Extent of adoption of mobile phone applications by smallholder dairy farmers in Tharaka Nithi County, Kenya

Abstract

In Kenya, smartphones are readily available at lower cost than before, allowing farmers to access agricultural information through mobile applications. However, despite increased ownership of smartphones and the availability of mobile applications, the overall usage of mobile applications is still low. This research aimed to assess the factors that determine the adoption of mobile phone applications among smallholder dairy farmers in Tharaka Nithi County of Kenya. A multistage stratified sampling procedure was used to interview 427 farmers. A Double Hurdle approach was employed to analyze the drivers of decision and extent of adoption of mobile applications. The study revealed that 51% of the respondents used mobile applications to access agricultural information. The decision to use a mobile phone application was influenced by the sex of the household head, age and level of education, distance to the market center, farming (as the primary occupation), access to credit, channels of access to information, farmer’s perception on the cost and ease of use of the mobile application. Subsequently, farm size and the various channels through which farmers access information about the applications positively influenced the number of mobile applications used. This study recommends strengthening the existing extension pathways to disseminate more information regarding the use of mobile applications among smallholder farmers. Supporting public-private partnerships will be crucial in increasing the utilization of mobile phone applications.

Keywords:

• agricultural information
• mobile applications
• smallholder dairy farmers
• smartphones

Public interest statement

In Kenya, smartphones are now more affordable, which means farmers have a chance to use mobile applications to access agricultural information. A study involving 427 smallholder dairy farmers in Tharaka Nithi County, found that 51% of the farmers used mobile applications to get agricultural information. More men had mobile applications than women and the difference was significant. As well, younger farmers adopted mobile applications than the older farmers. There were differences in the adoption of mobile applications in the education level of the farmer, the distance to the market, farming as the main occupation, access to credit, and how farmers learned about the applications. Farm size and the different ways farmers accessed information about the applications also affected the number of applications adopted. The study suggests that we need to improve how we share information about agricultural applications with farmers. The public and private organizations need to work together to create awareness of mobile applications.

1. Introduction

Dairy production is a crucial sector that contributes to global food security, nutrition, and household livelihoods (Derbe et al., 2022). Dairy farming is recognized as a pathway towards achieving global food security. The demand for dairy products is projected to increase, driven by factors such as rising disposable income, population growth, urbanization, and changes in dietary preferences resulting in higher calorie intake (Food and Agriculture Organization of the United Nations, 2018). In Kenya, it is estimated that 28% of households engage in dairy production, ranking it second after indigenous cattle in livestock production (Kenya National Bureau of Statistics, 2019). Despite this relatively small proportion, smallholder dairy farmers contribute more than 88% of the total milk production in Kenya, estimated at 3.9 billion liters per year (Kenya Dairy Board, 2019).

Despite the high demand for dairy products in Kenya, the dairy sector operates below its potential (Ochieng et al., 2020). It is also projected that Kenya’s population will rise to 96 million people with 41 million living in the urban areas by the year 2050 (Food and Agriculture Organization of the United Nations, 2019). The rising trend therefore calls for more attention to the production of dairy products in an attempt to bridge the gap between demand and supply. Some of the identified setbacks facing smallholder dairy farmers in Kenya include; limited access to information on inputs such as feeds, breeds, markets, health, and inadequate technical skills. Studies have asserted that access to information through different extension pathways plays a crucial role in agricultural development-related programs to deal with the challenges affecting production (Anna & Helene, 2014; Ragasaa & Mazundab, 2018; Rawlins et al., 2014). Extension services’ core objective is to provide relevant information to farmers to improve productivity (Danso-Abbeam et al., 2018).

In Sub-Saharan Africa, extension services have primarily been offered through extension officers, researchers, or service providers having physical contact with farmers (Ameru et al., 2018; Thiam & Matofari, 2018). However, according to Norton and Alwang (2020), extension has been evolving, driven by structural changes in agriculture that accompany economic development, new agricultural technologies, new information and communication technologies, uncertain and dwindling public sector funding for extension, and decentralized governments with increased involvement of local governments in the delivery of extension services. In an attempt to improve extension service provision and tackle challenges that limit productivity, there has been an effort to use Information and Communication Technology (ICT) by both public and private sector players (Hamid & Mohammad, 2018; Joyous & McNamara, 2018; Larochelle et al., 2019; Omulo & Kumeh, 2020; World Bank, 2017).

“Information and Communication Technology” may be defined as the convergence of electronics, computing, and telecommunications and includes radio, television, telephones (fixed and mobile), computers, internet, and are evolving almost on a daily basis (Patel, 2018). Extension services through mobile phones are mostly on weather advisory, crop and livestock production, record keeping, financial services, market information, and data analytics (Haruna et al., 2018). When using mobile phones, extension services can be accessed and delivered using various means, including Short Message Services (SMSs), Unstructured Supplementary Service Data (USSD), mobile applications (apps), voice or browsing web-based applications (Emeana et al., 2020; Quandt et al., 2020). Several studies have been made to assess the use of mobile phones in agricultural extension with empirical evidence that the use of mobile phones is effective in delivering agricultural information and has contributed to: improved production (Aker & Ksoll, 2016; Casaburi et al., 2019; Haruna et al., 2018; Mwita et al., 2020); adoption of technologies (Cole & Fernando, 2016) and market participation (Aker, 2010; Minkoua et al., 2018; Tadesse & Bahiigwa, 2015).

Access to information from mobile phone applications requires that a farmer must have a smartphone or a tablet, and internet connectivity. Mobile applications are software programs intended to operate on mobile devices like smartphones and tablets (Kumar & Karthikeyan, 2019). Ownership of a smartphone is a key determinant in the use of mobile applications (Krell et al., 2020; Thar et al., 2021). Increased availability of agricultural-related mobile applications coincides with increased ownership of smartphones in developing countries. For instance, in Kenya, the Communications Authority of Kenya (2021) reports that as of September 2021, 41.6% of the adult population owning a mobile phone had access to broadband internet. On the other hand, the Kenya Agriculture and Livestock Research Organization (KALRO) has developed mobile applications to support agricultural production, the apps include the Kenya Agri-Observatory Platform (KAOP), Livestock Selector, Digital Veterinary among others (KALRO, 2022). Availability of these applications adds to the many mobile applications developed and disseminated by the private sector such as WeFarm, Icow, FarmDrive, Vet Africa, and uLima among many in an attempt to diversify the various channels of access to information among smallholder dairy farmers in Kenya (Emeana et al., 2020). In addition, web-based applications such as social media, YouTube, and search engines such as Google are readily available for use by smallholder dairy farmers to access agricultural information.

There is limited evidence to explain the reason why, despite the increased ownership of smartphones among smallholder farmers and the availability of useful mobile applications, their utilization for accessing agricultural information remains low in Kenya. Previous studies have mainly focused on the decision to use mobile phones (Aker & Ksoll, 2016; Ameru et al., 2018; Haruna et al., 2018; Katengeza et al., 2011; Okello et al., 2012; Okoroji et al., 2021). Other research has centered on the level of usage of Short Message Services (SMSs) (Casaburi et al., 2019; Fafchamps & Minten, 2012; Kassem et al., 2020; Larochelle et al., 2019; Nakasone, 2013). This study contributes to the adoption literature by identifying the determinants of the decision to adopt and the extent of adoption of mobile applications among smallholder dairy farmers in Tharaka Nithi County, Kenya.

2. Review of past studies on the adoption of agricultural technologies

The explanatory variables in this study were selected based on a review of related literature. Empirical studies have shown that the decision to adopt and use agricultural technologies is determined by socio-economic characteristics, farmer perceptions, technology attributes, and access to information.

2.1. Socio-economic characteristics

Studies done in Kenya have found that age, gender, household size, occupation, distance to output market, owning a phone, education level, value of assets, number of crop enterprises, farming experience, crop income, membership in farmer organizations are the factors that explained the use of ICT tools for agricultural transaction purposes (Krell et al., 2020; Mwenda et al., 2022; Okello et al., 2012). Mwangi and Kariuki (2015), in their review of previous studies on technology adoption, reported that economic and institutional factors, as well as human-specific factors, play a crucial role in determining technology adoption.

In India, a study by Ali and Ghildiyal (2023) found that males with higher age, education, and income were more likely to adopt digital financial services. Similarly, farmer’s age, education level, and farm size influenced farmers choice of information sources in a separate study in India (Surabhi & Mamta, 2016). In Myanmar, limited internet access and lack of digital knowledge were identified as the primary barriers to the adoption of agricultural applications (Thar et al., 2021).

A study conducted by Hoang (2020) in Vietnam examined the adoption of mobile phones for the marketing of cereals among smallholder farmers. The findings indicated that young male smallholders with higher levels of education and who were members of community-based organizations, had higher income levels, and participated in credit programs, had higher adoption of mobile phones for marketing purposes.

2.2. Farmer perceptions

Wyche and Steinfield (2016) conducted a study to investigate the barriers to adopting mobile services in Western Kenya. Their findings revealed that farmers’ perception of mobile phones solely as devices for voice communication hindered their utilization for accessing agricultural information. In a qualitative study conducted among smallholder farmers in Kenya and Zambia, a mismatch between the design of mobile phone applications and farmers’ perceptions and usage of their devices was found (Wyche et al., 2015). In the Kenyan highlands, a study found out that the level of trust farmers had in ICT-based pest information services influenced their adoption (Mwenda et al., 2022). Mwangi and Kariuki (2015) highlight the significance of farmers’ perceptions of innovations in determining the adoption of agricultural technologies.

2.3. Attributes of technology

The cost of technology, the profitability of its use, and ease of use were identified as key determinants of agricultural technology adoption by several studies (Diaz et al., 2021; Okoroji et al., 2021; Rehman et al., 2016).

2.4. Access to information

A study conducted to examine the effectiveness of dissemination pathways in driving the adoption of “Push and Pull” Technology in Western Kenya found that access to information played a crucial role in promoting the use of “Push and Pull” Technology among farmers (Murage et al., 2012). An adoption gap of 2% to 16% was observed in a study that aimed at understanding the level of adoption of Climate-Smart Agriculture (CSA) technologies and practices, as well as the drivers needed to encourage large-scale uptake of CSA in West Africa. This gap was attributed to the incomplete diffusion of information and the lack of awareness regarding CSA technologies and practices (Ouédraogo et al., 2019). Access to information was a key determinant in the adoption of various combinations of CSA technologies in a study done in Zimbabwe on the adoption patterns of Climate-Smart Agriculture in integrated crop-livestock smallholder farming systems (Mujeyi et al., 2022). Overall, these studies underscore the significant role of information access in driving the adoption of agricultural technologies and practices.

3. Methodology

3.1. Study area

This study was conducted in Tharaka Nithi County located in the Eastern part of Kenya (Figure 1). Tharaka Nithi County was purposively selected as it was involved in the piloting of Disruptive Agricultural Technologies (DAT), an initiative led by the World Bank and the Government of Kenya aimed at bringing 1 million farmers into an impactful digital platform (Kim et al., 2019). DAT involves the integration of digital technologies into crop and livestock management, enabling agricultural stakeholders to enhance their operational efficiency. One aspect of DAT involves the utilization of mobile phone technology for dissemination and access to extension services, leading to increased productivity.

Figure 1. Map of the study area.

Tharaka Nithi County covers an area of 2,662.1 km² with a population of 393,177 (Kenya National Bureau of Statistics, 2019). The County consists of five Sub-Counties and a total of 15 Wards, it is divided into four major agroecological zones namely upper highlands, lower highlands, upper midlands, and lower midlands (Ministry of Agriculture et al., 1982). The key enterprises in the County include livestock keeping (exotic dairy cows, exotic beef cows, indigenous cattle, shoats, and chicken), tea, coffee, bananas, macadamia, avocadoes, maize, beans, green grams, sorghum, pearl millet, cowpeas and pigeon peas (Tharaka Nithi County Integrated Plan, 2018–2022 –, 2018–2022).

3.2. Research design

A cross-sectional research design was applied to collect primary data during the period from December 2021 to March 2022. Secondary data was obtained from the Ministry of Agriculture offices situated in Kajuki, Marima, Kieganguru, and Kathwana within Tharaka Nithi County.

3.3. Sampling procedure

A multi-stage stratified sampling procedure was used to identify households practicing dairy farming. Dairy farmers were purposely targeted since they were key beneficiaries of the targeted DAT services in the County. The sampling procedure involved several stages. First, the selection of the county (Tharaka Nithi) was done purposively. In the second stage, six Wards where dairy farming was practiced were chosen randomly. The number of respondents per Ward was proportionate to the population of dairy farmers in that specific Ward. Moving to the third stage, two Locations within each Ward were randomly selected, and dairy farmers from these Locations were interviewed. The minimum required sample size was determined using the Cochran (1977) method as shown:

(1) $n=\frac{Z^2\left(p\right)\left(q\right)}{d^2}$(1)

Where n = Sample size; Z = Statistical certainty, equals 1.96 for 95% confidence level; p was estimated as 0.5 to give the maximum sample size; q = The weight variable and was computed as 1-p and d = the degree of precision or margin of error assumed at 5%. Though the formulae gave a sample size of 384 respondents, a total of 427 households were interviewed.

3.4. Criteria for inclusion and exclusion

Participation in the study was contingent upon the respondent ownership of a smartphone, dairy cow(s), and the respondent’s agreement to voluntarily participate after the consent form was read aloud by the interviewer.

3.5. Ethical considerations

Prior to data collection, a research permit was acquired from the National Commission for Science, Technology & Innovation in Kenya (License No: NACOSTI/P/22/16655). During interviews, a thorough consent procedure was conducted with all prospective participants prior to questioning.

3.6. Data collection

Data was collected from the primary decision maker within each household who was engaged in dairy farming. A team of enumerators collected the data using a structured questionnaire administered through a mobile application.

3.7. Data analysis

Data was received in a delimited file format, and STATA used for analysis. Descriptive statistics were utilized to conduct basic analysis, which were presented in the form of tables and figures. Econometric modeling techniques were applied to assess the factors influencing the adoption decision and the extent of use of mobile applications.

3.8. Econometric model specification and estimation procedure

According to Wooldridge (2020), the choice of an econometric model depends on the type of data, as well as the specific goals and objectives of the study. Logit or Probit regression methods are commonly used to model factors influencing the adoption of technologies (Gujarati, 2004). Heckman selection model, Tobit model, and Double hurdle models are used to simultaneously model adoption and decision to adopt technologies (Okello et al., 2012). When using the Tobit model, the assumption is that the variables explaining the choice of technology and the number of technologies taken up are similar (Oluoch-Kosura et al., 2001). When using the Heckman model, the assumption is that non-adopters will never adopt a technology while in a double hurdle model, non-adopters are considered potential adopters (Jones, 1989). Heckman model becomes inappropriate in this study since the non-adopters of mobile applications may adopt owing to factors such as awareness creation and training.

The double hurdle model assumes that farmers are faced with two choices in technology adoption, that is adoption and the extent, each of which is determined by a different set of explanatory variables. In this study double hurdle model is used to identify factors affecting the probability of adopting mobile applications and the extent of adoption measured by the number of applications the farmer uses. Double-hurdle model is a parametric generalization of the Tobit model, in which two separate stochastic processes determine the decision to adopt and the extent of adoption of technology.

The double hurdle model has been used to determine factors influencing the adoption of technologies and the number of technologies adopted such as Mihretie et al. (2022) and Mujeyi et al. (2022). Several studies have also used the Double hurdle model to identify determinants of adoption and intensity of adopting technology (Beshir et al., 2012; Zeleke et al., 2021).

To determine the most appropriate model for assessing the factors influencing the adoption decision, and the extent of use of mobile applications, the Likelihood Ratio (LR) tests were done for Tobit and the combination of Probit and Truncated models. Probit regression on the choice to use followed by a Truncated model on the users was used for the Double hurdle model following Cragg (1971).

Probit model equation is given as follows:

(2) $P\left(Y=\frac{1}{X}\right)=\mathrm{\varnothing }\left({\beta }_0+{\beta }_1{X}_1\right)$(2)

Where $P$ denotes the probability of a farmer adopting a mobile application, $Y$ choice of mobile app; Φ is the cumulative distribution function of the standard normal distribution. β₀ is the intercept and ${\beta }_1$is the regression coefficient: $X_1$ is the explanatory variable.

Index equation:

(3) $Y^{\ast }=\{1ifY>0,andis0ifY\le 0\}$(3)

(4) $Y{\beta }_0+{\beta }_1{X}_1+{\beta }_2{X}_2+\dots +{\beta }_n{X}_n+\epsilon \left(\mathrm{W}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e}\epsilon \mathrm{i}\mathrm{s}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{e}\mathrm{r}\mathrm{r}\mathrm{o}\mathrm{r}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{m}\right)$(4)

Equation of the truncated model at zero:

(5) ${Y_i}^{\ast }=x_i\beta +u_i$(5)

Where

(6) $Y_i=Y_i{ }^{*},if{Y}^{*}00,if{Y}^{*}\le 0$(6)

Where $Y_i$ is the number of applications which depends on the latent variable $Y^{\ast }$ being greater than zero, $x_i$vector of explanatory variables, $\beta$ is the vector of parameters to be estimated, and $u_i$ is the error term.

The log likelihood function for the double-hurdle model is represented by equation 7 as follows:

(7) $L=\sum _{0}In\left[1-\varphi \left(\alpha Z_i^{\prime }\right)\left(\frac{\beta X_i^{\prime }}{\sigma }\right)\right]+\sum _{+}In\left[\varphi \left(\alpha Z_i^{\prime }\right)\frac{1}{\sigma }\phi \left(\frac{Y_i-\beta X_i^{\prime }}{\sigma }\right)\right]$(7)

Where ${\sum }_0$ is the summation over the zero observations; ${\sum }_{+}$ is the summation over positive observations; $\varphi$ is the standard normal cumulative distribution function and $\phi$ is the probability distribution function.

The Tobit model was specified as follows (Dougherty, 2011):

(8) $\hat{y}={\beta }_0+{\beta }_i{X}_i+\epsilon$(8)

(9) $Y=\hat{y}if{\beta }_0+{\beta }_i{X}_i+\epsilon >0$(9)

(10) $Y_i=0if{\beta }_0+{\beta }_i{X}_i+\epsilon \hspace{0.17em}\le \hspace{0.17em}0$(10)

Where Ŷ= latent variable which is not observable; $Y_i=$observed dependent variable; $X_1$ is the vector of explanatory variables; ${\beta }_0$ is the intercept and${\beta }_i$ the vector of parameters to be estimated while $\epsilon$ is the error term.

Tobit likelihood function was specified as follows:

(11) $L=\prod _{Y_i^{\ast }>0}f\left(\frac{Y_i-{\beta }_i{X}_i}{\sigma }\right)\prod _{Y_i^{\ast }\le 0}F\left(\frac{-{\beta }_i{X}_i}{\sigma }\right)$(11)

Where f is the density function and F the cumulative distribution function of $Y_i^{\ast }$; ${\mathrm{\Pi }}_{Y_i^{\ast }>0}$ is the product over those $i$ for which $Y_i^{\ast }>0$, and ${\mathrm{\Pi }}_{Y_i^{\ast }\le 0}$ is the product over those $i$ for which $Y_i^{\ast }\le 0$.

The likelihood ratio statistic was computed following Greene (2000) as follows:

(12) $\mathrm{\Gamma }=-2\left[In{L}_T-\left(In{L}_P+In{L}_{TR}\right)\right]\sim X_k^2$(12)

Where $L_T$ is the Tobit model likelihood; $L_P$ and $L_{TR}$ is the Probit model likelihood and Truncated Model likelihood respectively while $k$ are the independent variables.

3.9. Description of variables

To estimate the factors influencing adoption of mobile phone applications, a binary variable was coded 1 if a farmer used any mobile application to access agricultural information and 0 if otherwise. The number of mobile applications in use was used to measure the extent of adoption. The explanatory variables used in this study are the socio-economic factors, sources of information about mobile applications and farmer perceptions (Table 1).

Table 1. Definition of explanatory variables

Variable	Definition of variable	Hypothesis
Dependent Variable
Adoption	Decision to adopt a mobile phone application (Yes =1, otherwise 0)
Extent of use	The number of mobile applications the farmer uses
Independent Variables
Age	Age recorded in the number of years (continuous variable)	-/+
Education	Years spent in school taken as cumulative from lowest to highest level (continuous variable)	+
Sex	Recorded as a dummy variable, representing the sex of the farmer (1= Male, 0 = female).	-/+
Occupation	Farmers key occupation (dummy variable 1= Farming, 0=Otherwise)	-/+
Income	Total household annual income in Ksh (continuous variable), estimated from all sources of income for the household	+
Farm size	Farm size in Ha (continuous variable), collected from title deeds.	-/+
Access to credit	Access to formal credit (financial and/or inputs) a dummy variable (1= farmer has gotten credit in the last 12 months, 0=Otherwise)	+
Distance to the main road	Distance from the farm location to the nearest tarmac road (km) (continuous variable), calculated using QGIS	-
Distance to the market	Distance from the farm location to the nearest market (km) (continuous variable), calculated using QGIS	-
Farmer organisation	Farmer belongs to a farmer’s organization (dummy variable 1= Yes, 0=Otherwise)	+
Sources of information about mobile applications
Public extension	1= farmer got information from the government extension officer, 0 otherwise	+
Private extension	1= farmer got information from the private extension service providers, 0 otherwise	+
Researcher	1= farmer got information from officers from the National Research Organizations and the Universities, 0 otherwise	+
Social media	1= farmer got information from social media, 0 otherwise	+
Ease of using an app	Perception on the ease of using mobile applications a dummy variable (1= mobile applications are easy to navigate 0=Otherwise)	+
Cost of using an app	Perception on affordability of mobile applications a dummy variable (1= affordable 0=Otherwise)	+

4. Results

4.1. Descriptive statistics

Table 2 presents descriptive statistics for the socioeconomic characteristics of the sampled household in the study area. The mean age was 46 years with a household size of 4 persons per household. Average distances to the nearest market centre and main road were 2.4 km and 5.29 km, respectively. The average number of years of schooling was 12 indicating that the majority of farmers were able to read and write. The average household income was Ksh 161,149 per annum and a mean of 2 contacts with extension service providers.

Table 2. Socioeconomic characteristics of the sampled households (continuous variables)

Characteristic	Mean	Std. dev.
Age of the respondent in years	46	13
Years in school	12	3
Household size	4	1
Total household annual income (Ksh)	161,149	97,598
Distance from the farm to the tarmac road (km)	5.29	2.83
Distance from the farm to the nearest market (Km)	2.40	1.69
Farm size in Ha	0.71	0.58
Number of cows kept by respondents	2	1
Number of contacts with extension providers for last 12 months	2	1
Total number of farmers N = 427

The result in Table 3 revealed that 51% of the respondents used mobile application(s) with 46% of men using mobile applications and 54% of women using mobile applications. The majority (86% of the respondents) were members of farmer organizations and 56% had contact with extension service providers. Out of all the respondents, 85% used the Internet and 77% were aware that they could use mobile phones to access agricultural information. Credit for use in agriculture was accessed by 60% of the respondents.

Table 3. Socioeconomic characteristics of the sampled households (dummy variables)

Characteristic	Yes	No
Use of mobile applications	51%	49%
Gender (Yes =Male; No = Female)	46%	54%
Respondent belong to a farmer organization	86%	14%
Contact with extension service providers in the last 12 months	56%	44%
Respondent awareness of using mobile phone to access agric infor	77%	23%
Access to internet	85%	15%
Access to formal credit in the last 12 months	60%	40%
Total number of farmers N = 427

A total of 22 mobile apps were in use by farmers including web-based Google Search and YouTube as shown in Figure 2 and Annex 1. Google search engine was the most common application which was used by 33% of the farmers who participated in the study. Farmers also used social media to access agricultural information, the most common social media application that was in use was WhatsApp, used by 20% of the farmers, and Facebook 19%. Farmers belonged to social media groups that shared or exchanged agricultural information. Other common applications were KARLO Selector used by 12% of the farmers interviewed, YouTube used by 11%, and Kenya Agri-Observatory Platform (KAOP) used by 9.6%. On average, farmers were using one mobile application.

Figure 2. Common mobile applications used to access agricultural information by farmers in Tharaka Nithi, Kenya.

Table 4 and 5 show significant differences between individuals who use mobile phone applications for agricultural purposes and those who do not. For example, 54% of male users utilized agricultural applications, while only 38% of non-users were male. This suggests that there are more male users than female users. In terms of access to credit, 69% of agricultural app users had access to credit compared to 49% of non-users. Interestingly, 53% of non-users were aware that they could use their mobile phones to access agricultural information. Additionally, 67% of agricultural app users had contact with extension service providers, while only 45% of non-users had such contact.

Table 4. Differences in characteristics of users versus non-users of mobile applications (dummy variables)

Characteristic	Users		Non-users	Pearson Chi-Square
	N=216		N=211
Gender	Male	54%	38%	10.685***
	Female	46%	62%
Respondents belong to a farmer organization	Yes	89%	83%	3.208
	No	11%	17%
Access to formal credit for last 12 months	Yes	69%	49%	17.991***
	No	31%	51%
Access to internet	Yes	100%	69%	79.916***
	No	0%	31%
Awareness of using mobile phone to access Agric Information	Yes	100%	53%	133.675***
	No	0%	47%
Contact with extension service provider	Yes	67%	45%	19.432***
	No	33%	55%

Significance of mean difference at ***1%.

Table 5. Differences in characteristics of users versus non-users of mobile applications (continuous variables)

Characteristic	Users	Non-users	t –test
Age of the respondent in years	43	49	−4.785***
Years in school	12	11	5.397***
Household size	4	3	1.864
Total household annual income (Ksh)	179,368	142,498	3.70***
Distance from the farm to the main road (Km)	5.15	5.43	−1.048
Distance from the farm to the nearest market (Km)	1.89	2.93	−6.732***
Farm size in Ha	0.79	0.63	2.936
Number of cows kept by respondents	2	2	−0.372
Number of contacts with extension providers in last 12 months	2	1	4.431***
Total number of farmers (N)	216	211

Significance of mean difference is at ***1%.

Users of mobile applications were significantly younger with a mean age of 43 years compared to non-users 49 years (Table 5). There were significant differences between the users and non-users of agricultural applications in the years of schooling with users having spent 12 years and non-users 11 years. Users of agricultural applications had more income than users, were closer to markets, and had more contacts with agricultural extension service providers (Table 5).

4.2. T-Statistics on double hurdle (probit and truncated model versus the Tobit model)

Based on the Log Likelihood Ratio test result of Г = 101.6, which is higher than the critical value of the χ2 distribution at a 0.05 level of significance (as shown in Table 6), it is concluded that the Tobit model should be rejected. This implies that the decision to use and the extent of adoption of mobile applications were made separately.

Table 6. Log likelihood Ratio test for Tobit and Probit + Truncated models

	Probit	Truncated	Tobit
Log likelihood	−172.27	−193.63	−416.7
No. of observations	427	216	427
Test statistics: Г= 101.6 > χ^2 0.05,16 = 26.29

Goodness of fit test for Probit showed that the model fitted the data well with a χ² (410) of 391.11 and P > χ² = 0.74. The area under Receiver Operating Characteristic (ROC) curve test was 0.89 (Figure 3).

Figure 3. Area under receiver operating Characteristic curve test.

Goodness-of-fit test after probit model

Table

No. of observations	427
Number of covariate patterns	427
Pearson χ^2 (410)	391.11
Prob > χ^2 = 0.7411

4.3 Determinants of the adoption decision and extent of adoption of mobile applications among smallholder dairy farmers in Tharaka Nithi County

According to Table 7, the sex of the household head had a significant effect on the use of mobile phone applications. If the farmer was male, there was an expected 8% likelihood of using an app, when all other factors were held constant. The results showed that as farmers aged by one year, the likelihood of using mobile applications decreased by 0.3% (given the instantaneous rate of change). Similarly, an increase of one unit in distance to the market decreased the probability of using mobile applications by 7%. Marginal effect results showed that an additional year of schooling increased the probability of using mobile phone applications by 2%. Furthermore, if farming was the primary occupation, there was a 5% increase in the probability of adopting mobile phone applications. The results also indicated that access to credit for agricultural use increased the probability of adopting mobile phone applications by 10%.

Table 7. Double hurdle results on factors influencing the decision to adopt mobile phone applications

	Probit model		Truncated model
Variable	Marginal Effects	Std. Error	Coefficient	Std. Error
Sex of household head	0.08**	0.04	−0.17	0.10
Age of household head	−0.003*	0.002	0.002	0.005
Distance to main road	0.01	0.01	0.005	0.02
Distance to the market	−0.07***	0.01	0.01	0.04
Education of Household head	0.02***	0.01	−0.01	0.02
Occupation	0.05**	0.02	0.01	0.05
Household Income	−0.02	0.03	0.05	0.05
Farm size	0.04	0.04	0.20**	0.09
Group membership	0.01	0.05	0.21	0.18
Access to Public extension	0.18***	0.07	0.18	0.13
Access to Researcher	0.33***	0.10	0.54***	0.17
Access to Private extension	0.21***	0.05	0.32***	0.12
Access to social media	0.18***	0.05	0.31***	0.12
Access to Credit access	0.10***	0.04	0.13	0.12
Ease of using an app	0.21***	0.06	0.02	0.23
Cost of using an app	0.17***	0.04	−0.13	0.14

***significant at 1%, **significant at 5%, *significant at 10%.

An instantaneous rate of change of one-unit increase in land holding, the probability of using an additional mobile application increased by 0.2. Furthermore, the source of information about mobile applications also appears to play a role in adoption, with farmers who learned about mobile applications from National Research Organisations having the highest probability of adoption at 33% and 54% probability of adopting an additional mobile application. Other factors such as the perception of how easy or difficult it is to navigate through the mobile phone app, and the perception of the cost of using the mobile phone application had a significant effect on adoption (Table 7).

5. Discussion

The results of the double hurdle regression showed the factors hypothesized to drive the use of mobile phone applications. Among the socio-economic characteristics included, sex, age, distance to the nearest market, education, and primary occupation were key variables that significantly influenced the adoption of mobile applications, with no significant influence on the number of applications the farmer uses. These findings are consistent with a previous study that found out that male farmers were more likely to use ICT-based pest information services in central highlands of Kenya (Mwenda et al., 2022). The possible explanation is that male-headed households have more access to productive resources compared to women (Ingutia & Sumelius, 2022; Kamara et al., 2022). Earlier studies have also shown gender differences in technology adoption, particularly on the use of mobile phones in Kenya (Wyche et al., 2019) and Krell et al. (2020).

According to the study, older household heads were less likely to adopt mobile applications. A possible justification for this finding is that older household heads are less receptive to new technologies compared to the younger farmers. Given their previous experience in farming, older dairy farmers are reluctant to adopt innovations given the risks associated with them. Risk aversion has been identified as a major factor that hampers the adoption of new technologies in developing countries. For example, Abdulai and Huffman (2005) and Mwangi and Kariuki (2015) showed that as farmers grow older, they become risk-averse and conservative to new technologies. Besides, younger farmers normally spend more time and effort on mobile applications compared to older farmers. Earlier studies have also supported the negative relationship between the age of the farmer and the use of agricultural technologies in different part of the World (Agwu et al., 2008; Daberkow & McBride, 2003; Simtowe et al., 2016: Denkyirah et al., 2016; Udimal et al., 2017; Xie & Huang, 2021).

Distance to the nearest market had a negative effect on the adoption of mobile phone applications, suggesting that farmers nearer the market had a higher probability of accessing information about the availability of mobile phone applications that would enhance adoption. Higher transportation costs will be incurred if the market center is far thus subjecting farmers’ access to information about mobile applications to difficulty. Thus, a shorter distance to the market is potentially an incentive for the adoption of mobile applications. Distance to the market may be considered a proxy for access to information in Sub-Saharan Africa (Atinafu et al., 2022; Belay & Mengiste, 2021; Darkwah et al., 2019).

Years of schooling had a positive and significant effect on the probability of adoption of mobile phone applications. The justification of this finding is that education remains a helpful tool since it enables the farmer to make informed choices of the available technologies. As well, higher levels of education help the farmer to comprehend and speculate on the benefits embedded in the adoption of a technology. Past studies have shown that the education level of a farmer increases his/her ability to acquire and use information relevant to the adoption of a new technology particularly in developing Countries (Kumar et al., 2018; Li et al., 2019; Mwaura et al., 2019; Namara et al., 2003; Wu & Licata, 2022).

Having farming as a primary occupation had a positive and significant effect on the adoption of mobile phone applications. The possible explanation is that farmers will be willing to invest more in the source of information that is channeled through the mobile application in an attempt to gain more knowledge on new production technologies and other complementary input. This could also be attributed to the degree of reliance on farming by farmers who highly depend on it for their livelihoods as found (Olusayo et al., 2019).

Access to credit had a positive and significant effect on the adoption of mobile phone applications. The possible explanation is that a farmer would adopt a mobile app since access to credit (an indicator of financial constraint) enables the farmer to meet various costs such as buying a smartphone, purchasing data to enable installation of the mobile apps etc. A study by Balana and Oyeyemi (2020) in Nigeria showed that access to credit by farmers may aid technology adoption due to increased financial capacity. Further, studies in other parts of the World showed that access to credit was a determinant of technology adoption such as a study by Mariyono (2019) in Indonesia; in Pakistan (Rehman et al., 2019, and in Ethiopia (Fikire et al., 2022; Workineh et al., 2020; Zegeye et al., 2022).

Larger farm sizes were positively associated with the adoption of mobile phone applications. This finding may be attributed to bigger farm sizes accommodating more intensive systems of production that require significant capital, labor, management, and advanced technologies. In this case, larger land holdings provide an opportunity to seek more information related to investments, as demonstrated by studies conducted by Langyintuo and Mekuria (2008) in Africa, Paustian and Theuvsen (2017) in Germany, and in China (Hu et al., 2022).

Farmers have been introduced to mobile applications or have increased their usage of mobile applications through learning from researchers, private extension service providers, and social media. The study highlights the importance of the means by which farmers are introduced to these apps as it influences their usage. A previous study (Murage et al., 2012) demonstrated the significance of the pathways used to disseminate agricultural technologies in driving their adoption among farmers in Kenya.

Farmers who thought that it is easy to navigate through mobile phone apps were more likely to adopt them, as were those who felt the cost of use was manageable. Previous research supports these findings, highlighting the importance of perceptions of ease of use and cost in the adoption of new technologies (Diaz et al., 2021; Okoroji et al., 2021; Rehman et al., 2016).

6. Conclusions and recommendations

This study identified the factors driving the adoption of mobile phone applications for accessing agricultural information among smallholder dairy farmers. The results indicated that the decision to adopt mobile phone applications was influenced by sex, age, education level, distance to the market center, farming as the primary occupation, access to credit, and the means by which farmers learned about mobile applications. Perceptions of cost and ease of use were also identified as key determinants of adopting mobile phone applications among smallholder dairy farmers. Additionally, the study found that farm size and the method of disseminating information about mobile applications had a positive impact on the number of applications adopted by dairy farmers.

Smallholder dairy farmers can benefit greatly from mobile phone applications for accessing agricultural information. This study offers useful insights on how to promote these apps effectively. Farmers should be educated on the available applications and taught how to use them, including understanding the costs involved.

To help smallholder dairy farmers access mobile applications, it’s important to improve the current extension pathways and spread awareness of their availability. Public-private partnerships can help encourage farmers to adopt these applications. Developing policy guidelines that involve the private sector in educating farmers about the applications is crucial. Future research should focus on how farmers use these applications and how they impact dairy production.

Acknowledgments

The authors acknowledge funding from the World Bank and the Government of Kenya through the Kenya Climate Smart Agriculture Project (KCSAP).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data is available upon a reasonable request.

Additional information

Funding

The work was supported by the Kenya Climate Smart Agricultural Project.

Notes on contributors

Samuel P. Mutuma

Samuel P. Mutuma is a PhD student in Agricultural Economics at Kenyatta University. He specializes in value chain development, design and implementation of agribusiness projects, and delivery of extension services to farmers.

Wangare L. Ngare

Wangare L. Ngare, PhD Specialist in Agricultural Economics and Agricultural Marketing. She is a Senior Lecturer at the Cooperative University of Kenya.

Eric K. Bett

Eric K. Bett, PhD An agribusiness specialist and Senior Lecturer of Agricultural Economics and Agribusiness at Kenyatta University. He heads the Department of Agricultural Economics, Kenyatta University.

Christopher N. Kamau

Christopher N. Kamau Tutorial Fellow at the Department of Agricultural Economics, Kenyatta University. His research interests are in production economics, agricultural marketing, adoption of agricultural technologies, and impact assessment.

Annex 1: Mobile Applications used by the respondents

1. WhatsApp

2. Facebook

3. Google_search

4. Youtube

5. Digicow

6. Icow

7. Digifarm

8. Farmcare

9. Farm_kenya

10. Sokoyetu

11. KAOP

12. KALRO_selector

13. KALRO_TIMPS

14. KALRO_Ichicken

15. Farming_Appkenya

16. Mkuilma_Bora

17. Mkulima_online

18. Farmers_Tips

19. Mkulima_young

20. I_shamba

21. Pig_farmer

22. Kenya_met