2.1. Broadband Internet Overview

Broadband networks are defined by the ITU as those that are able to deliver the Internet at a speed faster than the Integrated Services Digital Network principal rate of 1.5 Mbps or 2.0 Mbps (International Telecommunication Union [ITU], 2003). Broadband networks provide broadband Internet. Digital subscriber lines, cable modems, fiber optics, broadband overpower lines, satellites, and wireless radio, all of which are divided into fixed and wireless categories, are just a few of the several underlying technologies that can be used to access these networks. Fixed broadband technologies have proven costly and challenging to build in developing countries, especially in rural areas; hence, wireless mobile technologies play a central role in broadband service provision (Prieger, 2015).

Broadband technology is regarded as a general-purpose technology that serves a broad range of uses (Naughton, 2016). It has been able to revolutionize people’s lives in many aspects, such as education, industries and manufacturing, health, governance, and the environment; good examples are seen in developed countries (Koutroumpis, 2019). According to studies, mobile broadband penetration boosts countries’ gross domestic product (GDP); for every 10% penetration increase, middle-income countries’ GDP increases by 1.8%, while low-income countries’ GDP increases by 2% (Bahrini & Qaffas, 2019; Ghosh, 2017; Katz & Callorda, 2019). However, until broadband is adopted and used at its full potential, the technology is of little use.

2.2. Broadband in Tanzania

Tanzania has maintained its status among the LMICs, with a gross national income of $1,136 and GDP growth of 4.2% in 2021 (World Bank, 2022). There are 61.7 million people living in the nation, and 64.05% of them live in rural areas (Tanzania National Bureau of Statistics [NBS], 2022). In Tanzania, 98% of the population subscribes to telecommunications services, with mobile networks accounting for 99.86%. Tanzania has a 50.6% Internet penetration rate, with the majority of users accessing it through mobile devices (Tanzania Communications Regulatory Authority, 2022).

Research claims that 86% of rural residents lack Internet access, compared to 44.6% of urban residents, making them the most deprived group (Mothobi et al., 2017). Moreover, 48% of the Tanzanian population lives under a broadband footprint, but they do not use it (GSMA, 2021a). Initiatives to push Internet uptake, especially in rural areas, were put in place, such as the Universal Communication Services Access Fund and National Information and Communication Technologies Broadband Backbone, but Internet penetration is not growing at a faster pace compared to the early Internet days of 15% between 2011-2015, as opposed to 6% between 2016-2020 (Tanzania Communications Regulatory Authority, 2021). With such statistics, Tanzania is a better candidate to exemplify other LMICs in this study.

2.3. Adoption Theories

As there is still a broadband usage gap globally, establishing successful strategies for promoting end-user acceptance and the subsequent use of broadband remains a fundamental challenge (GSMA, 2021b; International Telecommunication Union [ITU], 2021; International Telecommunication Union [ITU] and United Nations Educational, Scientific and Cultural Organization, 2021). Many theoretical models have been established to comprehend user adoption of technologies. In their theory of reasoned action (TRA), Fishbein and Ajzen (1975) highlighted that the main determinants of technology adoption intentions are attitudes toward behavior (ATB) and the impact of subjective norms (SN). Despite its application in several settings, TRA failed to consider volition control as one of the factors driving adoption behavior. To address the issue of volitional control, Ajzen (1985) established TPB, which adds perceived behavioral control (PBC) to existing constructs as a driver of the intention to engage in a behavior. Moreover, as an extension of TRA, Davis (1989) introduced the technology acceptance model (TAM), which predicts a person’s adoption and utilization of technology based on perceived usefulness and perceived ease of use. Although TAM is among the most popular adoption models, the attitude construct was left out in the final conceptualization, casting doubts on it, as Ajzen et al. (2018) argued that attitude is an important key predictor of behavior. Furthermore, the diffusion of innovation (DOI) theory developed by Rogers (1995) holds that the adoption of technology is influenced by social systems, innovation, time, and communication channels. However, Jokonya (2017) argues that DOI is deterministic in nature and concentrates more on the technical side of technology, ignoring the social context. Technology adoption is said to go beyond technical aspects as it must incorporate social, economic, and political factors (Sopha et al., 2017).

Given its scientific legitimacy, TPB is a notable behavioral action model compared to others when it comes to developing strategies and designing intervention mechanisms (Ajzen & Schmidt, 2020). TPB considers all the elements that could directly or indirectly influence behavior through behavioral intention, which has been proven to be the strongest predictor of one’s involvement in a certain activity (Canova & Manganelli, 2020). Three variables: ATB, SN, and PBC are used in the TPB to determine intention. ATB refers to one’s opinion of a behavior, whether favorable or unfavorable, whereas SN refers to the perceived influence of friends, family, or other individuals on decision making. The concept of PBC refers to how one perceives their level of control over their own conduct.

Several studies have employed TPB to investigate the intention towards using or adopting certain technologies. Recent studies are listed herein: blockchain cryptocurrency adoption in an e-government context (Schaupp & Festa, 2018), mobile learning adoption by medical students (Azizi & Khatony, 2019), blockchain adoption as a supply chain technology (Kamble et al., 2019), adoption of embedded telemedicine support services (Gantz, 2020), adoption of low-carbon agriculture technologies in China (Yang et al., 2022), ICT adoption among intensive shrimp farmers (Ulhaq et al., 2022), adoption of green technology in Pakistan (Zeng et al., 2022), and technology adoption among small retailers (Aithal et al., 2023). The aforementioned studies justify the role that TPB can play in examining the factors influencing technology adoption. However, to the best of our knowledge, there is limited research on intention-based studies using TPB in rural broadband Internet adoption.

2.4. Hypothesis Definition

Since this study’s goal is to examine rural dwellers’ intention to adopt broadband Internet using TPB, its hypotheses revolve around the aforementioned theory with the addition of demographic features. The central tenet of the theory of TPB is that intentionally planned behaviors are predicted by the intention toward such behaviors expounded by ATB, SN, and PBC. This subsection formulates the hypotheses for intention exploration adopted in this study, as summarized in Fig. 1.

Fig. 1

Hypotheses summary.

According to TPB, attitude acts as an evaluative inclination toward behavior because of its relationship with intention (Ajzen et al., 2018). The idea of attitude, which expresses positive or negative thoughts about a certain behavior, is seen as a key factor determining the intention in many earlier studies on technology adoption, including Azizi and Khatony (2019), Mazambani and Mutambara (2020), Yang et al. (2017), and Shalender and Sharma (2021), to mention but a few. Therefore, the definition of hypothesis H1 is

H1: The behavioral intention of LMIC rural residents to adopt broadband Internet and their attitude toward doing so are significantly and positively correlated.

The SN stems from the belief that people’s intentions to engage in specific behaviors are influenced by the opinions of significant others. Studies by Lin et al. (2020), Rana et al. (2019), Chu and Chen (2016), Masud et al. (2016), and Jing et al. (2019), to name just a few, have shown that SN significantly affects one’s intention to use technology. Therefore, the definition of hypothesis H2 is

H2: The behavioral intention of LMIC rural residents to adopt broadband Internet and their SN toward doing so are significantly and positively correlated.

Despite having a good attitude toward technology, some people may not think they are capable of using it; it is until they believe they have control over how technology is used that their intention to use it rises (La Barbera & Ajzen, 2021). Several researchers, e.g., Lin et al. (2020), Anser et al. (2020), and Mohr and Kühl (2021) have found strong empirical evidence for this association. Therefore, the definition of hypothesis H3 is

H3: The behavioral intention of LMIC rural residents to adopt broadband Internet and their perceived behavioral control toward doing so are significantly and positively correlated.

Not only do the TPB’s constructs influence intention to adopt technologies, but demographic features have also been proven to do so. Numerous studies have demonstrated the impact of age, income, and education on technology use, to name a few: Yu et al. (2017), Puspitasari and Ishii (2016), Lee and Coughlin (2015), and Sharma (2015). Therefore, the definition of hypothesis H4 is

H4: The behavioral intention of LMIC rural residents to adopt broadband Internet and their age, education, and income toward doing so are significantly and positively correlated.

3.1. Study Design

This study employed a cross-sectional survey approach, following Creswell and Creswell (2017) recommendations, to examine users’ intentions to use broadband Internet in LMICs using Tanzania as an example. Due to its quick turnaround, data standardization, and affordability, survey research was the method of choice for gathering data (Babbie, 2021).

3.2. Target Population & Participant Recruitment

This study targeted people living in rural areas of LMICs, as exemplified by Tanzania. The study was interested in places that have a broadband footprint but poor Internet usage. The authors had to fetch information from service providers concerning rural areas with broadband services, but characterized by poor usage/adoption. People invited to participate in the survey were recruited from rural areas with such features as the villages listed herein: Olkung’wado (Arusha), Mgwashi (Tanga), Teule (Tanga), Jitengeni (Tanga), Chemba (Dodoma), Chinangali II (Dodoma), and Mlali (Dodoma). With the assistance of village executives, a non-probability voluntary response sampling was adopted to recruit participants. The survey was planned to be administered to 175 participants (25 people from each village). Jenkins and Quintana-Ascencio (2020)’s recommendations for a minimal number of participants in studies involving regression served as the basis for sample size determination. Out of the expected 175 participants, only 155 consented to participate in the survey (almost an 88.57% response rate), which was sufficient to capture the intended data.

3.3. Survey Development & Administration

Since the study employs TPB to examine the intention to adopt broadband Internet, the survey development and administration followed the guidelines of Ajzen (2006). Nonetheless, a well-established questionnaire must go through properly defined steps: the elicitation stage, pilot research, and a final questionnaire (Taherdoost, 2016). The survey started with the elicitation process, involving 30 participants from two villages among the villages of interest (Olkung’wado and Chinangali II). Exploiting focus group discussions, beliefs, and opinions regarding the topic under study were elicited using open-ended questions. Thereafter, the authors applied content analysis to summarize the themes that later paved the way for the initial questionnaire, with 45 items divided into three sections: demographic features, broadband usage, and TPB’s constructs. A 5-point Likert scale, with 1 representing “strongly disagree” or the equivalent and 5 representing “strongly agree” or the equivalent, was used to score the potential answers to the TPB constructs.

To identify questionnaire issues and flaws in the question-answering process as well as other potential measurement errors, we invited three experts (survey methodologists) to analyze the questionnaires. The questionnaire was refined during this process, which decreased the number of items to 40, leading to the creation of a pilot questionnaire. A pilot survey was conducted with 30 individuals (number recommended by Viechtbauer et al. [2015]) to confirm the testing instrument. To capture different perspectives, the aforementioned individuals were sourced from Chemba and Mlali, which are different villages from those used in the elicitation stage. The pilot test led to modification of the questionnaire to contain 35 items. The survey tool is attached as Appendix A.

With survey administration, which took three weeks, a research administered technique was opted as rural dwellers in most of the developing countries (Tanzania in this study) are characterized by poor literacy and digital skills (World Bank, 2016). With this technique, a researcher can clarify questions, ensure the completion of a questionnaire, and obtain a higher response rate, as well as greater control of the environment. Both the pilot test and the survey itself were conducted using a free open-source tool for data collection called KoboToolbox.

3.4. Measures

3.5. Data Analysis

IBM SPSS Statistics version 26.0 (IBM Co., Armonk, NY, USA) was used for statistical analysis of the study data. Using the Kolmogorov-Smirnov test, the data used to assess the TPB’s construct were examined for normality to determine if any items had p-values greater than 0.05, which would not have rejected the null hypothesis of normality. Cronbach’s alpha was used to assess the internal consistency of the survey. According to Abu-Bader (2021), a measurement is considered reliable if its Cronbach’s alpha coefficient is greater than 0.7.

Since the dependent variable (intention) was measured in an ordinal fashion, ordinal logistic regression was selected as a suitable analysis technique (Williams, 2016). As the study was correlational, the outcome was defined as the variable of interest, and the hypotheses were tested based on the defined variables (age, income, level of education, ATB, SN, and PBC) (Lau & Kuziemsky, 2016). With regard to TPB, Equation (1) represents the relationship between the anticipated outcome and the hypothesized variable; henceforth, coefficients are presented as odds ratios. If the parallel regression assumption is affirmed, the estimated coefficients (β_n) are translated as odds ratios, with β₀ and ε being the regression constant and error term, respectively.

4.1. Socio-Demographic Characteristics

The survey participants were 59 females (38.1%) and 96 males (61.9%). The majority age group involved in the survey was 26-45 years of age (51.6%). Moreover, only five participants attended college/university, and the majority of them ended up in primary school (64.5%). With regard to monthly income, the majority of participants (49.7%) earn between 50,000-100,000 Tanzania Shillings (Tsh.). The aforementioned statistics are summarized in Fig. 2. Most of the participants owned a mobile phone (91.6%), but only 18.1% of them had phones that could access the Internet, which happened to be the same percentage of Internet users among the participants.

Fig. 2

Demographic features. (A) Sex distribution. (B) Age distribution. (C) Education level. (D) Income distribution (Tsh.). Tsh., Tanzania Shillings.

4.2. Reliability Test

A reliability test was conducted for the entire survey (excluding demographic features) and the individual TPB constructs. Cronbach’s alpha values are listed in brackets for the entire survey (0.913), ATB (0.888), SN (0.714), and PBC (0.712). The aforementioned statistics are summarized in Fig. 3. When demographic features were considered, the Cronbach’s alpha value for the entire survey was 0.920. The fact that every survey measure exhibited internal consistency with Cronbach’s alpha values above the suggested 0.7 indicates that the measures are reliable.

Fig. 3

Reliabilty analysis summary. (A) Entire survey. (B) ATB measure. (C) SN measure. (D) PBC Measure. ATB, attitudes toward behavior; SN, subjective norms; PBC, perceived behavioral control.

4.3. Validity Test

Owing to the ordinal nature of the survey data, the monotonic relationship between pertinent variables (intention, ATB, SN, PBC, education, age, and income) was evaluated using Spearman’s correlation (Schober et al., 2018). There were positive and significant correlations between intention and ATB measure, rs=0.53, n=155, p< 0.01, intention and SN measure, rs=0.56, n=155, p<0.01, intention and PBC measure, rs=0.35, n=155, p<0.01, intention and education, rs=0.58, n=155, p<0.01, and intention and income, rs=0.34, n=155, p<0.01. Meanwhile, age and intention were negatively and significantly correlated (rs=-0.22, n=155, p=0.006). The test results are listed in Table 1.

4.4. Descriptive Statistics

Table 2 presents TPB constructs. The average ATB of the participants was 3.59, with a 0.46 standard deviation. SN had the lowest mean score (3.34) and lowest standard deviation (0.40), according to the results. The PBC mean score and standard deviation were found to be 3.75 and 0.34, respectively. With a standard deviation of 0.66, intention obtained a mean score of 4.12.

4.5. Ordinal Logistic Regression Predicting Intention to Adopt Broadband

Before examining the individual TPB predictors in the model, we needed to ascertain whether the model improved the ability for outcome prediction. According to the model-fitting data in Tables 3, 4, the final model fit was considerably better than that of the null model (p<0.05), demonstrating that the model accurately predicts consumers’ intentions to adopt broadband Internet centered on the marginal likelihoods of the resultant categories. The Pearson and Deviance chi-square tests provide a useful measurement of a model’s goodness-of-fit; both p-values were >0.05, indicating a good match between the model and the data (Tables 3, 4).

When dealing with ordinal logistic regression the pseudo-R² is treated as the rough analog of R² using three estimations: Nagelkerke (1991), McFadden (1972), and Cox and Snell (1989). Among them, the model with the largest pseudo-R² value is assumed to be the best; hence, upon analysis, the pseudo-R² values (Nagelkerke, 1991) indicate that TBP constructs describe the variance in the intention to adopt broadband Internet by 50.3% (Table 5).

This present article’s key emphasis is on TPB measures, which were established by the parameter estimates shown in Table 6. The link between the predictor variables (ATB, SN, and PBC) and outcomes (intention) can be explained as follows:

i. ATB was a significant positive predictor of intention to adopt the Internet (p<0.05). The log odds of a rural resident falling into a higher category of intent to adopt broadband were projected to increase by 1.938 for every unit increase in ATB.

ii. SN was a highly significant positive predictor of intention (p<0.05); the expected increase in the log probability of a rural resident of an LMIC falling into a higher category of intention was 2.144 for every unit increase in the corresponding construct score.

iii. The least significant positive predictor among the TPB measures was PBC (p=0.013). The expected increase in the log chances of a rural resident of an LMIC falling into a higher category of intention is 1.437 for every unit increase in the PBC score.

Moreover, the selected demographic features (age, education, and income) were also checked to determine their influence in predicting the intention to adopt broadband Internet in rural settings. Both variables were significant predictors (p<0.05); however, age showed a negative correlation (β=-0.725) while education and income showed positive correlations (Table 7).