combination Reliability
Behavior perception	0.5257	0.7678
Compatibility perception	0.4064	0.7279
Social impact	0.5388	0.7732
Self-efficacy	0.4439	0.7607
Facilitation conditions	0.5652	0.7191
Behavior intention	0.5764	0.7923

show that the AVEs of the latent variables are greater than the evaluation standard of 0.4, ranging from 0.4064 to 0.5764. The CR of the latent variables are greater than the evaluation standard of 0.7, ranging from 0.7191 to 0.7923.

In

Table 6. Results of discrimination validity test of the scale.

Latent Variable	BP	CP	SI	SE	FC
Behavior perception (BP)	0.526	-	-	-	-
Compatibility perception (CP)	0.065	0.406	-	-	-
Social impact (SI)	0.148	0.154	0.539	-	-
Self-efficacy (SE)	0.029	0.187	0.059	0.444	-
Facilitation conditions (FC)	0.176	0.104	0.135	−0.125	0.565
Average variance variation extraction	0.725	0.637	0.734	0.666	0.752

, the values on the main diagonal are the square root of the AVE of each latent variable, and the remaining values are the correlation coefficients between two latent variables, which need to be analyzed by Pearson correlation. For each dimension, the root value of the AVEs of all dimensions is greater than the correlation coefficient between dimensions, indicating that each dimension has good discriminant validity [78,79]. Table 6 shows that the square root of the AVE value in the last row is greater than the correlation coefficient between dimensions in this study, indicating that the discrimination validity of each latent variable is better than the AVE values of each latent variable.

3.2. SEM Fit Test

According to the results of exploratory factor analysis, a relationship is formed between the observed variables and latent variables [42]. After modifying and fitting the model, an influence path is formed, as shown in Figure 2. The overall fitness test index and the standard value of SEM require the following criteria to be met: a chi-square freedom ratio (X²/df) less than 3.0; a root mean square error of approximation (RMSEA) less than 0.05 (excellent) or less than 0.08 (good); goodness of fit index (GFI), incremental fit index (IFI), Tucker–Lewis index (TLI) and comparative fit index (CFI) greater than 0.90; and the parsimonious goodness-of-fit index (PGFI) and the parsimonious normed fit index (PNFI) greater than 0.50. If the above criteria are met, the SEM is well fitted [39,77,80]. The criteria test results of the overall adaptability of the model are shown in

Table 7. Fitting index results of the structural equation model.

Inspection Index	Adapt to Standard or Critical Value	Fitted Value	Adaptation Judgment
Absolute fitness index
X2/df	<3.00	1.899	Yes
RMSEA	<0.05 is excellent; <0.08 is good	0.056	Good
GFI	>0.90	0.914	Yes
Value-added adaptability index
IFI	>0.90	0.926	Yes
TLI	>0.90	0.906	Yes
CFI	>0.90	0.924	Yes
Reduced fitness index
PNFI	>0.50	0.685	Yes
PGFI	>0.50	0.741	Yes

X²/df, chi-square freedom ratio; RMSEA, root mean square error of approximation; GFI, goodness of fit index; IFI, incremental fit index; TLI, Tucker–Lewis index; CFI, comparative fit index; PNFI, parsimonious normed fit index; PGFI, parsimonious goodness-of-fit index.

. The X²/df value is 1.899, appropriately less than 3; RMSEA is 0.056, less than 0.08; and GFI is 0.914, greater than 0.9; therefore, the results were well adapted. Regarding the value-added adaptation index, the results of the IFI, TLI, and CFI are all greater than 0.9, indicating good adaptation. The PNFI and PGFI are 0.926 and 0.906, respectively, both of which are greater than 0.50, indicating good adaptation. Overall, the scale and actual data used in this study are well adapted to the structural model, and the estimated results have strong reliability.

3.3. SEM Estimation Results

Figure 2 shows the structural equation model with standardized path coefficients based on the maximum likelihood estimation method, built in the AMOS 26.0 software according to the above theoretical analysis framework. Building SEM in AMOS software is a key step to calculating the results of our study. Each ellipse includes the latent variables: Behavior perception (BP), Compatibility perception (CP), Social impact (SI), Self-efficacy (SE), Facilitation conditions (FC), Behavior intention (BI). Each rectangle includes the observed variable of each latent variable (see Table 1). Each observed variable of latent variables has a measurement error [63,64], shown as error variables “e1, e2, … e19” in Figure 2. In our study design, BP, CP, SI, SE, FC are exogenous latent variables, and BI is an endogenous latent variable. A causal relationship between exogenous latent variables and endogenous latent variables is indicated by a single arrow [42]. The residual variable of the endogenous latent variable is represented by “res1”. The exogenous latent variables do not need to set residual variables, but define the covariation relationship between each of them, so the relationship is depicted with double arrows [63,64].

The tabulation of results and the test of research hypotheses is shown in

Table 8. Standardized path coefficient and hypothesis test of structural equation model.

Hypothesis	Path	Standardized Path Coefficient	T Value	p Value	Results
H1	BP → BI	0.331 ***	3.906	***	Accept
H2	CP → BI	0.204 *	2.338	0.019	Accept
H3	SI → BI	0.154	1.818	0.069	Reject
H4	SE → BI	0.032	0.375	0.708	Reject
H5	FC → BI	0.076	0.642	0.521	Reject

* stands for p < 0.05; *** for p < 0.001. BP, behavior perception; BI, behavior intention; CP, compatibility perception; SI, social impact; SE, self-efficacy; FC, facilitation conditions.

. The standardized path coefficient of farmers’ behavior perceptions on their intention to adopt improved maize technology is 0.331, with a t value of 6.314, and a p value of less than 0.001. Under the significance level of 5%, Hypothesis 1, which posits that behavior perception positively impacts farmers’ intention to adopt improved maize technology, is accepted. The standardized path coefficient of the influence of compatibility perception on farmers’ intention to adopt improved maize technology is 0.204, with a t value of 2.338, and a p value of 0.019. Under the significance level of 5%, Hypothesis 2, which posits that compatibility perception positively impacts farmers’ intention to adopt improved maize technology, is accepted.

However, the influence of social impact on farmers’ intention to adopt improved maize technology was not significant (standardized path coefficient of 0.154); therefore Hypothesis 3 is rejected. The influence of self-efficacy on farmers’ intention to adopt improved maize technology was not significant (standardized path coefficient of 0.032); therefore, Hypothesis 4 is rejected. The influence of facilitation conditions on farmers’ intention to adopt improved maize technology was not significant (standardized path coefficient of 0.032); therefore, Hypothesis 5 is rejected.

3.4. FsQCA Variable Selection and Calibration

Based on the above, our study selects behavior perception (BP), compatibility perception (CP), social impact (SI), self-efficacy (SE), and facilitation conditions (FC) as antecedent variables. When performing an fsQCA analysis, the variables involved are first calibrated (Calibrate) to improve the interpretability of the results [77]. First, the average of the five antecedent variables, behavior perception, compatibility perception, social impact, self-efficacy, and facilitation conditions, is calculated. Second, Ragin’s proposed 5% (Fully out), 95% (Fully in), and the intersection point 50 was followed. The standard of % (Crossover Point) was used for data calibration [59], and the data were calibrated using the Calibrate function in fsQCA 3.0 software [77].

3.5. FsQCA Necessity and Sufficiency Analysis of a Single Antecedent Variable

According to the influence of different combinations of independent variables on the dependent variable, different truth tables are generated, and the necessity and sufficiency of the antecedent conditions of each variable are analyzed to obtain the necessary conditions for a single factor [77] (see

Table 9. Sufficient and Necessary Analysis of Antecedent Variables.

Variables	Consistency	Coverage
BP (value = 1)	0.729746	0.764336
~BP (value = 0)	0.540575	0.568826
CP (value = 1)	0.717179	0.724589
~CP (value = 0)	0.568249	0.620828
SI (value = 1)	0.736096	0.721531
~SI (value = 0)	0.540107	0.610364
SE (value = 1)	0.677073	0.681216
~SE (value = 0)	0.589973	0.647495
FC (value = 1)	0.745188	0.686835
~FC (value = 0)	0.437433	0.533376

BP, behavior perception; CP, compatibility perception; SI, social impact; SE, self-efficacy; FC, facilitation conditions.

). The fit index is similar to the significance of the coefficient in the regression analysis, that is, the p-value; this coefficient refers to the degree of consistency between the conditional variable and the result, that is, the extent to which a certain result requires the existence of a certain variable [80]. It can be seen from Table 9 that the highest coincidence degree of facilitation conditions (FC) is 0.745188, which does not meet the standard value of 0.9 of the necessary condition, indicating that there is no indicator that can become a necessary condition for farmers to adopt a behavior intention. Simultaneously, each antecedent variable’s consistency and coverage were lower than 0.8. This shows that the necessity and sufficiency of each single antecedent condition do not meet the absolute necessary condition standard. Furthermore, each antecedent condition cannot be regarded as a sufficient condition for the result, indicating that the farmers’ intention to adopt behavior is affected by overlapping multi-factors and multi-variables rather than a single cause, and each antecedent condition is not a sufficient condition for an outcome [59,81].

3.6. FsQCA Conditional Combination Analysis

When conducting the fsQCA analysis, according to the recommendation of Fiss [57], this study set the consistency threshold to 0.8, and referred to the recommendation of Du and Jia [45] to set the Proportional Reduction in Inconsistency (PRI) consistency threshold to 0.70. The threshold of case frequency is set to 2, and at least 80% of the sample size is reserved. In this study, the simplified solution and the intermediate solution in the standardized analysis results were combined, and the final configuration results are shown in

Table 10. Antecedent variable structure of farmers’ adoption intention.

	P1	P2	P3
Behavior perception	●	●
Compatibility perception	●		●
Social impact		●	●
Self-efficacy	⊗	⊗	●
Facilitation conditions	●	○	●
Consistency	0.891	0.896	0.900
Raw coverage	0.349	0.363	0.400
Unique coverage	0.037	0.051	0.168
Solution consistency	0.874
Solution coverage	0.569

. Among them, ● and ○ indicate that the condition exists; ⊗ indicates that the condition does not exist; blank indicates that the condition may exist and may not exist in the configuration. ● and ⊗ are core conditions, ○ are auxiliary conditions. The antecedent configurations with the same core conditions are classified into one category, and three types of antecedent configuration patterns that trigger farmers’ adoption intention are found. It can be seen in Table 10 that the overall consistency level of the model that triggers farmers’ adoption intention is that the consistencies of each antecedent condition configuration are higher than 0.8, the overall consistency is 0.874, and the overall coverage rate is 0.569. The model has a good interpretation effect. The core variable that exists in all three modes is “self-efficacy.” The main difference is that Pattern 1 and Pattern 2 include “self-inefficacy,” while Pattern 3 includes “self-efficacy.” Therefore, adoption intention can be classified into Self-inefficacy and High Self-efficacy patterns.

3.6.1. Self-Inefficacy Pattern

A core variable that exists in the Self-inefficacy Pattern is high behavior perception. The antecedent configuration of Pattern 1 is mainly “BP, CP, ~SE, FC,” where ~ represents the logical operation, “Non.” Among them, BP, CP, ~SE, and FC are all core variables, with the consistency of 0.891, raw coverage of 0.349, and unique coverage of 0.037. In the case of farmers with self-inefficacy, high behavior perception, high compatibility perception, and high facilitation conditions trigger the intention to adopt technology. The antecedent configuration of Pattern 2 is mainly “BP, SI, ~SE, FC,” where ~ represents the logical operation, “Non.” Among them, BP, SI, and ~SE are core variables, and FC is an auxiliary variable, and the consistency, raw coverage, and unique coverage are 0.896, 0.363, and 0.051, respectively. In the case of farmers with self-inefficacy, high behavior perception, high social impact, and high facilitation conditions can trigger the intention to adopt technology.

3.6.2. Self-Efficacy Pattern

The antecedent configuration of the self-efficacy pattern is “CP, SI, SE, FC,” where CP, SI, SE, and FC are all core variables and the consistency is 0.900, raw coverage is 0.400, and unique coverage is 0.168. These patterns indicate that high compatibility perception, high social impact, and high facilitation conditions can trigger farmers’ intention to adopt technology when their self-efficacy is high.

4. Discussion

This study aims to analyze farmers’ intention to adopt improved maize technology and the key influencing factors thereof from micro-farmer household survey data, and to qualitatively analyze the combination paths that affect farmers’ intention to adopt improved maize technology. There are four main findings and contributions from this study.

(1) It was found that behavior perception had a significant positive effect on behavioral intention (with a standardized path coefficient of 0.331, p < 0.1%), and the degree of influence is the strongest of all variables. This indicates that improving maize yield, quality, income, or the technology itself being easy to learn and apply, will significantly enhance farmers’ enthusiasm for adopting improved maize technology. This is consistent with the conclusions of previous studies, which propose that the more positive the behavioral attitude, the stronger the behavioral intention [82]. Compatibility perception had a significant positive effect on behavioral intention (with a standardized path coefficient of 0.204, p < 0.5%), indicating that compatibility perception has a significant impact on farmers’ intention to adopt the technology. When farmers perceive that improved maize technology conforms to their own values, practical experience, and environment and is suitable for local use, their adoption enthusiasm will be higher. This is consistent with conclusions from previous studies that propose that the more positive a person’s compatibility perception, the stronger their behavioral intention.

(2) The influence of social impact, self-efficacy, and facilitation conditions on maize technology adoption intention was positive but not significant. This is not consistent with previous studies [33,39,40]. However, it does not indicate that farmers’ technology adoption is unrelated to the influence of friends, neighbors, and rural organizations or farmers’ self-efficacy and the facilitation conditions of technology adoption. This result may be because improved maize technology, as an exotic agricultural technology, will be the result of interdependence among various factors when it is promoted in Tanzania, which is a complex social phenomenon [83]. Factors such as social impact, self-efficacy, and facilitation conditions may affect farmers’ behavioral intention to adopt improved maize technology. Unlike the above analysis, these factors do not have a strong independent positive influence on behavior perception as they produce a kind of “joint effect” [83].

(3) To further analyze the situation that the possible independent variables are interdependent and work together, our research used the qualitative comparative analysis tool, fsQCA, to analyze the configuration relationship between the factors that affect farmers’ intention to adopt technology. The results are as follows: the first configuration category is the self-inefficacy of farmers, that is, when farmers do not think they have enough ability to use improved maize technology. The combination of factors that trigger farmers’ intention to adopt technology in Pattern 1 (coverage rate of 0.349) is “behavior perception, compatibility perception, self-inefficacy, facilitation conditions.” This indicates that even when farmers think they do not have enough ability to use improved maize technology, if they realize the usefulness and ease of use of the technology, that it is well suited to the local area, and the availability of facilitation conditions to access relevant knowledge and obtain technical guidance, farmers will be willing to adopt the technology. Pattern 2 (coverage ratio of 0.363) is “behavior perception, social impact, self-inefficacy, facilitation conditions.” This indicates that even when farmers think they do not have the ability to use improved maize technology, if they recognize the usefulness and ease of use of the technology, its positive social impact (including the core conditions of friends and relatives around them and the promotion of the technology in villages, towns, and organizations), and the auxiliary role of facilitation conditions, the farmers’ intention to adopt the technology will be triggered.

(4) The second configuration category is the self-efficacy of farmers, that is, when farmers believe that they have sufficient ability to use improved maize technology. The combination of factors that trigger farmers’ intention to adopt technology (coverage rate is 0.400) is “compatibility perception, social impact, self-efficacy, facilitation conditions.” This indicates that when farmers think they have self-efficacy, the perception of higher compatibility between improved maize technology and traditional maize technology, their own values, practical experience, environment, positive social impact, a higher level of compatibility, and good facilitation conditions can trigger farmers’ intention to adopt improved maize technology. The coverage rate of this path is greater than that of the two modes of self-inefficiency, which indicates that the explanatory power of this pattern is greater than that of the two patterns of self-inefficiency [79].

5. Conclusions

This study conducted a questionnaire survey on 282 maize farmers in Peapea Village, Batini Village, and Mbuyuni Village in the Morogoro Region of Tanzania, where the China–Tanzania Village-based Learning Center for Poverty Reduction project is located. Based on the TPB, the SEM method was used to study the influencing factors of farmers’ intention to adopt improved maize technology. Through the analysis, it was concluded that farmers’ behavior perception and compatibility perception positively impacted China’s agricultural technology choice intention. The fsQCA method was employed to explore the combination paths that affect farmers’ intention to adopt improved maize technology. Through analysis, it was concluded that there are three patterns that can trigger farmers’ intention to adopt technology, namely “behavior perception, compatibility perception, non-self-efficacy, facilitation conditions,” “behavior perception, social impact, non-self-efficacy, facilitation conditions,” and “compatibility perception, social impact, self-efficacy, facilitation conditions.” The results show that the influencing factors of farmers’ intention to adopt improved maize technology are multi-dimensional and complex. To increase crop yield and promote the use of improved agricultural technology in Tanzania, an effective combination path should be selected to influence farmers’ intention to adopt such technology.

Although our study provides insights into strengthening industry sustainability by improving technology usage, there are several limitations. First, we focused on the perspective of behavior attitudes, subjective norms, and perceived behavioral control. However, there could be other factors that influence intention. Second, we only used data from a one-time survey. In the future, if conditions permit, long-term observation and data collection should be conducted. Third, our study only analyzed the influencing factors that affect farmers’ intention to adopt improved maize technology. Future research can further analyze the path of and factors that influence the shift from adoption intention to actual adoption.

There are also methodological limitations in our study. (1) SEM is a confirmatory factor analysis rather than an exploratory factor analysis. It means that in the measurement model part, which latent variable corresponds to each observed variable is decided in advance, and the path between the structural model variables is set based on prior theoretical assumptions. The research tests whether each observed variable can effectively measure its corresponding latent variable. This is different from exploratory factor analysis, in which the decision regarding which latent variable corresponds to each observed variable is determined by the data itself [84]. (2) Even a well-fitted SEM model may have problems with low-order components and ignore important variables [84]. (3) It is impossible to know the sequence of the conditions in the configuration through the QCA method, and it is impossible to know whether the configuration obtained by our study will change over time [85,86]. Therefore, in future research, we can use other models and add other quantitative variables to further analyze the results of the study. It is also necessary to combine dynamic panel data with QCA to further effectively explain the degree of change in conditional configurations over time.

Based on the above empirical analysis, the following policies are recommended.

First, the promotion and training should consider the ease of use, usefulness, and benefit improvement of the maize technology. Simultaneously, it should highlight the alignment of the technology with local planting traditions, habits, and values, so that local farmers do not reject the cognition of new technologies at first sight. Subsequently, the opportunities for farmers to experience the technology through demonstration households or the demonstration center should increase farmers’ perception ability. The existing literature also emphasizes the importance of cooperation between Chinese extension experts and local governments [7,8,9,25]. To expand the scope of the demonstration, a very important point is to strengthen the role of the local extension agents in the village, especially through interactions with the leadership of the village council and the village committee, so that they can become an important force in organizing training and extension for farmers, and mobilize them to promote improved maize technology.

Second, farmers should be subdivided according to their self-efficacy, and different promotion methods should be adopted for different groups. This recommendation was not mentioned in the previous literature. For farmers who perceive that they lack the ability to master improved maize technology, a possible combination method is to increase the opportunities for farmers to receive technical guidance in addition to strengthening the publicity and guidance regarding the technical usefulness, ease of use, and compatibility of the technology; another combination method is to allow farmers to receive technical guidance. Farmers who recognize the usefulness and ease of use of the technology can enhance the social impact of improved maize technology by strengthening propaganda and promotion of the technology in villages, towns, and organizations.

Third, for farmers who perceive that they have the ability to master improved maize technology, strengthening their perception of technology compatibility and strengthening the popularization of relevant knowledge and advantages can facilitate increased intention for adoption. The latter can be achieved by making use of local organizations and media to publicize and expand the scale and scope of training and extension, thereby, improving farmers’ cognitive levels regarding the technology. Furthermore, these strategies can play a positive role in social influence and the process of technology popularization. Stable technical support should be provided to ensure that farmers can conveniently obtain the relevant knowledge of improved agricultural technology.

If farmers’ intention to adopt improved maize technology can be increased to a greater extent through the above recommendations, it can accelerate the improvement of agricultural technology in Tanzania, thereby increasing agricultural productivity, improving agricultural production at the village level, and reducing poverty.